Archives of Computational Methods in Engineering (2022) 29:5525–5567

https://doi.org/10.1007/s11831-022-09776-x

REVIEW ARTICLE

Computer Based Diagnosis of Some Chronic Diseases: A Medical

Journey of the Last Two Decades
Samir Malakar1 · Soumya Deep Roy2 · Soham Das2 · Swaraj Sen3 · Juan D. Velásquez4,5 · Ram Sarkar3

Received: 16 March 2022 / Accepted: 22 May 2022 / Published online: 15 June 2022
© The Author(s) under exclusive licence to International Center for Numerical Methods in Engineering (CIMNE) 2022

Abstract
Disease prediction from diagnostic reports and pathological images using artificial intelligence (AI) and machine learning
(ML) is one of the fastest emerging applications in recent days. Researchers are striving to achieve near-perfect results using
advanced hardware technologies in amalgamation with AI and ML based approaches. As a result, a large number of AI and
ML based methods are found in the literature. A systematic survey describing the state-of-the-art disease prediction methods,
specifically chronic disease prediction algorithms, will provide a clear idea about the recent models developed in this field.
This will also help the researchers to identify the research gaps present there. To this end, this paper looks over the approaches
in the literature designed for predicting chronic diseases like Breast Cancer, Lung Cancer, Leukemia, Heart Disease, Dia-
betes, Chronic Kidney Disease and Liver Disease. The advantages and disadvantages of various techniques are thoroughly
explained. This paper also presents a detailed performance comparison of different methods. Finally, it concludes the survey
by highlighting some future research directions in this field that can be addressed through the forthcoming research attempts.

1 Introduction visit the clinic and the health professionals advise them
through their expertise in diagnosis. However, in this age-
In this digital era, many organizations have been established old way of medical diagnosis, patients face various difficul-
across the globe which provide continuous health monitor- ties owing to the increase in the number of health related
ing facilities for humans. In the traditional method, patients problems as well as the population, especially in developing
countries. This scenario sometimes leads to improper care
* Samir Malakar of a patient, which can even prove fatal.
[email protected] To this end, technology provides an alternative to the tra-
Soumya Deep Roy ditional system. Hence, it plays a significant role in health-
[email protected] care systems by incorporating a large number of computer
Swaraj Sen aided supporting systems and tools. This bonding has not
[email protected] only improved the quality of patient care but also reduced
Juan D. Velásquez the cost of treatment by imparting efficient allocation of
[email protected] medical resources. The main components of technology-
Ram Sarkar enabled healthcare systems are medical experts, hardware
[email protected] and software. However, designing an automatic system that
can predict the disease from electronically available medi-
1
Department of Computer Science, Asutosh College, Kolkata, cal data is very challenging. The huge social impact of this
India
research field motivates researchers from various domains
2
Department of Metallurgical and Material Engineering, like computer science, biology, medicine, statistics, and drug
Jadavpur University, Kolkata, India
design. These researchers are continuously trying to come up
3
Department of Computer Science and Engineering, Jadavpur with a near perfect system for better patient care.
University, Kolkata, India
In this context, it is worth mentioning that with the grow-
4
Departament of Industrial Engineering, University of Chile, ing availability of digital records and data, the last two dec-
Santiago, Chile
ades have observed an exhaustive adoption of data mining
5
Instituto Sistemas Complejos de Ingeniería (ISCI), Santiago, and machine learning (ML) techniques [89] in healthcare/
Chile

13
Vol.:(0123456789)
5526 S. Malakar et al.

patients care systems. Healthcare, one of the most crucial processes insulin. Type-II Diabetes accounts for more than
sectors of our society, is availing the facility of digitization 90% deaths of all the Diabetes cases [155]. It mainly arises
of medical records and the emergence of Electronic Health out of unhealthy lifestyle choices. Over 30 million people
Records (EHRs). The large-scale availability of EHRs has have been diagnosed with Diabetes in India and are attrib-
led to a surge in Computer aided Diagnosis and Detection uted as the direct cause of 1.6 million deaths as of 20163.
(CADD) systems which, in general, employ various ML Diabetes along with hypertension is responsible for
algorithms to accurately predict the presence of a particular Chronic Kidney Diseases (CKD). In CKD, the malfunc-
disease in a subject. These CADD systems help in removing tioning kidneys fail to filter waste from the blood leading to
subjectivity in EHR and/or histopathology image analysis, waste accumulation and may eventually lead to renal fail-
thereby minimizing the prediction error. According to the ure. A major impediment to CKD diagnosis is that the early
third global survey1 on electronic health (eHealth), con- stages of CKD show no symptoms. Around 100,000 patients
ducted by the World Health Organization (WHO) in 2016, are diagnosed with end stage kidney disease every year4 in
there has been a steady growth in the adoption of EHRs India. Unhealthy lifestyle choices also cause Liver Diseases.
over the past 15 years and a 46.00% global increase in the The liver is the largest organ in the body and Liver diseases
past five years. mainly arise from excessive consumption of substances
A disease is an abnormal condition that affects mostly like alcohol, harmful gases, contaminated food, pickles
a part of an organ and it is not caused by some external and drugs [148]. With 259,749 deaths in 2017 in India and
injury. In medical science, there are many categories of dis- around 1.3 million deaths worldwide from cirrhosis alone,
eases like acute, infectious, heredity, and chronic. Chronic Liver Disease is one of the major health issues worldwide5.
diseases generally persist for a longer period (3 months or
more) in human organs. In general, such diseases cannot be 1.1 Motivation and Contributions
prevented using vaccines or cured by medications. How-
ever, early detection of these diseases can save many human All the facts mentioned earlier demand the need for early
lives. Chronic diseases such as cancer, heart disease, and detection of such chronic diseases, which could save a large
diabetes are the leading causes of death of human beings. number of human lives. As a result, several researchers
With 784,821 deaths in India and an estimated 9.6 million around the globe engaged their time to find the solutions
deaths globally in 2018, cancer is considered as one of the for detecting chronic diseases at their early stages. Also, the
most fatal diseases. Cancer generally involves abnormal cell abundant data gathered from EHRs, medical diagnosis and
growth with the potential to spread to other organs in the medical imaging alongside the technological and technical
human body. The most common forms of cancer include improvement in ML and artificial intelligence (AI) have led
Breast Cancer, Lung Cancer, Bronchial Cancer, Leukemia, to significant research in Bioinformatics, Biomedical imag-
Prostate Cancer, etc. It comes as no surprise that the earliest ing and CADD systems. There has been extensive research
efforts in CADD [58] started with mammography for the in the domain of disease prediction. Since 1997, such meth-
detection of Breast Cancer, and these techniques were later ods have been published in different refereed journals and
applied to other forms of cancer as well. conferences [163]. As a result, voluminous research articles
Heart Disease, also called Cardiovascular Diseases are present in the literature. Accordingly, several review arti-
(CVDs), is another group of diseases that causes a large cles [5, 57, 59, 78, 91, 92, 92, 93, 103, 134, 177, 201] are
number of deaths every year. Diseases that fall under the also present in the literature on chronic disease prediction.
umbrella of CVDs include coronary artery disease, heart We have summarized the types of methods these review arti-
rhythm problems (also known as arrhythmias), and congeni- cles considered in Table 1. From Table 1, we can observe
tal heart diseases among others. According to Abdul-Aziz that these articles were prepared by highlighting the problem
et al. [3], one in four deaths in India is due to CVD. Accord- domain narrowly i.e., these were prepared either by high-
ing to a WHO report2 more people (over 17.9 million each lighting a specific chronic disease [90, 93, 93, 177] in most
year) die from CVDs worldwide than from any other cause. of the cases or in a few cases, a specific category of ML or
Diabetes is one of the prime risk factors for CVDs. It is of AI-aided techniques [5, 91, 92, 103, 201]. Sometimes, the
two types: Diabetes insipidus or Type-I Diabetes in which authors discussed specific ML or AI-aided techniques for
the pancreas produces little to no insulin and Diabetes mel-
litus or Type-II Diabetes which affects the way the body
3
https://​www.​who.​int/​news-​room/​fact-​sheets/​detail/​diabe​tes
4
https://​www.​india​today.​in/​educa​tion-​today/​gk-​curre​nt-​affai​rs/​story/​
1
https://​www.​who.​int/​gho/​goe/​elect​ronic_​health_​recor​ds/​en/ world-​kidney-​day-​hyper​tensi​on-​and-​diabe​tes-​two-​major-​causes-​of-​
2
https://​www.​who.​int/​health-​topics/​cardi​ovasc​ular-​disea​ses/#​tab=​ kidney-​disea​ses-​14778​41-​2019-​03-​14
5
tab_1 https://​en.​wikip​edia.​org/​wiki/​Cirrh​osis

13
Computer Based Diagnosis of Some Chronic Diseases: A Medical Journey of the Last Two Decades 5527

Table 1  Coverage of the present survey in comparison with the past surveys
Review article Remarks

Fatima et al. [57] 1. Research works related to the prediction of heart, Diabetes, liver, dengue and hepatitis were included for
discussion
2. On average, 4 to 5 works for each disease were included for discussion
3. No critical analysis was provided to highlight the future research directions for such diseases
Rigla et al. [177] 1. Research findings on Diabetes detection from diagnosis reports were considered
2. Classical ML approaches like Neural Network (NN), Decision Tree (DT), Support Vector Machine (SVM)
based methods were included
3. Important factors like notable research findings, future research directions, critical analysis of the surveyed
works were missing
Jain and Singh [103] 1. Importance was given to describing FS techniques rather than disease detection methods
2. Most of the works used (more than 90%) for critical analysis were on Diabetes prediction
3. Image based disease predictions were absent as all the cited works were on diagnosis report based repositories
3. Disease specific discussion was missing
Ahmadi et al. [5] 1. Works considered for review are more fuzzy technique centric rather than disease-centric which means fuzzy
techniques were prioritized rather than the disease detection
2. Published papers till 2017 were considered for discussion
3. Only 52 works on different (>10), i.e., on an average 3 to 4 works per disease, and diagnosis report based
disease were used
Hosni et al. [91] 1. Seven research questions, mostly classifier ensemble oriented, were set, and the published works that address
each of the research questions mentioned, were investigated. So this review is entirely focused on ensemble
based methods
2. Only the works on Lung Cancer disease detection from diagnosis reports were considered
3. Ensemble methods that were considered followed classical ML algorithms
Hosni et al. [90] 1. Some of the notable research outcomes related to Breast Cancer detection from diagnosis reports were consid-
ered for analysis
2. No description/critical analysis of the methods could be found. Rather categorical analyses of the works were
made without individualization
3. No description of used datasets and performance comparison and hence the beginners may find it difficult to
get the complete idea over the problem domain
Fernández et al. [59] 1. Research initiatives for Diabetes disease detection were considered
2. Disease detection techniques that employed classifier ensemble techniques were only included
3. Better performing single classifier works were excluded based on the selection strategy of the research works
4. No critical analysis of the works was provided
Hosni et al. [92] 1. Extended version of the work [91]
2. Extension in terms of paper selection strategy and increase of one research question
3. Statistics related to the questions were discussed. There was no focus on disease detection methods.
Gupta and Gupta [78] 1. Investigated the performance of Artificial NNs (ANNs), Restricted Boltzmann Machine, Deep Autoencoders,
and Convolutional Neural Networks (CNNs) for screening Breast Cancer
2. Described only 17 previously published research works
3. Results on only the Surveillance, Epidemiology and End Results (SEER) dataset were investigated to suggest a
suitable solution
Manhas et al. [134] 1. Reviewed image based cancer detection techniques
2. Five types of cancer - Cervical, Oral, Breast, Brain and Skin were included in the review
3. Diagnosis based cancer detection techniques were overlooked
Sharma and Rani [201] 1. A set of 12 research questions are identified for cancer research
2. Gene expression data based research was highlighted most
3. Research related diagnosis reports were also included
Houssein et al. [93] 1. Breast Cancer screening techniques were discussed
2. Only research work that dealt with histopathological, mammogram images were included
3. Diagnosis report based methods were also included

13
5528 S. Malakar et al.

Table 1  (continued)
Review article Remarks

Reshmi et al. [175] 1. Image based Breast Cancer detection techniques were included
2. Only research works that dealt with histopathological images were included and thus minimized the research
spectrum
3. Other aspects of Breast Cancer detection were not discussed
Proposed, 2022 1. Several diseases that include Breast Cancer, Lung Cancer, Leukemia, Heart Disease, Diabetes, Liver Disease
and CKD have been covered
2. Research endeavors made over the last two decades involving both image based and diagnostic report based
disease screening methods have been studied
3. Various important ML strategies like missing value imputation, feature reduction, FS, classifier combination,
fuzzy logic along with DL based strategies have been considered
4. A comprehensive description of each work along with its shortcomings and possible scope for future improve-
ment is provided. Also a comparative performance analysis of the methods for each disease has been provided
5. The disease specific research trends are analyzed and accordingly some future research directions for each
disease are suggested

a particular disease [78, 90, 177]. Hence, it can be safely Cancer, Heart Disease, Diabetes, Liver Disease and CKD
concluded that although surveys published enlisting research respectively. Section 4 is subdivided into three subsections
efforts for specific diseases or specific ML techniques, such for describing Breast Cancer, Lung Cancer and Leukemia
surveys fail to shed light on the prevailing trend of research respectively, each of which further contains subsections for
across multiple diseases or ML approaches. image and diagnostic report based disease prediction tech-
To this end, the present survey is a significant endeavor as niques. Whereas, Sect. 5 has separate sections for the single
it not only encapsulates the concerted research efforts in the classifier based approach and the Ensemble based approach
specific diseases but also attempts to reflect on the current in Heart Disease detection. For an easy reference, the organi-
trends in research across chronic diseases like Breast Cancer, zation of disease specific discussions made in this article is
Leukemia, Lung Cancer, Heart Diseases, Diabetes, CKD shown in Fig. 1. Some important future research directions
and Liver, and AI based methods including missing value are discussed in Sect. 9. Finally, this survey is concluded in
imputation, feature reduction, feature selection (FS), clas- Sect. 10.
sifier combination, fuzzy logic, and ML and Deep Learn-
ing (DL) based approaches. This survey also chronicles the
diagnostic report based and image based approaches for the 2 Generic Chronic Disease Prediction
said diseases. A comparative study of our survey with some Method
recent review articles is reported in Table 1. Information in
this table conveys that the present survey not only covers the A chronic disease prediction system accepts data from a new
similar methods described in state-of-the-art review articles subject as input and generates some specific report about
but also includes a wider range of AI based applications in the status of the disease observed in the subject. In gen-
several disease prediction systems. It is to be noted that in eral, the status of the subject is labelled either positive or
the present article we only concentrate on automatic disease negative. Status is positive if the subject is affected by the
prediction systems (an integral part of technology-enabled disease for which the test is made. Otherwise, the subject is
health care systems) that use several ML and DL schemes labelled negative. In some recent systems, the positive cases
for the prediction purposes. are further classified into more categories based on the stage
of the disease. For example, in the Breast Cancer Histol-
1.2 Organization of the Article ogy (BACH) image dataset6 Aresta et al. [17] labelled posi-
tive Breast Cancer stages into three stages: benign, in situ
The overall organization of this survey is shown in Table 2.
Section 2 provides the detail working procedure of a generic
disease prediction system and a generic diagnostic report
based system and an image based system. Section 3 dis-
cusses the commonly applied evaluation metrics used for 6
https://​iciar​2018-​chall​enge.​grand-​chall​enge.​org/​Datas​et/
assessing the performance of a disease prediction system.
Sections 4, 5, 6, 7 and 8 report the research endeavors in

13
Computer Based Diagnosis of Some Chronic Diseases: A Medical Journey of the Last Two Decades 5529

Fig. 1  Organization of disease prediction methods discussed in this article. Numbers indicate sections and subsections where the particular top-
ics are discussed

carcinoma and invasive carcinoma while in the Breast Can- 2.1 Diagnosis Report based Method
cer Histopathological Database (BreakHis)7 positive Breast
Cancer cases labelled into eight sub-categories. In the early years of research related to computerized disease
In Fig. 2, we have provided a schematic diagram which prediction systems, the patients’ data (diagnostic reports)
shows how a generic disease prediction system works (here was mostly stored electronically in text form. As a result,
a chronic disease). The figure highlights the communication a set of systems have been designed over time. In these
among subjects, doctors, pathological laboratories and AI systems, in general, a classifier is used to learn from these
assisted decision support systems. The clinical data (a set stored textual data. Most popular diagnosis report based
of diagnostic reports) for a subject may have three different datasets available electronically contains missing values,
forms: (i) input from the subject (e.g., height, weight, age, insignificant attributes and redundant information about the
and sex, etc.), (ii) input from the doctors or experts (e.g., subject which do not help the classifiers to learn from these
heart rate, body mass index, and disease specific observa- data sufficiently rather many times reduce their efficiency.
tions, etc.), and (iii) input from a pathological laboratory Therefore, preprocessing techniques that include missing
(e.g., blood cell counts, pathological image(s) and X-ray value imputation or data filtering (used as a substitute for
image, etc.). It is noteworthy to mention that the diagnos- missing value imputation) and data reduction have been
tic parameters vary from one disease to another. The col- employed by the researchers before feeding the data to a
lected clinical data are analysed by an AI-assisted disease classification model. The preprocessed training samples
prediction system and the system generates some forms of are then fed to the selected classifier to generate a learning
recommendations to be suggested by the doctor for the sub- module that contains optimally tuned hyperparameters of
ject, which are here considered as the presence or absence the classifier. In the testing phase (or prediction phase), a
of disease. The doctors analyse the recommendations and diagnosis report is collected from an unknown subject and
suggest treatment for the subject. In this survey, based on then preprocessed based on the available information from
the nature of clinical data, we broadly divide the disease the training stage. Finally, the preprocessed data are fed to
prediction systems into diagnosis report based systems and the saved learned model to decide whether the subject is
pathological image based systems. carrying the disease or not. A generic model of diagnosis
report based disease prediction system is shown in Fig. 3.

7
https://​web.​inf.​ufpr.​br/​vri/​datab​ases/​breast-​cancer-​histo​patho​logic​
al-​datab​ase-​break​his/

13
5530 S. Malakar et al.

Table 2  The overall organization of the article

Topic Work Refs.

1: Introduction
1.1: Motivation and Contributions
1.2: Organization of the Article
2: Generic Chronic Disease Prediction Method
2.1: Diagnosis Report based Method
2.2: Pathological Image based Method
3: Evaluation Metrics
4: Cancer Prediction Methods
4.1: Breast Cancer Prediction Methods
4.1.1: Diagnosis Report based Methods [6, 36, 43, 49, 66, 67, 72, 73, 98, 109, 121, 122, 129, 149, 165, 166, 202–204]
4.1.2: Pathological Image based Methods [28, 30, 37, 38, 40, 44, 52, 77, 79, 105, 172, 178, 185, 189, 195, 199, 212, 228,
230]
4.2: Lung Cancer Prediction Methods
4.2.1: Diagnosis Report based Methods [47, 49, 53, 87, 132, 151, 153, 161, 169, 180, 183]
4.2.2: Pathological Image based Methods [9, 13, 16, 104, 117, 152, 198, 209, 216, 218, 222, 231]
4.3: Leukemia Prediction Methods
4.3.1: Gene Expression based Methods [27, 55, 63, 70, 94, 114, 187, 188]
4.3.2: Pathological Image based Methods [2, 125, 142–145, 170, 176, 181, 196, 197]
5: Heart Disease Prediction Methods
5.1: Single Classifier based Methods [8, 43, 56, 60, 64, 69, 81, 97, 100, 101, 106, 110, 113, 136, 158, 186, 194, 200,
208]
5.2: Ensemble based Methods [14, 48, 130, 138, 146]
6: Diabetes Prediction Methods [19, 39, 43, 46, 54, 81, 82, 84, 99, 102, 111, 154, 160, 168, 173, 182, 202–204,
206, 221, 227]
7: CKD Prediction Methods [11, 12, 15, 18, 21, 35, 41, 74, 81, 108, 164, 167, 186, 190, 207, 219, 223, 225]
8: Liver Disease Prediction Methods [1, 4, 22, 43, 45, 80, 83, 112, 118–120, 174, 210, 220]
9: Future Research Directions
9.1: Exploring more DL Models
9.2: Designing Sophisticated Data Processing Strategies
9.2.1: Handling Missing Values
9.2.2: Exploring more Dimensionality Reduction Strategies
9.2.3: Designing new FS Strategies
9.3: Fusing Handcrafted Features with Deep Features
9.4: Handling Class Imbalance Problem
9.5: Generating Large-scale Datasets
9.5.1: New Data Collection
9.5.2: Generating Synthetic Data
9.5.3: Preparing Multi-modal Data
9.6: Medical Image Segmentation
9.7: Maintaining Ethical and Legal Aspects
10: Conclusion

2.2 Pathological Image based Method the images. As shown in Figs. 4 and 5 after preprocessing,
the general trend is to find out the Region of Interest (ROI)
A diagnosis system that uses pathological images collected to pinpoint areas which can be used to extract the features.
from the subjects to predict disease, in general, follows two In the case of DL based methods, the high level features
different approaches: feature engineering based approach are extracted from ROI images through the use of a set of
(see Fig. 4) and DL based approach (see Fig. 5). However, convolutional and pooling operations while in the case of
methods following any of these approaches commonly use the handcrafted features the domain knowledge contributes
some preprocessing techniques to obtain better features from significantly while extracting features. In the case of feature

13
Computer Based Diagnosis of Some Chronic Diseases: A Medical Journey of the Last Two Decades 5531

Fig. 2  Steps involved in a general decision support system

Fig. 3  A block diagram repre-

senting steps used, in general,
for predicting a disease from a
given set of diagnosis reports

engineering based methods, many standard classifiers are 3 Evaluation Metrics

used to predict the disease from unknown samples. Whereas
in the case of DL based methods, it is generally performed The disease prediction systems described in this article used
using multi-layered NNs that automatically extract features some classification models for disease diagnosis. Research-
from the input data. ers used some standard metrics like accuracy (ACC), pre-
cision (P), recall (R), F1-score (F1), Specificity (S) and

13
5532 S. Malakar et al.

Fig. 4  A block diagram representing steps used, in general, for predicting a disease from a pathological image following handcrafted features
extraction and shallow learner based classifiers

Fig. 5  A block diagram representing steps, in general, used for predicting a disease from pathological images using some DL based models

Area under Receiver Operator Characteristic (ROC) Curve performance of a model while predicting the result. It is a
(ROC-AUC score or simply AUC score) of a classification N × N (N is the number of classes) dimensional matrix. It
model while citing the performance of their model. These helps to summarize and visualize the performance of a pre-
evaluation metrics can easily be measured using the confu- dictive model. It consists of four metrics: True Positive (TP),
sion matrix or error matrix which describes the complete

13
Computer Based Diagnosis of Some Chronic Diseases: A Medical Journey of the Last Two Decades 5533

True Negative (TN), False Positive (FP), and False Negative 3.4 F1‑Score (F1)
(FN), which are defined below.
F1-score represents the trade off between recall and preci-
– TP indicates the number of positive samples that are pre- sion scores, and the harmonic mean of precision and recall
dicted correctly by the classification model is considered as the F1-score value. When the FP and FP are
– TN indicates the number of negative samples that are equally important then the F1-score metric is very useful in
predicted as negative samples by the classification model. order to know about the model’s prediction capability. It is
– FP indicates the number of negative samples that are calculated using Eq. 4.
predicted as positive samples by the classification model
P∗R
– FN indicates the number of positive samples that are pre- F1 − Score = 2 ∗ (4)
dicted as negative samples by the classification model P+R

3.1 Accuracy 3.5 Specificity

One of the most commonly used metrics for reporting the Specificity, also known as TN rate, is the proportion of nega-
performance of a disease prediction method is accuracy. It tive samples that are correctly classified i.e., it is the fraction
indicates how often a disease prediction model correctly pre- of TN and TN + FP . It is calculated using Eq. 5.
dicts a sample. It is calculated as the fraction of the number
of correctly classified samples (i.e., TP + TN ) and the total TN
S= (5)
number of test samples (i.e., TP + TN + FP + FN ) and it is TN + FP
defined by Eq. 1.
TP + TN 3.6 ROC‑AUC Score (AUC)
ACC = (1)
TP + TN + FP + FN
ROC curve is an evaluation metric for classification prob-
If the test set suffers from the imbalanced class problem [32, lems and disease diagnostic tasks, where the curve repre-
131] which is very common for medical datasets as normal sents the probability curve that plots the TP rate against the
cases overwhelm disease cases then this metric may mislead FP rate at various threshold values. It is used to show the
the overall model performance. Thus, the use of other met- trade-off between sensitivity and specificity in a graphical
rics like P, R, F1, S, and AUC scores become necessary and way. In other words, the AUC score describes how an adjust-
thus used widely in the past works. able threshold causes changes in two types of errors: false
positives and false negatives. AUC score works as a quan-
3.2 Precision titative measure of the ROC curve based evaluation metric.
It tells us how much a model is capable of distinguishing
Precision, also known as Positive Predictive Value (PPV), between classes. A higher value of AUC means the model is
represents the classifier’s capability to predict positive cases better at class prediction. The readers are suggested to read
as the positive. It is calculated using Eq. 2. the article by Carrington et al. [32] for a more insightful
TP explanation of the same.
P= (2)
TP + FP
4 Cancer Prediction Methods
3.3 Recall
In the current section, we have mostly discussed recent DL
Recall, alternatively known as sensitivity, is the fraction of and ML aided cancer detection methods. Here, three types
positive samples that are classified by the model correctly of cancer viz., Breast Cancer, Lung Cancer and Leukemia
with respect to the total number of actual positive cases pre- prediction systems have been discussed.
sent in the test dataset. Recall is calculated using Eq. 3.
TP
R= (3)
TP + FN

13
5534 S. Malakar et al.

Table 3  Performance Work Ref. Technique Used Performance metrics

comparison of recent prediction
models on WBC datasets ACC​ P R F1 S

Zheng et al. [234] K-means aided feature transform + SVM 0.94 – 0.96 – 0.92
Kamel et al. [109] GWO FS + SVM 1.00 – 1.00 – 1.00
Mafarja et al. [129] Hybridized WOA and SA 0.97 – – – –
Ghosh et al. [66] SMO-X FS + k-NN 1.00 – – – –
Huang et al. [98] GA-RBF SVM (small scale data) 0.98 – – 0.99 –
RBF-SVM (large scale data) 0.99 – – 0.99 –
Kumari and Singh [122] k-NN 0.99 – – – –
Kumar et al. [121] Lazy k-star/ Lazy IBK. 0.99 0.99 0.99 0.99 –
Prakash and Rajkumar [165] HLFDA-T2FNN 0.98 – – – –
Preetha and Jinny [166] PCA-LDA + ANNFIA 0.97 – – – –
Sandhiya and Palani [186] ICRF-LCFS 0.94 – – – –
Chatterjee et al. [36] SSD-LAHC + k-NN 0.99 – – – –
Ahmed et al. [6] RTHS + k-NN 0.99 – – – –
Guha et al. [72] CPBGSA +MLP 0.99 – – – –
Guha et al. [73] ECWSA + k-NN 0.95 – – – –
Ghosh et al. [67] MRFO + k-NN 1.00 – – – –

4.1 Breast Cancer Prediction Methods first generated n (< number of attributes) possible cluster
centres using the k-means algorithm on the training dataset
We have elucidated some of the existing Breast Cancer dis- and then the distances of test samples from these n cluster
ease detection techniques present in the literature. We have centres were considered as transformed features. Later to
come across some quality research works related to both perform the disease detection, the transformed feature vec-
diagnosis reports and image based methods for Breast Can- tors were classified using the SVM classifier. The authors
cer detection which are discussed hereafter. have proved experimentally that it outperformed the Genetic
Algorithm (GA) and Particle Swarm Optimization (PSO)
4.1.1 Diagnosis Report based Methods based FS techniques while using SVM as a classifier. Kamel
et al. [109] showed the importance of a good FS technique
In literature, a significant number of research articles used for this research problem. In this work, outliers were elimi-
the Wisconsin Breast Cancer (WBC)8 dataset in Breast Can- nated utilizing the outer line method. They utilized Grey
cer detection research. It contains a diagnosis report con- Wolf Optimization (GWO) [141] for FS. This FS technique
stituting 9 attributes of 699 subjects (241 positive subjects yielded an extraordinary result using the SVM classifier. In
and 458 negative subjects). The values of all these attributes another work, Mafarja et al. [129] proposed a hybrid FS
are scaled to a range of 1 to 10 depending on their propor- approach, where the authors utilized the Whale Optimiza-
tion. A class attribute (value is 2 for benign tumor and 4 for tion Algorithm (WOA) [140] and Simulated Annealing (SA)
malignant tumor) is also present in the dataset. By thorough [116]. The authors obtained very promising results for FS
analysis of the state-of-the-art methods, we have observed where only 4 attributes from 9 were selected. In another
that there are many works on this dataset using different ML work, Fruit Fly Optimization Algorithm (FOA) [156] was
based approaches. However, in most of the cases, methods used by Shen et al. [204] for selecting diagnostic attributes.
involving SVM have outperformed the rest. Table 3 lists the FOA iteratively obtains the optimal features. Moorthy and
performance of various research efforts on the WBC dataset. Gandhi [149] proposed another hybrid FS technique using
Some of the important methods are discussed here. Analysis of Variance (ANOVA) and WOA based FS tech-
The FS based techniques are mostly employed in the last niques. With this model, the accuracy of the dataset with the
10 years for Breast Cancer detection while considering the SVM classifier increased by 2.10% while compared to SVM
WBC dataset. In 2014, Zheng et al. [234] designed a feature based classification without FS.
transform method using the k-means algorithm where they Ghosh et al. [66] introduced a swarm based FS technique
named social mimic optimizer using an X-shaped trans-
fer function (called SMO-X). With SMO-X, the authors
8
https://​archi​ve.​ics.​uci.​edu/​ml/​datas​ets/​breast+​cancer+​wisco​nsin+​ gained the maximum accuracy with the k-nearest neighbors
(origi​nal) (k-NN) classifier with the help of 33.33% of actual features.

13
Computer Based Diagnosis of Some Chronic Diseases: A Medical Journey of the Last Two Decades 5535

In another work, Prakash and Rajkumar [165] used a hybrid established that in the case of small datasets, boosted SVM
FS technique termed Hybrid Local Fisher Discrimination with polynomial kernel and GA performed the best while for
Analysis (HLFDA) [61]. After the FS, the dataset that has large datasets, SVM with RBF kernel and SVM with polyno-
a near optimal feature subset was fed into Type 2 Fuzzy mial kernel performed the best. Missing values in data can
NN (T2FNN) for classification. Preetha and Jinny [166] put a detrimental effect on the classifier. To ameliorate this,
combined two powerful feature transform strategies: Prin- Choudhury and Pal [43] introduced a novel method of miss-
cipal Component Analysis (PCA) and Linear Discriminant ing value imputation using autoencoder NN. In this method,
Analysis (LDA). They put the dataset individually in PCA first the autoencoder NN was trained on multiple datasets
and LDA first and then combined the resultant transformed like WDBC without any missing values for the attributes
features to obtain the final feature set, which was used for present there and then the learned autoencoder was used to
disease detection. Performances of three classifiers viz., predict missing values for other datasets that contain some
Adaptive Neuro-Fuzzy Inference Systems (ANFIS), SVM missing values. As an initial guess of the missing value, the
and Multilayer Perceptron (MLP) were investigated and nearest neighbor rule was used and then refined by minimiz-
out of which ANFISs outperformed others. Recently, Dey ing the reconstruction error. This was based on the hypoth-
et al. [49] proposed a hybrid FS technique called Learn- esis that a good choice for a missing value would be the one
ing Automata based Grasshopper Optimization Algorithm that can reconstruct itself from the autoencoder. Recently,
(LAGOA) where the premature convergence of Grasshop- Shaw et al. [202, 203] used methods to handle imbalance
per Optimization Algorithm (GOA) was taken care of. The class problems and subsequently improved the cancer pre-
Learning Automata (LA) was used to adjust GOA param- diction performance. In the work [203], the authors pro-
eters adaptively while a two-phase mutation technique was posed an ensemble approach to handle the class problem
utilized to increase the exploitation capability of GOA. while in [202] they solved the problem using an evolutionary
One can also find a number of hybrid FS methods [6, 36, algorithm. In [202], the ring theory-based algorithm was
67, 72, 73] where the researchers employed their method on hybridized with the PSO algorithm to select the near-optimal
WBC dataset. For example, Chatterjee et al. [36] designed majority class samples from the training set. Such an initia-
a hybrid FS method which improved the local search capa- tive improved the final Breast Cancer detection performance
bility of the Social Ski Driver (SSD) algorithm with the of the classifiers. Apart from these major works, many
help of the Late Acceptance Hill Climbing (LAHC) method. authors like Kumari and Singh [122] and Kumar et al. [121]
The method is named as SSD-LAHC algorithm. In another have given prominence to comparative studies. Kumari and
work, Ahmed et al. [6] proposed Ring Theory based Har- Singh [122] applied correlation based FS and established
mony Search (RTHS) algorithm which coupled well-known the superiority of k-NN to SVM and LR classifiers while
Harmony Search (HS) algorithm and Ring Theory based Kumar et al. [121] compared 12 classifiers, namely Adap-
Evolutionary Algorithm (RTEA). Guha et al. [72] designed a tive Boosting (AdaBoost) algorithm, decision table, J-Rip, a
Clustering based Population in Binary Gravitational Search DT classifier (J48), k-NN, Lazy K-star, LR, Multiclass Clas-
Algorithm (CPBGSA). In this work, the authors decided sifier, MLP, NB, RF and Random Tree (RT). The authors
initial population using a clustering method to overcome the recommended the use of four classifiers, namely RF, RT,
premature convergence of the Gravitational Search Algo- Lazy k-star, and k-NN that yielded the best accuracy during
rithm. In another work, Guha et al. [73] proposed Embedded their experiments.
Chaotic Whale Survival Algorithm (ECWSA) using Whale Performances of some state-of-the-art works on the WBC
Optimization Algorithm (WOA) aiming at better classi- dataset have been recorded in Table 3. These results depict
fication accuracy. They used a filter method to refine the that the better performing methods used FS techniques
selected features by WSA. Ghosh et al. in [67] compared the before classifier training. Also, the methods like Shaw et al.
performance of Manta Ray Foraging Optimization (MRFO) [202], Ghosh et al. [67], and Dey et al. [49] provide state-of-
with varying transfer functions. In this work, the perfor- the-art results on UCI Breast Cancer Wisconsin (Diagnostic)
mances of four variants of S-shaped and V-shaped transfer Dataset9. Hence, we can say that the use of the FS method
functions were studied. will be considered a standard norm before using a classifier
Apart from the FS based techniques, we also find the use for diagnosis report based disease prediction systems.
of other classical ML approaches. For example, Huang et al.
[98] performed a detailed comparative study between SVM
with the varying kernel (i.e., Liner, Polynomial and Radial
Basis Function (RBF)) and ensemble methods (bagging and
boosting) having SVM as classifier. They also experimented
with GA to establish the effect of FS on classification in 9
https://​archi​ve.​ics.​uci.​edu/​ml/​datas​ets/​breast+​cancer+​wisco​nsin+​
the WBC dataset. Through extensive experiments, they (diagn​ostic)

13
5536 S. Malakar et al.

Fig. 6  Sample images from BreakHIS dataset : Eight classes of Breast Cancer histopathological images

4.1.2 Pathological Image Based Methods ultrasound, magnetic resonance imaging, mammography,

and biopsy have been developed and employed in the past.
It has already been mentioned that Breast Cancer is a deadly Among these, the histopathological image based diagnosis
disease and early detection of the same is very crucial. Vari- resulting from needle biopsy has turned out to be the gold
ous diagnostic methods, including physical examination, standard in diagnosing Breast Cancer. However, DL and ML

13
Computer Based Diagnosis of Some Chronic Diseases: A Medical Journey of the Last Two Decades 5537

Fig. 7  Sample images from BreakHIS dataset: Papillary Carcinoma (malignant) tumor seen at different magnification levels

based CADDs are steadily finding their way into diagnostic [42] and DenseNet [96], all of which were extensively used
studies. Deep NN (DNN) is approaching or even surpassing in image based Breast Cancer prediction. Shallu and Mehra
the accuracy and precision of histopathological diagnosis [199] presented a comparative study employing three pre-
conducted by pathologists when identifying important fea- trained networks, VGG-16, VGG-19 and ResNet-50 for
tures in diagnostic imaging studies. fine-tuning and full training. During the preprocessing of
The DL and ML aided research works on Breast Cancer BreakHis images, data augmentation was carried out by
prognosis considered here have proposed histopathological rotation of images about their centre. The effect of image
image based classification while using the BreakHis and scaling and the influence of training data size on overall
BACH datasets mostly for conducting experiments. The system performance was emphasized using fully trained
BreakHis dataset is made up of 9,109 histopathological (where weights were randomly assigned) and fine-tuned
images (2,480 benign and 5,429 malignant images) of breast CNNs. However, overfitting caused by the enormous capac-
tumor tissue. Images have been amassed from 82 patients ity of the network was not rectified using the freezing layers
using four different magnifying factors (40X, 100X, 200X, due to space constraints. The fine-tuned pre-trained VGG-16
and 400X). Each image has been stored in PNG format with model based feature extraction along with the LR classifier
a resolution of (700 × 400) pixels. Some samples from this outperformed the rest of the pre-trained networks. In another
dataset are shown in Figs. 6 and 7. The BACH dataset con- work, Yan et al. [228] proposed a hybrid model involving
tains microscopy images that are made available through the CNN and recurrent NN (RNN) models. They released a
grand challenge under International Conference on Image larger and more diverse dataset10 consisting of 3771 Breast
Analysis and Recognition (ICIAR 2018). This dataset con- Cancer high resolution and annotated H&E stained histo-
sists of 400 Hematoxylin and Eosin (H&E) stained breast pathological images. Each image was labelled as normal,
images of dimension (2048 × 1536) labelled with one of the benign, in situ carcinoma, or invasive carcinoma. Instead of
four classes, namely normal, benign, in situ carcinoma and using the entire dataset, the dataset they used had the same
invasive carcinoma. Images have been annotated by medical size as the BACH dataset. The authors prepared a study and
experts. Some samples from the BACH dataset are shown comparative analysis report on different combinations of
in Fig. 8. patch-wise and image-wise DL based methods. In the pre-
In the last few years, CNNs have undergone a substantial processing stage, patches were extracted from each image
amount of tweaking and evolution. This has left us with a
plethora of CNN architectures like VGG-16 [205], Incep-
tion-v1 [214], Inception-v3 [215], ResNet-50 [85], Xception 10
http://​ear.​ict.​ac.​cn/?​page_​id=​1616

13
5538 S. Malakar et al.

followed by data augmentation. A fine-tuned Inception-V3 on ImageNet-37, followed by fine-tuning on BreakHis data-
was employed to extract features from each patch which set. Sudharshan et al. [212] shed light on Multiple Instance
were then fed into a bidirectional long short-term memory Learning (MIL) and Single Instance Classification (SIC)
(BLSTM) to coalesce the features. problems while dealing with the Breast Cancer detection
Coming to ResNets, Jannesari et al. [105] investigated problem. This work divulged that MIL allows obtaining
the performances of ResNet extensions and all versions comparable or better results than SIC without labeling all the
of the Inception model (i.e., v1-v4). Data augmentation images. In this work, MIL methodologies like Axis-Parallel
methods like flipping, rotating, cropping and random resiz- Hyper Rectangle (APR), diverse density, citation-k-NN,
ing were applied to BreakHis dataset images. Saturation, and SVMs with linear, polynomial and RBF kernels, non-
brightness and contrast were applied to incorporate color parametric MIL and Multiple Instance Learning based CNN
distortions. Fine-tuned ResNet-152 competently classified (MILCNN) were applied to the dataset. For each methodol-
malignant and benign cancer types with the highest accu- ogy, a grid search was used to tune their respective hyper-
racy. Unlike many studies where a single magnification level parameters. After successive evaluations, it transpired that
was used, this work used all the magnification levels i.e., the non-parametric MIL and MILCNN-APR performed
40X, 100X, 200X and 400X of BreakHis data. Rakhlin et al. adroitly and yielded the best results while APR and citation-
[172] extracted deep features using three pre-trained CNN k-NN did not perform up to the mark.
networks, namely ResNet-50, Inception-v3 and VGG-16 It is well-known that the performance of weak classifiers
from strain normalized images. To encounter the overfitting can be improved by using classifier ensemble techniques.
problem the authors utilized an unsupervised dimensional- Thus, many researchers in the literature have designed
ity reduction mechanism along with data augmentation like classifier combination techniques for image based Breast
image scaling and cropping. Recently, Gupta and Chawla Cancer detection. For example, Yang et al. [230] proposed
[77] used pre-trained CNN models, namely, VGG-16, VGG- the Ensemble of Multi Scale CNN (EMS-NN) model. To
19, Xception, and ResNet-50 to extract features from his- form the ensemble, they utilized pre-trained models of
topathological images. CNNs were trained separately on DenseNet-161, ResNet-152 and ResNet-101. The pro-
different magnification factors (40x, 100x, 200x and 400x). posed methodology followed three stages, the first being,
The BreakHis dataset was used for performing experiments. Multiscale image patch extraction followed by Training
The top layers i.e., the FC layers of the CNN model were multiple DCNNs and Model selection and combination.
removed and replaced by traditional ML methods i.e., SVM Overfitting was attenuated by applying various data aug-
and LR. Features from pre-trained ResNet-50 with LR clas- mentation techniques that were used to inflate the size and
sifier outperformed other models on 40x and a 100x magni- color diversity. However, Chennamsetty et al. [40] used two
fication factor. In another work, Dey et al. [52] used a pre- pre-trained CNN models - DenseNet-161 and ResNet-101
trained DenseNet-121 model for feature extraction purposes that were employed on differently preprocessed histology
to detect Breast Cancer from thermal images. However, images. Brancati et al. [30] designed their ensemble model
instead of passing the original image to the CNN model, with different versions of the ResNet CNN architectures.
they preprocessed the image using edge detection methods. The notable contribution of the work is that the authors
The authors stacked edge images generated by Prewitt and reduced the problem complexity using a down-sampling
Roberts edge detection technique with the original gray-level technique where the image size was reduced by a factor of
image to generate a three channel image. On the other hand, k. It also used the central patch of size m ∗ m as input to the
Sánchez-Cauce et al. [185] proposed a multi-input CNN model which reduced the training complexity of the model
model for detection of breast cancer from multi-modal data further. Recently, Bhowal et al. [28] proposed a classifier
consist of patients’ information along with thermal images. combination method that used Choquet fuzzy integral as the
The above mentioned methods utilized the existing aggregator function. The aggregator was used to combine
CNN architectures to obtain better results. However, some the confidence scores returned by the CNN based classifi-
instances are found where researchers designed their own ers. The most notable aspect of the work is that the authors
CNN architecture instead of using the popular CNN archi- utilized coalition game and information theory to estimate
tectures as mentioned earlier. For example, Han et al. [79] the fuzzy measures used in Choquet integral with the help
designed Class Structure-based Deep CNN (CSDCNN) for of validation accuracy. In another work, Chouhan et al. [44]
the classification of Breast Cancer. Training instances were proposed emotional learning inspired feature fusion strategy
augmented to get rid of the imbalanced class problem. In to classify a mammogram image into normal and abnormal
their experiments, they made use of two different training classes. The authors used three static features: taxonomic
strategies. In the first case, they trained the CSDCNN from indices, statistical measures and LBP along with CNN
scratch on the BreakHis dataset while the second case was based features. Chattopadhyay et al. [38] designed a DL
based on transfer learning that initially pre-trains CSDCNN based method called Dense Residual Dual-shuffle Attention

13
Computer Based Diagnosis of Some Chronic Diseases: A Medical Journey of the Last Two Decades 5539

Table 4  Performance Work Ref. Classifier Performance metrics

comparison of recent prediction
models on various Breast ACC​ P R F1 S AUC​
Cancer datasets. Classifiers with
’*’ represent the classifiers with BreakHis Dataset
the highest ACC scores in the Han et al. [79] CSDCNN 0.93 – – – – –
respective paper Sudharshan et al. [212] MILCNN* 0.92 – – – – –
Shallu and Mehra [199] VGG 16 + LR 0.92 0.93 0.93 0.93 – 0.95
Jannesari et al. [105] ResNet V1 152 0.98 0.99 – – – 0.98
Gupta and Chawla [77] ResNet50+LR 0.93 – – – – –
Chattopadhyay et al. [38] DRDA-Net 0.98 – – – – –
BACH Dataset
Rakhlin et al. [172] LightGBM + CNN 0.87 – – – – –
Yang et al. [230] EMS-Net 0.91 – – – – –
Roy et al. [178] Self-designed (OPOD) 0.77 0.77 – 0.77 0.77 –
Roy et al. [178] Self-designed (APOD) 0.90 0.92 – 0.90 0.90 –
Sanyal et al. [189] Hybrid Ensemble (OPOD) 0.87 0.86 0.87 0.86 0.99 –
Sanyal et al. [189] Hybrid Ensemble (APOD) 0.95 0.95 0.95 0.95 0.98 –
Bhowal et al. [28] Choquet fuzzy integral and 0.95 – – – – –
coalition game based classifier
ensemble
Mics. Dataset
Yan et al. [228] Inception-V3* 0.91 – 0.87 – – 0.89
Dey et al. [52] DenseNet-121 0.99 0.99 0.98 0.98 – –

Fig. 8  Examples of different forms of microscopic biopsy images taken from the BACH dataset

13
5540 S. Malakar et al.

Network (DRDA-Net) for detection of Breast Cancer detec- 4.2 Lung Cancer Prediction Methods
tion from histopathological image.
All the methods described above used the entire image Like Breast Cancer, Lung Cancer detection methods are also
to detect the presence of Breast Cancer using the patho- divided into two categories in this work: diagnosis report
logical image i.e., all these methods passed the entire image based and pathological image based methods. Some research
to the detection system. However, multi-view analysis of attempts produced satisfactory results with ML algorithms
the pathological images might help in improving the final like SVM and AdaBoost. However, in most of the works,
result. With this objective, some authors followed patch- clear domination of DL based techniques is found for image
based approaches for the said task. For example, Roy et al. based as well as diagnosis report based methods.
[178] designed a CNN aided patch-based classification
model for classifying histology breast images. Patches car- 4.2.1 Diagnosis Report based Methods
rying distinguishing information were extracted from the
original images to perform classification, which followed The dataset that is very commonly used in diagnosis report
two different evaluation strategies, namely One Patch in One based Lung Cancer detection is from UCI-Irvine reposi-
Decision (OPOD) and All Patches in One Decision (APOD). tory11. It contains 32 instances of data where the num-
In another work, Sanyal et al. [189] proposed a similar detec- ber of attributes is 57 (1 class attribute and 56 predictive
tion like Roy et al. [178]. However, in this work, the authors attributes). It is an age-old dataset and contains a very lim-
utilized a novel hybrid ensemble approach for classification ited number of samples with a large number of diagnosis
purposes while Roy et al. [178] used a single CNN archi- attributes makes it a challenging dataset. However, some
tecture. This ensemble model used the confidence score works used Lung Cancer Data12 (LCD), survey Lung Can-
returned by base CNN models and the confidence scores cer Data13 (SLCD) of data from world repository, Thoracic
returned by the Extreme gradient boosting trees (XGB) clas- Surgery Dataset (TSD) of UCI repository [23], Michigan
sifier with the help of different features to make the final Lung Cancer Dataset14 (MLCD) and SEER15 dataset. LCD
decision at the patch level. Finally, the patch level results are dataset contains 1000 instances with 24 attributes having
again ensembled to take the image level decision. three different class labels: ’0’ is for a healthy person, ’1’
All the above discussed works mostly used deep fea- for a person with a benign tumor and ’2’ for a person with
tures that might contain some redundant/irrelevant fea- a malignant tumor. SLCD contains 309 instances but with
tures. Therefore, the use of some FS methods might help 16 attributes. Out of these 16 attributes, 14 attributes have
in improving the final output. Based on this, Chatterjee values 1 and 2 to represent NO and YES respectively while
et al. [37] proposed a deep feature selection technique. In for the gender M (for male) and F (for female) are used.
this method, the authors improved the Dragonfly Algorithm The values of attribute age are normalized by the min-max
(DA) with the help of the Grunwald-Letnikov method to normalization method. The TSD contains 17 attributes (14
perform FS. By doing this the authors improved the model’s having nominal values and 3 having continuous values) and
performance with fewer features. In computer vision, it is 470 instances. In this dataset class label, the attribute is rep-
also observed that fusing handcrafted features with deep resented by RiskYr which is a binary attribute. Risk1Yr is
features improves the overall performance. Relying on this true if the person is dead and false if the person is alive.
observation, Sethy and Behera [195] first concatenated LBP MLCD contains 96 instances (86 instances are cancerous
features with deep features extracted using the VGG19 CNN and 10 are healthy) and 7130 attributes. SEER dataset is a
model and then classified them using the K-NN classifier. gene expression based dataset which was released in April
The performances of some important state-of-the-art Breast 2016. It contains 149 attributes and 643,924 instances. The
Cancer detection methods on different datasets are summa- dataset named lungdata lists the performance of various
rized in Table 4. The performance of state-of-the-art meth- research efforts on various datasets discussed here. The
ods on different datasets is satisfactory but considering its non-availability of the diversified dataset for Lung Cancer
deadliness characteristics it remains an open research prob- disease detection for a long period might lessen the quality
lem. After analysing the performance trends of the methods of research works as compared to other cases.
studied here, we can safely comment that patch based meth-
ods assisted with DL performed better on BACH dataset.
So, in the future, this approach could be studied on other 11
http://​archi​ve.​ics.​uci.​edu/​ml/​datas​ets/​Lung+​Cancer
datasets even for designing a better Breast Cancer detection 12
https://​data.​world/​cance​rdata​hp/​lung-​cancer-​data
technique from pathological images. 13
https://​data.​world/​sta42​7ceyin/​survey-​lung-​cancer/
14
http://​leo.​ugr.​es/​elvira/​DBCRe​posit​ory/​LungC​ancer/​LungC​ancer-​
Michi​gan.​html
15
https://​rdrr.​io/​github/​cezden/​parSE​ER/​src/R/​dataL​oad.R

13
Computer Based Diagnosis of Some Chronic Diseases: A Medical Journey of the Last Two Decades 5541

Table 5  Performance Work Ref. Classifier Performance Metrics

comparison of recent prediction
models on various Lung Cancer ACC​ P R F1 Error AUC​
datasets
UCI -TSD Dataset
Danjuma [47] MLP 0.82 0.82 0.82 0.82 – 0.84
MLCD Dataset
Murty et al. [151] RBF-NN – – – – 0.19 –
UCI-Irvine Dataset
Patra [161] RBF-NN 0.81 0.81 0.81 0.81 – 0.75
Dey et al. [49] LAGOA+RF 0.86 – – – – –
LCD Dataset
Radhika et al. [169] SVM 0.99 – – – – –
Salaken et al. [183] Deep-ANN 0.80 – – – – –
Maleki et al. [132] Deep-ANN 1.00 – – – – –
SEER Dataset
Ali and Reza [180] RIPPER with AdaBoost 0.89 – – – – 0.95
Doppalapudi et al. [53] ANN 0.71 0.71 0.71 0.71 – 0.87
SLCD Dataset
Nasser and Abu-Naser [153] ANN 0.97 – – – – –

Most of the works in the literature performed experimen- best. The author applied the classifier to preprocessed data
tal comparative studies consisting of different shallow classi- obtained after inputting the missing values.
fiers. For example, Danjuma [47] compared the performance In addition to these comparative studies, there are a few
of three different classifiers: MLP, J48 and Naive Bayes works that have concentrated on method building. For exam-
(NB) on TSD from the UCI repository. The outcome of their ple, Ali and Reza [180] employed different preprocessing to
study showed that MLP produced the best result on these filter out redundant/noisy patient data as well as attributes
datasets. In another work, Murty and Babu [151] consid- from the SEER dataset. In the first phase of preprocessing,
ered four popular classifiers viz., NB, NN with RBF kernel the data of the patient who died due to cancer were only kept
(RBF-NN), MLP and DT in their work to study the perfor- and then some irrelevant attributes like the cause of death,
mances of these classifiers on two datasets: UCI repository ID_no, age and sex were removed. Later, a correlation based
and MLCD. RBF-NN outperformed all other classifiers. FS algorithm was employed to obtain the near optimal set of
Radhika et al. [169] used the Lung Cancer dataset from the attributes. After the preprocessing, the researchers selected
UCI repository and SLCD. In their research, they observed 46 attributes from 149 attributes and 17,484 instances from
that LR with a 7-fold cross validation regime and SVM with 643,924 instances. They applied various ensemble methods
a 10-fold cross validation strategy outperformed the other to this optimally selected dataset. Three classifiers: J48 with
classifiers on the UCI repository and SLCD dataset respec- the base learner, RF with Dagging and Repeated Incremental
tively. Recently, Patra [161] compared the performances of Pruning to Produce Error Reduction (RIPPER) with Ada-
four different classical ML algorithms: RBF-NN, k-NN, NB Boost were used and RIPPER with AdaBoost performed
and J48 on the Lung Cancer dataset of the UCI repository. best.
The experimental outcomes confirmed that NN produces In another work, Nasser and Abu-Naser [153] used the
the best accuracy with a 10-fold cross validation technique. SLCD dataset for their work. Out of 16 attributes, in this
Recently, Doppalapudi et al. [53] compared the performance research, the author used the first 15 attributes but excluded
of 6 classifiers (3 deep learning models, namely ANN, CNN the chest pain attribute. ANN with 3, 1 and 2 nodes in the
and RNN, and 3 shallow learners, namely, NB, SVM, RF) hidden layers was used. Salaken et al. [183] introduced a
for survivability prediction of Lung Cancer prediction using deep autoencoder based classification method. In this model,
SEER dataset. They experimentally showed that ANN per- deep features were first learned and then trained an ANN
formed the best among these classifiers while RNN and with the learned features. Hinton and Salakhutdinov [87]
CNN models followed the ANN classifier. In another work, used a DNN model with 80 and 10 features in the first and
Gultepe [75] made a study of six different classifiers namely, second layer respectively and the number of hidden neu-
k-NN, RF, NB, Logistic Regression (LR), DT, and SVM on rons=5. They used Sigmoidal as an encoder function and
UCI- Irvine repository and found that K-NN performed the got a promising result. Recently, Dey et al. [49] proposed
an FS method termed LAGOA for several disease diagnosis

13
5542 S. Malakar et al.

purposes. The selected features from LAGOA are classified Anifah et al. [16] used ANN with back-propagation and
using an RF classifier. The model was evaluated on the UCI- Gray Level Co-occurrence Matrix (GLCM) based features
Irvine Lung Cancer dataset. In another recent work, Meleki to classify 50 CT images obtained from the CIA. During
et al. [132] used a K-NN classifier along with GA based FS prepossessing, the images were binarized and enhanced like
approach for lung cancer prediction from the LCD dataset. in [152]. Finally, GLCM features were extracted from ROIs
Recently, Hsu et al. [95] proposed an FS scheme by com- to classify the images using the designed ANN. Vas and
bining GA with correlation based FS technique to select an Dessai [222] also proposed a Haralick feature assisted Lung
optimal set of features from UCI- Irvine repository. Cancer detection technique where image segmentation was
Some of the important findings by the researchers are performed using morphological operation. Images were col-
listed in Table 5. Like diagnosis report based Breast Cancer lected from the Manipal hospital and V.M.Salgaocar hospital
detection here also FS based methods dominate. However, both situated in Goa, India. The authors applied a median
to the best of our knowledge, methods assisted with FS and filter of size 3 × 3 to improve the input CT image quality.
deep learners for Lung Cancer diagnosis are still missing For ROI segmentation a four steps method was employed.
in the literature. However, the use of such methods could The steps are 1) complementing the image and opening the
improve the diagnosis performance of the ML/DL based image with periodic-line structuring elements, 2) filtering
method significantly. out the lungs by using maximum area, 3) performing a close
operation by disk structuring element to procure lung mask
4.2.2 Pathological Image Based Methods and 4) superimposing the lung mask on the original image.
7 Haralick features (second order statistical features) were
The practicalities of ML and DL based models have already extracted by applying Haar wavelets and creating GLCM
been discussed in the previous sections. With the rapid from the wavelet transformed images. Recently, Shakeel
advances in computational intelligence and GPUs, it has et al. [198] used deep learning instantaneously trained neu-
become easier for researchers to develop robust image classi- ral network (DITNN) classifier on top of several statistical
fiers. As a result, we find many recent methods that employed and spectral features designed by Sridhar et al. [209] aim-
image classification protocol for Lung Cancer detection from ing at designing a Lung Cancer detection model using CT
Computed Tomography (CT) images. In the period before images. The main contribution of the authors is proposing an
wider acceptability of transfer learning for learnable texture improved profuse clustering technique (IPCT) that segments
feature extraction, researchers [16, 117, 152] mostly used the affected regions from the CT image efficiently and thus
preprocessing to improve the visual quality of the CT images improves the overall detection result. Alzubaidi et al. [13]
followed by handcrafted feature extraction to classify with performed an experimental comparison between global and
the help of shallow learners. However, with the advent of local hand-crafted features used for Breast Cancer detection
the transfer learning concept, the research spectrum became from CT images. For the comparative study, 100 CT images
wider. For example, Kulkarni and Panditrao [117] utilized the were collected from different sources while 10 different fea-
CT images obtained from the National Institutes of Health ture extraction processes, namely intensity histogram, HOG,
(NIH) or National Cancer Institute (NCI) Lung Image Data- Gabor Haar wavelet, etc. This comparative study confirms
base Consortium (LIDC) datasets for detecting the stages of that the use of local Gabor features with the SVM classifier
Lung Cancer. In this work, before extracting the features, the performed the best. Recently, Akter et al. [7] designed a
quality of the CT images was enhanced using Median filter fuzzy rule based image segmentation technique to extract
followed by the Gabor filter. Next, image segmentation was suspicious nodules from a CT image. Next, they prepared
performed using marker based watershed segmentation [9] shape (2D and 3D) and texture (2D) based features from the
to extract the tumor regions. Finally, geometric features like extracted nodules and classified them using a neuro fuzzy
area, perimeter and eccentricity were fed to the SVM classi- classifier. In another work, Zhou et al. [236] first employed
fier to determine the stages of cancer. Nadkarni and Borkar a region growing mechanism to segment suspicious regions
[152] classified CT images as cancerous (abnormal) and non- from a CT image and then a CNN model to classify the
cancerous (normal) using an SVM classifier placed on top of segmented regions to decide the presence of Lung Cancer.
several extracted geometric features that include area, perim- In the era of DL with the concept of transfer learning, we
eter and eccentricity. Prior to the extraction of features, the observed some works that utilized the same in their work.
median filter was applied to eradicate the noise from the CT For example, Jakimovski and Davcev [104] gathered CT
images. Down the line, image enhancement was carried out images of 70 subjects from a local medical hospital and
using contrast adjustment, followed by image segmentation. labelled them from oncology specialists. Two piles were cre-
Annotated CT images of Lung Cancer were collected from ated, where 58 subjects had cancer, and 12 were diagnosed
the Cancer Imaging Archive (CIA) repository. Two sample cancer-free. The piles were split into training (90% of the
images taken from the CIA database are shown in Fig. 9. images) and testing sets. After binarization, images were

13
Computer Based Diagnosis of Some Chronic Diseases: A Medical Journey of the Last Two Decades 5543

Table 6  Performance Work Ref. Classifier Performance metrics

comparison of recent prediction
models on Lung Cancer ACC​ P R F1 S
Datasets
CIA
Tekade and Rajeswari [218] VGG 0.95 – – – –
Anifah et al. [16] DNN 0.80 – – – –
Alam et al. [10] SVM 0.97 – – – –
Shakeel et al. [198] DITNN 0.95 0.94 0.97 0.95 0.97
Zhang and Kong [231] MSDLF 0.99 – – – –
Mics. Datasets
Jakimovski and Davcev [104] DNN 0.75 – – – –
Taher et al. [216] Rule-based classifier 0.93 0.95 0.94 – 0.91
Vas and Dessai [222] DNN 0.92 – 0.88 – 0.97
Alzubaidi et al. [13] SVM 0.97 – 0.96 – 0.97
Akter et al. [7] Neuro-Fuzzy 0.90 0.86 – – 0.81

Fig. 9  Examples of Lung CT

images from CIA

fed into the DNN for classification. In another work, Taher The detected nodules were classified using a 4-channel CNN
et al. [216] came up with a rule based method that analyzed architecture. The summary of performances of all the meth-
the sputum samples obtained from the Tokyo Center of Lung ods studied here is provided in Table 6.
Cancer in Japan. The features extracted from the nucleus
region included nucleus-cytoplasm ratio, perimeter, density, 4.3 Leukemia Prediction Methods
curvature, circularity and eigen ratio. Diagnosis rules for
each of the features were derived. 100 sputum color images Considering the mortality rate of Leukemia, research-
were used to assess the rule based method. Finally, classifi- ers have made significant inroads into the early detection
cation was done using a rule based method. While Kulkarni of Leukemia using both gene expression and pathologi-
and Panditrao [117] used classical ML in their work, Tekade cal image. All genes present in gene expression data are
and Rajeswari [218] proposed a 3D multipath VGG-like not responsible for Leukemia, rather there are a few which
CNN architecture. In the preprocessing stage like image known as biomarkers. Thus, selecting the responsible genes
binarization, morphological operations were applied. After was vastly explored in literature. Old days method used filter
preprocessing, U-Net architecture was used for the segmen- methods while the current trend is to use different hybrid FS
tation of lung nodules, which were fed to the CNN model mechanisms for the said purpose. In the case of image based
for detection. Finally, the features were fed to an ANN for methods, a shift from the traditional ML approach to the
classification needs. Since identified lung nodules greatly DL approach is observed. In the following subsections, we
help in risk assessment of Lung Cancer, Zhang and Kong
[231] designed a Multi-Scene Deep Learning Framework
(MSDLF) to detect Lung Nodule from Lung CT images.

13
5544 S. Malakar et al.

Table 7  Performance Work Ref. Classifier Performance metrics

comparison of recent prediction
models on Leukemia Datasets. ACC​ R F1 S
All the methods were evaluated
on the Dataset provided by Hsieh et al. [94] IG-SVM 0.98 – – –
Golub et al. [70] Begum et al. [27] CBFS-Poly SVM 0.97 – – –
Kavitha et al. [114] FCBF-SVM RFE 0.93 0.95 – 0.98
Dwivedi [55] ANN 0.98 1.00 – 0.93
Gao and Liu [63] PSOFOA-ANNFIS 1.00 – – –
Santhakumar and Logeswari [187] ALO-SVM 0.91 0.95 0.93 0.95
Santhakumar and Logeswari [188] ALO-ACO-SVM 0.95 0.95 0.96 0.98
Baldomero-Naranjo et al. [24] Modified SVM 0.95 – – –
Mandal et al. [133] Tri-state FS 1.00 – – –

describe the methods that followed either gene expression (ALO), designed by Mirjalili [139]. After FS, the SVM clas-
or cell image based diagnosis. sifier was applied to the selected features. In another work
[188], these authors proposed a hybridized ALO with Ant
4.3.1 Gene Expression Based Methods Colony Optimization (ACO) technique to improve their
previous performance. Though these two methods are new
In this study, the classification of Leukemia into Acute approaches, their performance was not so good compared
Lymphoblastic Leukemia (ALL) or Acute Myeloid Leuke- to other works on the said dataset. In another work, Gao
mia (AML) is considered. The most common dataset which and Liu [63] gained the highest accuracy with the help of
has been used is the dataset16 provided by Golub et al. [70]. an extension SVM classifier termed Least Square SVM
This is a gene expression dataset. It consists of 72 Leuke- (LSSVM) where the authors first normalized the data and
mia patients (47 in the ALL category and 25 in the AML then classified using LSSVM. Next, the F-statistic method
category) and each data consists of 7129 gene expression was used as a filter method to lessen the complexity of SVM-
measurements. The researchers aimed to come up with a RFE. Both PSO and FOA tend to get stuck in local optima.
near optimal set of genes that are responsible for Leukemia Hence, FOA was used first and then PSO was used to reach
disease prior to designing a classifier based detection tech- the global optimum if FOA stuck to the local optimum.
nique. For example, Hsieh et al. [94] used Information Gain Finally, LSSVM is used on the select features and they got
(IG) to select important genes that play a significant role in an extraordinary result of 100% with 4 features.
identifying the presence of Leukemia. The authors selected Recently, Baldomero-Naranjo et al. [24] designed a modi-
150 genes out of 7128 genes using the IG. These 150 genes fied SVM classifier that deals with outliers detection and FS
were later used to classify genes using SVM with the RBF simultaneously. They considered ramp loss margin error [31]
kernel. In another work, Begum et al. [27] performed Con- in the newly designed SVM model to mitigate the influence
sistency based FS (CBFS) to select optimal bio-markers of outlier on the classifier and a budget constraint approach
(i.e., genes) that helped distinguish ALL from AML with similar to [123] to restrict the number of the selected fea-
SVM (with polynomial kernel) classifier. Contrary to the tures. In another work, Baliarsing et al. [25] designed a
above methods, Kavitha et al. [114] used Fast Co-relation hybrid FS method that combines SA with Rao Algorithm
based Filter-Solution (FCBF) as feature selector and SVM (RA) (termed as SARA). The SA based local search helps
with Recursive Feature Elimination (RFE) kernel as the in improving the exploration capability of the RA method.
classifier. In FCBF, first Symmetrical Uncertainty (SU) was They also used Log sigmoidal function as a transfer function
computed for every feature and then the correlated features to convert the continuous domain SARA into the discrete
were deleted based on a preset threshold value. domain. Mandal et al. [133] proposed a tri-state wrapper-
In another work, Dwivedi [55] made a comparison study filter FS method and evaluated the performance on the data-
among many classifiers like ANN, SVM, LR, k-NN and clas- set provided by Golub et al. [70]. In the first stage, four filter
sification tree. For cross-validation, researchers used to leave methods, namely Mutual Information, ReliefF, Chi-Square,
one out cross validation. The best result came out for ANN and Xvariance were considered to design an ensemble FS
with the leave one out cross validation technique. Recently, that reduces some irrelevant features from high-dimensional
Santhakumar and Logeswari [187] used Ant Lion Optimizer features. In the next stage, the highly correlated features

16
https://​web.​stanf​ord.​edu/​~hastie/​CASI_​files/​DATA/​leuke​mia.​html

13
Computer Based Diagnosis of Some Chronic Diseases: A Medical Journey of the Last Two Decades 5545

Fig. 10  Samples taken from

ALL_IDB images

Fig. 11  Example of a periph-

eral blood smear having (a)
lymphoblast and (b) lymphocyte
(ALL_IDB Dataset)

(a) Lymphoblast (b) Lymphocyte

Fig. 12  Separation of nucleus

and cytoplasm exercised by
Mishra et al. [143] in their work

13
5546 S. Malakar et al.

were filtered out by employing the Pearson correlation based isolated leukocyte cells and classified using SVM, NB,
method. Finally, the authors employed WOA to select the k-NN and NN with Back Propagation (BPNN). It was found
final set of biomarkers. Table 7 lists the performance of vari- that the first 50 DCT coefficients with the SVM classifier
ous research efforts discussed here. performed the best. Next year, in the work [143], they first
extracted GLCM features from segmented cells and then
4.3.2 Pathological Image Based Methods they employed probabilistic PCA for feature dimension
reduction. Finally, they used RF, k-NN, SVM and BPNN
Since Leukemia disease usually involves white blood cells classifiers on dimension reduced features for both nuclei as
and thus the nature of these cells helps in screening it. As well as cytoplasm detection from which it was inferred that
a result, the pathological image based Leukemia detection nucleus features were more suitable for accurate detection
algorithm relied on understanding the nature/proportion of of Leukemia. In another work [144], the authors performed
blood cells. For this category of Leukemia classification texture based cell classification for predicting Leukemia.
methods considered here have used the Acute Lymphoblas- Texture based features were extracted using Discrete Ortho-
tic Leukemia Image Database (ALL_IDB17). All images normal S-Transform (DOST) from each segmented region.
are in JPG format with 24-bit color depth and dimension This was followed by dimensionality reduction using LDA.
2592 × 1944 . Some sample images of ALL_IDB dataset For classification, the authors tried AdaBoost, SVM, BPNN,
are shown in Figs. 10 and 11. The dataset has two parts: RF and k-NN out of which AdaBoost yielded the best result.
the ALL_IDB 1 and ALL_IDB 2. The ALL_IDB 1 data- Similar to Mishra et al. [142–144], Shafique et al. [196,
set with 108 samples is used primarily for differentiating 197] also proposed two different solutions for classifying
patients from non-patients. It contains about 39000 blood Leukemia but they used state-of-the-art transfer learning
cells, where the lymphocytes have been labelled by expert models for feature extraction purposes in [196] while in
oncologists. The ALL_IDB 2 dataset is primarily used to [197] they preferred hand-crafted color and shape based fea-
classify different types of Leukemia. It contains a collection tures and SVM classifier. In transfer learning, the last layer
of cropped areas of interest of normal and blast cells (see was changed first to detect white blood cells in the images
Fig. 11) that belong to the ALL_IDB1 dataset (see Fig. 10). and then to 4 output channels to categorize the subtypes of
ALL_IDB2 images have similar grey level properties to the Leukemia. This process was done for RGB, HSV, YCbCr
images of the ALL_IDB1, except for the image dimensions. and HCbCr image types and used data augmentation during
Hereafter we are discussing some significant works in Leu- training. The results depicted that the best classification was
kemia disease detection. observed in the case of RGB. In another work [197], the
Abdeldaim et al. [2] utilized the ALL_IDB 2 dataset to authors performed white blood cell extraction using a series
detect different types of Leukemia disease. In this work, the of preprocessing steps (conversion to CMYK and histogram
images were first converted into a CMYK color model and equalization based contrast enhancement), white blood cells
then used histogram based threshold calculation method to suppression using Zack’s algorithm, segmentation of image
separate the lymphocyte cells from the non-lymphocyte cells using watershed algorithm prior to classification. Whereas,
where roundness and solidity characteristics of the cells were Rahman and Hasan [170] detected white blood cells in four
considered. After this, 30 shape-based features, 15 color major steps which involved the conversion of the image from
based features and 84 texture based features were extracted RGB to HSV, histogram equalization based contrast stretch-
from detected lymphocyte cells and stacked to classify the ing, background removal and detection using the watershed
cells using k-NN, SVM (with RBF, Polynomial and Linear algorithm. The authors then performed image cleaning to
kernels), DT, NB classifiers. k-NN outperformed the other remove leukocytes situated at the edges and nucleus and
classifiers. Mishra et al. [142–144] proposed a series of solu- cytoplasm separation using the bounding box technique.
tions in the said field with varying features and classifiers. In After this, the authors performed feature extraction to obtain
all these works, they utilized the Weiner filter followed by 70 features: 16 morphological features, 36 texture features
histogram equalization based contrast enhancement on the and 18 color features. The classification was performed
images from ALL_IDB 2 dataset. Next, grouped leukocytes using an ensemble of classifiers having SVM, DT and k-NN
were segmented using marker based watershed segmentation as base classifiers.
algorithm [125]. Some such segmented outputs are shown In contrast to the above methods, Rawat et al. [176] used
in Fig. 12). the French American British (FAB) classification system
In the work, proposed by Mishra et al. [142], Discrete for ALL_IDB 2 dataset. According to the FAB, there are
Cosine Transform (DCT) features were extracted from three types of Leukemia in the ALL_IDB 2 dataset (say,
L1, L2 and L3). Based on this, the authors designed a hier-
archical classification model. In the first stage, normal cells
17
https://​homes.​di.​unimi.​it/​scotti/​all/ and cancerous cells were classified. In the next stage, from

13
Computer Based Diagnosis of Some Chronic Diseases: A Medical Journey of the Last Two Decades 5547

Table 8  Performance Work Ref. Classifier Performance metrics

comparison of recent prediction
models on ALL_IDB dataset ACC​ P R F1 S

ALL_IDB 1
Mishra et al. [142] SVM 0.99 0.99 1.00 0.99 0.99
Shafique et al. [197] SVM 0.93 – – – –
Rahman and Hasan [170] SVM+DT+k-NN 0.93 – 0.96 – –
Mohammed et al. [145] SVM 0.97 – – – –
ALL_IDB 2
Abdeldaim et al. [2] k-NN 0.96 – – – –
Mishra et al. [142] SVM 0.89 – 0.84 – 0.94
Shafique et al. [196] DNN 0.99 – – – –
Rawat et al. [176] ANFIS 0.96 – – – –
Mishra et al. [143] RF 0.99 0.97 1.00 0.98 0.98
Sahlol et al. [181] SVM 0.98 1.00 0.97 0.985 0.97

Table 9  Performance Work Ref. Classifier Performance metrics

comparison of some prediction
models on Cleveland dataset ACC​ P R F1 S ERR

Single classifier based methods

Al-Milli [8] BPNN 0.92 – – – – –
Sonawane and Patil [208] BPNN 0.985 – 0.98 – 0.95 0.02
Gavhane et al. [64] MLP – 0.91 0.89 – – –
Karayılan and Kılıç [113] BPNN 0.96 0.95 0.95 – – –
Ismaeel et al. [100] ELM-NN 0.87 – – – – –
Medhekar et al. [136] NB 0.90 – – – – –
Jabbar et al. [101] NB 0.86 – – – – –
Patel et al. [158] J48 – – – – – 0.15
Dwivedi [56] LR 0.85 0.85 0.89 – 0.81 –
Sen [194] SVM 0.84 – – – – –
Shao et al. [200] MARS-LR 0.84 – – – – –
Jayaraman and Sultana [106] PBAMNN 1.00 – 1.00 – 1.00 –
Gokulnath and Shantharajah [69] GA-SVM 0.88 – – – – –
Feshki and Shijani [60] PSO-FFBP 0.92 0.92 0.93 – – –
Harimoorthy and Thangavelu [81] Improved SVM 0.90 0.81 0.97 – 0.87 0.10
Ensemble based methods
Das et al. [48] Multi-NN 0.89 – – – – –
Miao et al. [138] Boosting 0.97 0.97 1.00 0.98 0.60 0.03
Mohan et al. [146] HRFLM 0.88 0.90 0.93 0.90 0.83 0.12
Maji and Arora [130] DT-NN 0.78 – 0.78 – – 0.23
Amin et al. [14] Average voting 0.87 – – – – –

amongst the cancerous cells, L1 and non-L1 cells were based on classification accuracy. Mohammed et al. [145]
classified. Similarly, L2 and L3 were classified in the third proposed a two-stage method where first white blood cells
stage. The authors first extracted 11 geometrical, 15 chro- were segmented using a series of preprocessing techniques
matic and 45 statistical texture features from preprocessed like median filtering, histogram equalization and hard
images followed by PCA to obtain reduced feature vectors. thresholding and then features like shape, statistical, tex-
In each stage of the hierarchical classification model k-NN, ture and DCT features were extracted from the segmented
Probabilistic NN (PNN), SVM, Smooth SVM (SSVM) and images to classify the cells as abnormal or healthy. For
ANN based Fuzzy Inference System (ANFIS) were applied. classification, the authors used an SVM classifier. Recently,
The experimental outcome showed that SVM performed best Sahlol et al. [181] proposed a wrapper FS method termed

13
5548 S. Malakar et al.

as statistically enhanced Salp Swarm Algorithm (SESSA), classifier based systems include the use of variants of NN
where SVM was used as a classifier, to select deep features, models, FS technique prior to classification, missing value
extracted using VGG architecture, for the classification of imputation, and comparative study to come up with a better
pre-segmented white blood cells. Recently, Sumi et al. [213] model. Since all the diagnostic attributes present in the data-
proposed a self-made CNN-architecture to detect presence of set might not be essential always while screening the Heart
Leukemia. To train the model, they generated synthetic sam- disease, hence many authors have utilized FS algorithms to
ples through a data augmentation process. In another work, build a state-of-the-art prediction model. Relatively older
Khandekar et al. [115] utilized an object detection algorithm methods that utilized some form of feature reduction mecha-
[147] known as the You Only Look Once (YOLO) algorithm nism mostly used the filter FS techniques. Next, we observe
(version 4) to extract and classify ROIs simultaneously. a shift from filter to wrapper FS methods and at present
Performance comparison of all the methods discussed hybrid FS techniques are being proposed. Many works also
above has been listed in the Table 8. It is to be noted that consider this UCI Heart disease dataset to show the effec-
these performances were not obtained following uniform tiveness of newly developed FS models.
experimental setups and thus are not directly comparable. In the literature, there are a large number of methods that
However, after closely analyzing the results, it is observed used NN based models. A number of works like Al-Milli
that hand-crafted features provided better performance over [8], Sonawane and Patil [208] and Gavhane et al. [64] cited
deep feature techniques [181, 196]. The reason might be the state-of-the-art results on Heart Disease classification
that the CNN models used in literature are of larger depth problem by tuning the parameters or increasing the number
and thus tend to extract texture features which in turn failed of hidden layers in an MLP with backpropagation algorithm.
to catch the shape information. Hence, in the future, the use For example, in [8] though the iteration size is fixed there
of some better CNN models that can highlight shape and were 3 hidden layers with 8, 5, and 2 neurons. Whereas in
texture information may be useful. Even the use of FS and some experiments only one hidden layer with a different
classifier ensemble techniques are not been largely explored number of neurons produced comparatively better results.
to date. For example, the numbers of neurons in hidden layer were
20, 100 and 8 in Sonawane and Patil [208], Gavhane et al.
[64] and Karayılan and Kılıç [113] respectively.
5 Heart Disease Prediction Methods Jabbar et al. [101] and Kanchan et al. [110] established
the benefits of FS for classifying Heart Disease. Jabbar et al.
One of the common datasets which have been used by many [101] utilized the One-R FS technique, proposed by Holte
researchers is the Heart disease dataset from the UCI reposi- [88] produced better results (combining with NB) compares
tory18. This is a diagnostic report based dataset that con- to Chi-square, Gain Ratio and ReliefF. Shao et al. [200] and
tains 76 attributes in total (1 class attribute and 75 predictive Feshki and Shijani [60] employed FS strategy prior to actual
attributes). However, most of the published works used a classification using LR and Fed Forward BP (FFBP) NN
subset of 14 attributes to conduct their experiment. In par- respectively. Shao et al. [200] used a rough set while in [60],
ticular, the Cleveland dataset is the one that has been very PSO was employed for FS purpose by the authors. Using
often used by researchers for Heart disease prediction. The PSO Al-Milli [8] showed the importance of good FS tech-
numbers 0 and 4 indicate no Heart Disease and end of life nique as PSO-FFBP produced better results only after utiliz-
respectively while the numbers 1, 2 and 3 indicate the cor- ing it (utilized multiple hidden layers) in terms of accuracy
responding attack count. The dataset consists of 303 subjects with a single classifier. In [186], the authors selected the best
and their corresponding details. In this study, we categorize feature using the ICRF-LCFS algorithm, after that CNN was
the Heart Disease detection techniques into two major cat- applied with temporal features with dense layers and activa-
egories based on whether a method employs a single classi- tion function as RELU.
fier or an ensemble of classifiers in the classification system. In another work, Jayaraman and Sultana [106] used Arti-
Table 9 lists the performance of various research efforts on ficial Gravitational Cuckoo Search Algorithm (AGCS) for
the UCI heart dataset discussed here. FS. As a classifier authors utilized the Particle Bee Opti-
mized associative Memory NN (PBAMNN), where the fea-
5.1 Single Classifier Based Methods tures were examined in the layers of the NN and this exami-
nation was done using appropriate weights and sigmoid as
In this section, we have discussed the works which utilized a the activation function. Kanchan et al. [110] and Karayılan
single classifier for Heart Disease classification. The single and Kılıç [113] employed PCA for dimensionality reduction.
The latter reduced the number of attributes from 13 to 8 and
the number of neuron in the hidden layer was decided by
18
https://​archi​ve.​ics.​uci.​edu/​ml/​datas​ets/​heart+​disea​se an iterative process. In [69], the authors utilized GA based

13
Computer Based Diagnosis of Some Chronic Diseases: A Medical Journey of the Last Two Decades 5549

Table 10  Performance Work Ref. Classifier Performance metrics

comparison of recent prediction
models on PIMA Indians ACC​ P R F1 S AUC​
Diabetes dataset. Classifiers
with ’*’ represent the classifiers Patil et al. [160] DT – – 0.79 – 0.93 0.95
with the highest ACC scores in NirmalaDevi et al. [154] k-NN 0.97 – 0.97 – 0.97 –
the respective paper Chen et al. [39] DT 0.90 – 0.87 – 0.91 –
Wu et al. [227] LR 0.95 0.95 0.95 – – 0.98
Sisodia and Sisodia [206] NB 0.76 0.76 0.76 0.76 – 0.82
Dutta et al. [54] RF – – 0.84 0.84 – –
Zhu et al. [237] LR* 0.97 0.97 0.97 – – –
Daanouni et al. [46] DNN 0.90 – – – – –
Rakshit et al. [173] ANN 0.83 – – – – –
Ashiquzzaman et al. [19] MLP 0.88 – – – – –
Saji and Balachandran [182] MLP 0.70 – – – – –
Jahangir et al. [102] MLP 0.89 0.85 0.88 – – –
Hasan et al. [82] AB+XGB – – 0.79 – 0.93 0.95
Kannadasan et al. [111] Encoder part of 0.86 0.90 0.87 0.89 0.83 –
autoencoder
Vaishali et al. [221] MOE NSGA II 0.83 – – – – –
Mansourypoor and Asadi [135] RLEFRBS* 0.84 – – – – –
Harimoorthy and Thangavelu [81] Improved SVM 0.99 1.00 0.95 – 1.00 –

approach for selecting the optimal features. After FS, out of Apart from the NN based classifiers, some researchers
13 features, only 7 features were selected and then SVM was have used other classifiers like NB [136], J48 [158], LR
used for classification. Recently, Harimoorthy and Thangav- [56] and SVM [194]. In the work by Medhekar et al. [136],
elu [81] selected the features using Chi-square based filter the NB classifier was trained using 240, 276 and 290 sam-
method. However, their major contribution was that they ples out of 303 as training samples where the set with 240
improved the radial basis kernel of the SVM model where samples outperforms all others. Sen [194] experimentally
they iteratively decreased the margin size i.e., increased the showed that SVM performs better than DT, NB and k-NN.
cost during model training. Doing so they reduced the over- Whereas in the work by Dwivedi [56], it was shown that LR
all classification error. performs better than SVM, DR, NB and k-NN.
Missing values can have a detrimental effect on the qual-
ity of data. To ameliorate this, Choudhury and Pal [43] intro- 5.2 Ensemble Based Methods
duced a novel method of missing value imputation using
autoencoder NN. The dataset was trained on multiple data- In the literature, some Heart Disease prediction systems
sets like UCI Cleveland dataset without any missing values empowered by several ensemble techniques are also reported
for the attribute. The trained autoencoder was then used to which make use of various classifiers to build an integrated
predict missing values. As an initial guess of the missing classification system. Das et al. [48] proposed an NN ensem-
value, the nearest neighbor rule was used and then refined ble classifier in their research. They first rejected the attrib-
by minimizing the reconstruction error. This was based on utes which were not relevant to Heart Disease classification.
the hypothesis that a good choice for a missing value would This was followed by dividing the dataset into two or more
be the one which can reconstruct itself from the autoen- mutually exclusive subsets first and then NN was applied to
coder. The classifiers used for the imputation were SVM, each dataset to form an ensemble of NNs that were trained
LR, NB, MLP, PRZ, k-NN, CART, and PNN. The authors with different configurations. Miao et al. [138] proposed an
comprehensively established the superiority of their impu- ensemble method based on AdaBoost and weighted majority
tation strategy especially when there is a high number of vote. The researchers found that the boosting methods were
missing values. On the other hand, Ismaeel et al. [100] used very powerful compared to other ensemble methods.
fixed hidden layer neurons (100) and Extreme Learning Recently, Maji and Arora [130] proposed an ensemble
Machine (ELM) [97] to get rid of the slow training capabil- technique that ensembles DT (C4.5) and NN classifiers
ity of BPNN algorithms as ELM is a one-time pass process named Hybrid DT. In another work, Mohan et al. [146]
thought it would be fast but it failed to acquire accuracy like introduced a method named hybrid RF with a linear model
the work proposed by Sonawane and Patil [208]. (HRFLM) to set new benchmark recognition accuracy on

13
5550 S. Malakar et al.

the UCI Heart Disease dataset. In this model, a DT based k-means clustering and k-NN classifier for the prediction of
partition was applied first, which in turn generated leaf Diabetes. In this method, prepossessing was first done by
nodes with constraints and then the leaf nodes were pruned removal of inconsistent values followed by k-means cluster-
using an entropy score. Next, using the constraint of the ing to remove the misclassified data from the training set as
leaf node, the dataset was split into several datasets (same suggested by Patil et al. [160]. However, unlike Patil et al.
as the number of leaf nodes). Each dataset was classified [160], k-NN was used as the classifier. Though the usage of
separately using the linear method and RF. In contrast to an HPM was novel, the dataset shrinks further due to the
the mentioned methods, Amin et al. [14] proposed a new removal of misclassified data.
idea for FS, where they applied a brute force approach and Significant research developments with greater focus
all combinations of features having at least 3 features were on the data preprocessing like missing value imputation
tested. In their experimentation, with 13 input attributes, a [39, 43, 227], class imbalance handling [54, 202, 203],
total of 8100 combinations were selected and tested. After outlier removal [102], and dimensionality reduction [204,
the FS, a model with popular 7 different classifiers (NB, 221] have taken place over the years. We have described
SVM, k-NN, DT, LR, NN, and Vote (average voting of NB a few important works that dealt with such processing to
and LR)) was created and the dataset was fed into that. In improve prediction performance. Chen et al. [39] improved
this experiment, with 9 features, Vote method produced the upon the HPM model of their predecessors by improvis-
maximum classification accuracy. ing a new missing value imputation method. The authors
imputed missing values using the mean of the remaining
values of the respective features and used SVM as the clas-
6 Diabetes Prediction Methods sifier. The work by Wu et al. [227] was also along similar
lines, except that they used LR as the classifier. Choudhury
In this section, we chronicle some of the research efforts and Pal [43] introduced a novel method of missing value
undertaken by the researchers to classify patients as dia- imputation using autoencoder NN. The method autoencoder
betic or non-diabetic. The works on Diabetes prediction con- model trained was trained on the dataset without any missing
sidered in the current survey used diagnosis report based values. Next, the trained autoencoder was used to predict
classification utilizing the PIMA Indians Diabetes dataset19 missing values in other records. As an initial guess for the
prepared by the National Institute of Diabetes and Diges- missing value, the nearest neighbor rule was used and then
tive and Kidney Diseases. All the subjects considered in this refined by minimizing the reconstruction error. A variety
dataset were female and above the age of 21. The dataset has of missing value imputation techniques were considered by
8 features namely, the number of pregnancies, body mass Purwar et al. [168]. The authors implemented an HPM using
index (BMI), insulin level, glucose, blood pressure (BP), MLP. Missing values were imputed using 11 different tech-
skin thickness, Diabetes pedigree function and age, and a niques like case deletion, most common method, concept
single target variable: Outcome that uses 0 and 1 for non- common method, k-NN, weighted k-NN, k-means clustering,
diabetic and diabetic cases respectively. The research efforts Fuzzy k-means clustering, SVMs imputation (SVMI), sin-
on PIMA Indians Diabetes dataset in the past decade have gular value decomposition based imputation (SVDI), local
been enumerated in the Table 10. Some of the ideas pre- least square imputation and matrix factorization, each of
sented in these works are as follows. which were classified using k-means clustering. The dataset
One of the most cited papers in the field of Diabetes with the least misclassification rate was selected. This new
prediction is by Patil et al. [160]. The authors introduced a imputed dataset without the misclassified data was classi-
novel hybrid prediction model (HPM) for Diabetes which fied using MLP to predict Diabetes. However, the dataset
has been extensively used and upgraded by other researchers is imbalanced, misclassification rate is not an appropriate
to have a better performing model. The method performed metric to judge the missing value imputation technique.
normalization and missing value deletion in the preprocess- To deal with the problem of class imbalance, Dutta et al.
ing stage followed by data reduction using a k-means cluster- [54] used oversampling of the minority class samples. Vari-
ing algorithm. k-means clustering was used to classify the ous ML algorithms like SVM, RF and LR were used for
training data into two clusters: diabetic and non-diabetic. classification and RF yielded the best result. However, over-
In this way, the samples were demarcated into two clusters, sampling leads to the synthetic generation of data which
out of which the wrongly classified instances were omitted is not entirely reliable for sensitive domains like disease
from training followed by classification using DT (J48 algo- prediction. To deal with the missing values, Harimoorthy
rithm). Later, NirmalaDevi et al. [154] used an amalgam of and Thangavelu [81] removed missing values by ignoring
the missing fields. After using the X-square filter FS, the
authors detected the presence of disease using the modified
19
https://​www.​kaggle.​com/​uciml/​PIMA-​india​ns-​diabe​tes-​datab​ase redial basis kernel of SVM. It is to be noted that recently,

13
Computer Based Diagnosis of Some Chronic Diseases: A Medical Journey of the Last Two Decades 5551

Shaw et al. [202, 203] proposed two new methods (ensem- the least clustering error was used. SVM was used for clas-
ble method [203] and RTPSO for majority sample selec- sification. It is noteworthy that F-score does not provide any
tion [202]) for better handling imbalance class problem in mutual information among the features. Hence, it may lead
a dataset. to the selection of multiple irrelevant features which convey
Another ’impurity’ in real world data is the existence of the same information. In another work, Vaishali et al. [221]
outliers [34]. Significant research efforts have been devoted used GA for FS which reduced the feature dimension to half
to eliminating this problem in the domain of diagnostic its actual size. After the FS, two types of multi-objective
report based disease prediction. For instance, Jahangir et al. evolutionary (MOE) fuzzy rule based classification algo-
[102] designed an outlier detection technique to preproc- rithms namely, Non-dominated Sorting GA (NSGA II) &
ess the data. The outliers in the dataset were detected using elitist Pareto-based multi-objective evolutionary algorithm
the enhanced class outlier factor (ECOF) based method (ENORA) were employed. MOE NSGA II fuzzy algorithm
that improved the supervised class outlier factor (COF) yielded a better result. In the same line, Shen et al. [204]
based method. COF for a particular instance is measured applied FOA-SVM technique. This technique can select
using three different scores: the probability of being part optimum features with an iterative method till the number
of the class under consideration (i.e., nk , where n represents of iterations matches the maximum iteration present. The
the number of samples from its k neighbours belong to the FOA-SVM technique produced very good results compared
same class), the deviation of from the respective instances to methods like PSO-SVM and GA-SVM.
of the same class and the average distance from its neigh- Fuzzy systems were also employed by Mansourypoor and
bors. However, the tuning parameters COF was replaced by Asadi [135] where the authors designed a Reinforcement
a normalization factor in ECOF based method. After this, Learning-based Evolutionary Fuzzy Rule-Based System
the authors used an auto tunable NN that adjusts the learn- (RLEFRBS). The proposed model involved the building of
ing rate and the number of hidden units for classification. a Rule Base (RB) followed by rule optimization. The ini-
An inappropriate outlier deletion method may lead to the tial RB was constructed from numerical data out of which
removal of a large amount of training data which may cause redundant rules were eliminated based on the confidence
underfitting/overfitting of NN based classifiers. Hence, a bet- measure. This was followed by rules used for pruning.
ter outlier detection method like [33] could be used. Finally, GA was employed to select the appropriate subset
Zhu et al. [237] and Wu et al. [227] converted the num- of rules. Usage of fuzzy systems ensures high interpretable
ber of pregnancies into a categorical feature to distinguish models which makes such models acceptable in the research
between zero and non-zero pregnancies as a part of data community. Both the problems of data preprocessing and
preprocessing but followed different classification strategies. FS are dealt with elegantly by Hasan et al. [82]. The authors
Zhu et al. [237] first applied PCA on preprocessed data to studied the performances of two ensemble learners (viz.,
reduce the number of features and then they used k-means XGBoost, and AdaBoost) and MLP in three different sce-
clustering to delete the misclassified instances from the narios. In the first case, preprocessing techniques like Outlier
training set. However, Wu et al. [227] used k-means cluster- Elimination (OE), Missing Value Substitution (MVS), and
ing based training data deletion on the original data samples. standardization were applied with dimensionality reduction
Wu et al. [227] used the basic version of LR while Zhu et al. techniques like PCA, ICA and FS using correlation. In the
[237] used improved LR as the underlying classifier. Here, second method, MLP was designed after applying OE and
it is to be noted that by clubbing all women with a history MVS. In the third method, OE + MVS with AdaBoost +
of pregnancy into a single bin, the authors disregarded the XGBoost was found to be the best ensemble technique. It is
number of pregnancies. However, according to medical observed that the first method had the best specificity while
research [127], women with at least four pregnancies were the other method yielded the best values in sensitivity and
28% more likely to develop Diabetes compared to women AUC score.
who reported two or three pregnancies. Hence, converting Apart from these, we find extensive usage of MLP in
the number of pregnancies into a binary variable may lead Diabetes prediction. Saji and Balachandran [182] and
to medical inconsistencies. Ashiquzzaman et al. [19] made significant contributions to
Apart from data preprocessing, another familiar problem this domain. Saji and Balachandran [182] compared the out-
that researchers face is the problem of FS. Quite often, the comes of MLPs with various learning algorithms like Leven-
datasets are encumbered with irrelevant information which berg-Marquardt, BFGS Quasi-Newton, Resilient BP, scaled
needlessly increases the complexity of the model. FS has conjugate gradient, conjugate gradient with Powell/Beale
been widely explored for a long time starting with Ilango restarts, Fletcher-Powell conjugate gradient, Polak-Ribiere
and Ramaraj [99] who used an HPM approach. In this work, conjugate gradient, one step secant and variable learning
a hybrid FS technique that first ranked the features based on rate backpropagation. Levenberg-Marquardt yielded the best
F-score followed by optimum feature subset selection having results among all. Ashiquzzaman et al. [19] experimented

13
5552 S. Malakar et al.

Table 11  Performance Work Ref. Classifier Performance metrics

comparison of recent prediction
models on UCI CKD dataset. ACC​ P R F1 S AUC​
Classifiers with ’*’ represent the
classifiers with the highest ACC Pujianto et al. [167] SVM 1.00 – – – – –
scores in the respective paper Amirgaliyev et al. [15] SVM 0.94 – 0.94 – 0.95 –
Chetty et al. [41] k-NN* 1.00 – – – – –
Charleonnan et al. [35] SVM 0.98 – 0.99 – 0.98 –
Wibawa et al. [226] AB 0.98 0.98 0.98 0.98 – –
Sedighi et al. [190] NB 0.97 – – – – –
Sisodia and Verma [207] RF 1.00 1.00 1.00 1.00 – 1.00
Avci et al. [21] DT 0.99 0.98 0.99 0.98 – –
Almansour et al. [12] ANN 0.99 – – – – –
Wibawa et al. [225] SVM – – 0.99 – 1.00 –
Aljaaf et al. [11] MLP 0.98 – – 0.98 – 0.99
Gunarathne et al. [76] RF 0.99 – – – – –
Guia et al. [74] k-NN – – – 0.99 – –
Johari et al. [108] ANN 1.00 1.00 1.00 – – –
Polat et al. [164] SVM 0.98 0.98 0.98 – – 0.98
Harimoorthy and Thangavelu [81] Improved SVM 0.98 0.95 1.00 – 0.99 –

with MLPs with different dropout schemes. The model used 7 CKD Prediction Methods
2 hidden layers, one with 64 neurons and the other with 32
neurons. The 2 hidden layers had 25% dropout and the last The methods considered in this survey use diagnostic data
layer had 50% dropout. for CKD prediction based on the UCI CKD dataset20. The
Recently, Kannadasan et al. [111] used a stacked autoen- dataset has 25 diagnostic attributes and a single target vari-
coder based DNN model to predict Diabetes. A typical able. In all, there are 11 nominal and 13 categorical fea-
autoencoder model consists of two components namely tures. The research endeavors of the past about the UCI CKD
encoder and decoder. In this work, the authors used only the dataset have been compared in Table 11. Some of the ideas
encoder part. In this model, the input of the autoencoder was presented in the existing works have been discussed here.
the input of the encoder and the output of the encoder was An important aspect of any classification system is the
the output of the hidden layer of the autoencoder. The DNN quality of data which can be enhanced by cleaning and pre-
model was trained to minimize 3 errors: Mean Square Error processing techniques, some of which include missing value
(MSE), regularization and sparsity using backpropagation. imputation, outlier removal, and class imbalance rectification.
Some researchers only tried to evaluate the performance These areas have been explored by a number of researchers in
of the dataset on a single classifier. For example, Rakshit the case of designing diagnostic report based disease screen-
et al. [173] implemented a 2-class DNN after standardiza- ing methods, which is true for CKD detection also. For exam-
tion and missing value deletion, Daanouni et al. [46] com- ple, Pujianto et al. [167] cleaned the data by creating ’pure
pared four classifiers, namely DT, k-NN, ANN and DNN. clusters’ using the k-means clustering algorithm. The clus-
In [46], DNN with FS using Neighbourhood Components tering process generated two ’pure’ clusters containing only
Analysis (NCA) yielded the best result. On the contrary, a those instances whose labels corresponded with the cluster
comparative analysis of different algorithms was performed label. Data not belonging to any of the pure clusters were
by authors like Hashi et al. [84], and Sisodia and Sisodia removed prior to learning the classifier. The ’pure’ cluster-
[206]. Three shallow classifiers namely DT, SVM and NB merged dataset was validated using SVM with polynomial,
were used by Sisodia and Sisodia [206], out of which NB RBF and Sigmoid kernels. This method is similar to the
yielded the best result. one used by Patil et al. [160] in Diabetes detection and thus
had the same problem of dataset shrinkage as the former. In
another work, Harimoorthy and Thangavelu [81] modified the
redial basis kernel (discussed earlier) of SVM which in turn
decreased the classification error of the SVM model.
The problem of class imbalance was scrutinized by
Wibawa et al. [225]. They also introduced a novel method
20
http://​archi​ve.​ics.​uci.​edu/​ml//​datas​ets/​Chron​ic_​Kidney_​Disea​se

13
Computer Based Diagnosis of Some Chronic Diseases: A Medical Journey of the Last Two Decades 5553

of ELM using RBF, Linear, Polynomial and Wavelet ker- Hence, the selected feature subset may not be the optimal
nels. Various combinations of imbalance correction, FS and feature subset always. A popular method using the corre-
kernels were applied. ELM with RBF kernel with under- lation between features and target variables has been used
sampling correction FS yielded the best results. Though by Almansour et al. [12] and Wibawa et al. [226]. Alman-
undersampling may reduce the imbalance in the dataset, it sour et al. [12] deleted the missing values. After the pre-
shrinks an already miniature dataset and sometimes makes processing, the best hyperparameters for ANN and SVM
the model prone to overfitting. Hence, using a hybrid of were selected. Using these optimized hyperparameters, the
oversampling and undersampling algorithms may lead to authors computed the correlations among all the features
better classification. To eradicate the problem of missing and the target which were then arranged in descending order.
values, a three-fold approach was designed by Sisodia and Half the features having the best correlation with the target
Verma [207]. If the number of missing values was small variable were retained and tested with ANN and SVM. The
for an attribute, then the respective instances were deleted. process of halving and testing was continued until a single
However, if the number of missing values was large in com- feature was left. The feature subset which exhibited the best
parison to the total number of instances, the attribute itself performance was used to build the final model. It is note-
was deleted. If the number of missing values was moderate, worthy that the process of halving the number of features
then they were imputed. Finally, the dataset was classified may lead to the selection of a non-optimum feature subset.
using NB, SVM, J48, and three ensemble classifiers, namely, Sandhiya and Palani [186] selected the best features using
RF, Bagging, and AdaBoost. the ICRF-LCFS algorithm where those features are selected
Aljaaf et al. [11] filtered the data to weed out outliers and which contributed most in predicting good results. Before
also dealt with missing values. The outliers were replaced that, the authors employed the k-Means algorithm to group
with the highest or lowest permissible values. For missing them. After that T-CNN is applied to this. Wibawa et al.
values, a two-fold approach was adopted. If the missing val- [226] combined boosted classifier and FS for predicting
ues comprised less than 20% of the total data, the data was CKD. The features were first selected using Correlation FS
deleted. Otherwise, the missing value was imputed. This was (CFS). For the boosted classifier, it used the base learners
followed by FS which produced an optimal feature subset like NB, k-NN and SVM. Though collinearity is a very easy
of 7 features. These were trained using Classification and and readable way to judge the relevance of the feature, it may
Regression Trees (CARTs), LR, MLP and SVM classifiers. lead to multicollinearity as only the correlation of feature
Another significant research direction is the predominance variables with the target variable was taken into account.
of FS algorithms owing primarily to a large number of fea- This could result in the inclusion of redundant features.
tures which leads to what can be called a twin curse: an Owing to superior prediction capabilities, boosting algo-
increase in training time and the possibility of overfitting. A rithms have become extremely popular of late. Works by
number of authors have tried different FS strategies like IG, Wibawa et al. [226], Sedighi et al. [190] and Islam and
Best First Search (BFS), etc. Polat et al. [164] demonstrated Ripon [18] which combined FS and boosted classifier bear
the utility of FS algorithms by comparing the performance testament to this fact. Sedighi et al. [190] undertook a com-
of SVM with and without FS. By fixating on a single clas- prehensive comparative study of different filter and wrap-
sification algorithm: SVM, the authors comprehensively per based FS methods. The missing values were imputed
illustrated that FS invariably improved the classification using the k-NN model. Forward FS (FFS), Backward Feature
accuracy. The FS algorithms used were classifier subset Elimination (BFE), Bi-directional Search (BDS) and GA
evaluator, wrapper subset evaluator, correlation FS subset were used for FS and trained with AdaBoost using NB as the
evaluator, and filtered subset evaluator which employs either base learner. GA was found to perform the best among all
best first search or greedy stepwise methodology. The fil- the FS algorithms. However, according to some studies like
tered subset evaluator with the BFS method outperformed Ting and Zheng [219], using a boosting algorithm on NB
all other strategies. may not produce the desired improvement in classification
Chetty et al. [41] used BFS for FS followed by classifi- compared to NB. Introducing a tree structure in NB followed
cation. The reduced feature subset was used for CKD pre- by boosting can improve the classification significantly.
diction using NB, SVM and k-NN classifiers, out of which Apart from FS, feature extraction methods were also used
k-NN produced the best result. Like Chetty et al. [41], to reduce dimensionality. Though not extensively explored
Charleonnan et al. [35] used BFS algorithm for FS. The in literature, authors like Guia et al. [74] have used PCA. In
authors converted the nominal attributes to binary attributes this work, after missing value imputation, label encoding
in the training data followed by FS. k-NN, SVM, LR and and normalization, PCA was used for dimensionality reduc-
DT were used as the classifiers out of which SVM yielded tion. The first 11 principal components accounting for more
the best results. One of the disadvantages of the BFS is that than 95% of the variance, were used. SVM, DT, RF, Gauss-
it, being a greedy algorithm, may get stuck in local optima. ian NB, MLP, and k-NN were used as the classifiers out of

13
5554 S. Malakar et al.

which k-NN yielded the best accuracy. It is noteworthy that applied k-fold cross-validation for partitioning the training
PCA works best with strongly correlated features. Hence, data. Gogi and Vijayalakshmi [68] used various classifiers
using CFS followed by PCA could enhance the predictive like SVM, LR and DT for the classifying samples belonging
power of the model. Some authors like Avci et al. [21] and to ILPD. However, they applied missing value imputation
Amirgaliyev et al. [15] emphasized the comparative study with not a number (NAN) in the preprocessing stage. In their
of various algorithms. Avci et al. [21] provided a simple experiment, LR performed better than the rest and yielded
comparison of various ML algorithms like NB, K-Star, SVM the best classification metrics. In another work, Kumar and
and J48. It was found that the J48 algorithm had the best per- Katyal [120] reported that the C5.0 with adaptive boosting
formance. An SVM-based approach was used by Amirgali- outperformed the classical classifiers like NB, RF, k-NN,
yev et al. [15] where they used Linear SVM and the results actual C5.0 (a DT based classifier) and the k-means clus-
were compared for 10-fold cross-validation. In another tering algorithm in terms of accuracy, precision and recall
work, Johari et al. [108] made a comparative study of DT values while detecting liver disease.
and 2-class NN after feature normalization using min-max Contrary to the above-mentioned techniques, some
normalization. The NN yielded much better results across researchers have tweaked the native classifiers to obtain
various performance metrics. A similar study was done by better classification performance. For example, Tiwari et al.
Vashisth et al. [223] among MLP, SVM and NB where MLP [220] employed various Lazy classifiers like instance based
outperformed the other algorithms. k-NN using Log and Gaussian (IBKLG) weight kernels,
Locally Adaptive k-NN (Localk-NN) algorithms and Rough
Set Exploration System Library based k-NN (Rseslibk-NN)
8 Liver Disease Prediction Methods for stratifying ILPD. Comparatively, Localk-NN showed
the finest accuracy and recall measure while IBKLG over-
The majority of the research works found in the literature powered the rest of the lazy classifiers in terms of precision
for Liver Disease prognosis have used diagnosis report score. However, it could not handle circumstances where
based classification utilizing the Indian Liver Patient Data- the training data is insufficient or corrupted with noise. In
set (ILPD)21 and Liver Disorders Dataset (LID)22. Both the another work, Chua and Tan [45] upgraded the classification
datasets are available in the UCI repository. ILPD is a mul- capability K-NN classifier by introducing a fuzzy rule-based
tivariate dataset consisting of 583 patient records (441 male k-NN algorithm for Liver Disease classification utilizing
and 142 female). All the patients were from the northeast the LID dataset. Unlike conventional k-NN, the proposed
part of Andhra Pradesh, India. It has 10 attributes and 416 method’s initialization procedure operated the imprecise
out of 583 patients are classified as liver patients and the rest inputs (neighborhood density and distance) through the
are non-liver patients in the dataset. LID (originally known natural framework of the fuzzy logic system. Euclidean dis-
as the British United Provident Association (BUPA) dataset) tance measurement was adopted to attenuate dimensionality.
is a multivariate dataset consisting of 345 samples having Fuzzy rule-base’s performance was enhanced by a binary
7 attributes. Each sample constitutes the record of a male coded GA that concomitantly optimizes the parameters of
subject. The subjects were labelled as either class label 1 the rule consequent, antecedent membership functions and
(positive case) or 2 (negative case). Out of 345, there are 145 the feature weights. The decision boundary produced by
and 200 positive and negative subjects respectively. the proposed method was tunable and thus it could handle
There is a plethora of works found in the literature that circumstances where the training data is insufficient or cor-
have selected a classifier from a pool of available classi- rupted with noise. After evaluation, it was evident that fuzzy
fiers based on their merit for Liver disease screening while rule-based k-NN outperformed crisp and fuzzy k-NN.
experimenting on IPLD and/or LTD. For example, Kant and All the prior mentioned techniques followed a common
Ansari [112] utilized the Atkinson index, a popular measure trend of utilizing robust classifiers on the raw dataset with-
of inequality to select the initial centroid of the k-means out concluding the distinguishing capability of the attrib-
algorithm. The improved k-means clustering with the Atkin- utes present in the dataset. However, some researchers have
son index produced a thriving precision score on the ILPD focused on the selection and ranking of the features and
dataset. In another work, Adil et al. [4] used an LR classifier reported that the performance of most practical classifiers
that is simple and possesses low computational cost. They improves a lot when highly-correlated or irrelevant features
split the ILPD into training and testing data in a 7:3 ratio and are omitted from a given dataset. For example, Kulkarni and
Shinde [118] came up with a hybrid neuro-fuzzy classifica-
tion model to predict Liver Disease from the LID dataset
21
https://​archi​ve.​ics.​uci.​edu/​ml/​datas​ets/​ILPD+​(Indian+​Liver+​Patie​ where IG was used to select the feature subset that gives
nt+​Datas​et) the highest classification accuracy. In this work, ANN was
22
https://​archi​ve.​ics.​uci.​edu/​ml/​datas​ets/​liver+​disor​ders used to comprehend the membership value for fuzzy classes

13
Computer Based Diagnosis of Some Chronic Diseases: A Medical Journey of the Last Two Decades 5555

Table 12  Performance Work Ref. Classifier Performance metrics

comparison of recent prediction
models on UCI ILPD and ACC​ P R F1 S AUC​
UCI Liver Disorders datasets.
Classifiers with ’*’ represent the Methods on UCI Indian Liver Patient dataset
classifiers with the highest ACC Adil et al. [4] LR 0.74 0.72 0.74 0.69 – –
scores in the respective paper Gogi and Vijayalakshmi [68] LR* 0.95 – – – – 0.93
Abdalrada et al. [1] LR 0.72 – 0.90 – 0.78 0.75
Srivenkatesh [210] LR* 0.76 0.77 0.95 0.85 – –
Kant and Ansari [112] k-means 0.23 0.94 0.31 0.46 – –
Ramana and Boddu [174] Bagging 0.69 0.44 0.95 – 0.85 0.70
Nahar and Ara [83] SVM 0.73 – 0.96 – 0.96 –
Babu et al. [22] NB* 0.56 0.76 – – 0.95 –
Lakshmi et al. [124] C4.5 – – 0.78 – – –
Kumar and Katyal [120] C5.0 0.75 0.90 0.78 – – –
Tiwari et al. [220] k-NN 0.98 – – – – 0.98
Kumar and Sahu [119] RF* 0.79 0.69 0.61 0.65 0.87 –
Methods on UCI Liver Disorders dataset
Chua and Tan [45] k-NN 0.67 – – – – –
Kulkarni and Shinde [118] ANN 0.67 – – – – –
Haque et al. [80] ANN 0.85 – 0.85 0.82 0.89 –

of an input dataset. Next, they employed the sum aggre- function based feature ranking method. Using the sigmoid
gation reasoning rule to aggregate attribute belongingness function occurrence probability (say, p) of each attribute
so that the pattern belongingness to the given classes can was calculated. Attributes having a low p value (< 0.05)
be calculated. Obtained attributes were thereafter used for and odds ratio were considered to be less important and
high level decision making first and then utilizing defuzzi- eventually they were removed. Only 5-attributes viz., Age,
fication operation, the pattern was assigned to the predicted DB, SPGT, TP and Albumin were chosen for Liver Disease
class. Underlying attributes were ranked using some ranker prediction using the LR classifier.
method and then a few top ranked attributes which were Another problem that plagues researchers is the pres-
relevant to making predictions were utilized to train the clas- ence of missing values. Choudhury and Pal [43] introduced
sification model. a novel method of missing value imputation using autoen-
Babu et al. [22] experimented with different attribute coder NN. The dataset was trained on multiple datasets like
evaluator techniques available in the WEKA tool and applied the ILPD without any missing values for the attribute. The
them to ILPD while Nahar and Ara [83] used Singular Value trained autoencoder is then used to predict missing values.
Decomposition (SVD) based ranking algorithm available in As an initial guess of the missing value, the nearest neighbor
Matrix Laboratory (MATLAB) for predicting Liver Disease. rule is used and then refined by minimizing the reconstruc-
Going into detail, Babu et al. [22] employed a k-means clus- tion error. This is based on the hypothesis that a good choice
tering algorithm on the top-7 ranked features of the train- for a missing value would be the one that can reconstruct
ing dataset and checked the clustered validity of objects. itself from the autoencoder. The classifiers used for the
Mis-clustered data were omitted from the training dataset. imputation are SVM, LR, NB, MLP, PRZ, k-NN, CART
Finally, the refined train set was fed into various classifiers and PNN. The researchers comprehensively establish the
such as NB, K-NN and DT (C4.5 algorithm) for classifica- superiority of their imputation strategy especially for high
tion. NB classifier outdid k-NN and DT. However, Nahar rates of missing values.
and Ara [83] experimented with different top ranked fea- In the literature, some authors like Kumar and Sahu [119],
tures using SVM as a classifier. The authors experimentally Auxilia [20], Ramana and Boddu [174] and Srivenkatesh
showed that top-8 ranked features performed best for both [210] emphasized the comparative performance of various
the ILPD and the LID datasets. Pearson Correlation Coef- classifiers on the ILPD. Kumar and Sahu [119] performed
ficient (PCC) of the attributes was used by Haque et al. [80]. a set of experiments to check the performance of classifiers
In their work, they used RF and ANN as a classifier in order like RF, CART and Reduced Error Pruning Tree (REPTree)
to discern liver disorder patients from rest. Experimentally, with two different train test splits: 70%-30% and 80%-20%,
NN outperformed RF. Abdalrada et al. [1] used a sigmoid and inclusion/exclusion of FS technique. They obtained the

13
5556 S. Malakar et al.

best result using the FS method with RF on 80%-20% on the image based disease detection. This is probably due to the
train-test split. In [20], the author examined the performance concise nature of the datasets and the bottleneck of comput-
of classifiers like DT, ANN and RF to foresee a liver ail- ing capabilities. With the abundance of data at the disposal
ment in patients. DT outperformed the rest of the classifiers of researchers and the availability of superior computers,
based on some factual metrics. Srivenkatesh [210] inves- it is expected that DNNs will be explored exhaustively in
tigated performances of five different supervised learning the near future. Besides, fully trained DNNs like ResNet
algorithms such as NB, LR, SVM, RF and k-NN. LR outper- and Inception could be employed in the future. Also, some
formed the rest of the classifiers. Performance measures like hybrid models (e.g., [65, 224]) that feed some handcrafted
Kappa statistic value, Mean Absolute Error (MAE), Root features to the DL model could be another future direction.
Mean Square Error (RMSE), P and F1-score were utilized In addition to these, ensembles of different DL models bear-
to assess the performances of the classifiers. Also, a com- ing some complementary information (e.g., [50, 51, 162])
parative study by Ramana and Boddu [174] includes the could be applied to enhance the overall classification model.
classifiers like Bagging, k-NN, DT (J48 algorithm), JRip (a
RIPPER algorithm), MLP and NB classifiers. The bagging 9.2 Designing Sophisticated Data Processing
classifier achieved the highest accuracy followed by MLP, Strategies
J48, JRip, k-NN and NB. These comparative studies allow
us in perceiving the advantages and disadvantages of vari- From the above discussion, it is evident that data preprocess-
ous classifiers and provide insight into various techniques. ing had played a vital role over the years to enhance the end
Table 12 enumerates the performance of various research performance of the detection models. Such preprocessing
efforts over the last 10 years about Liver Disease prediction techniques included missing value imputation and feature
using ILPD and LID datasets. dimension reduction through FS. In this subsection, we first
revisit and analyse the existing works and then suggest some
ways which might improve the existing ones.
9 Future Research Directions
9.2.1 Handling Missing Values
From our thorough analysis of different computerized dis-
ease detection methods over the different datasets related The existence of missing values is a very common problem
to Breast Cancer, Lung Cancer, Leukemia, Heart Disease, in the case of diagnosis report based datasets as this is simi-
kidney disorder (i.e, CKD) and Liver Disease, it is clear lar to survey data. All the related information might not be
that AI and ML based approaches were extensively used available for all subjects as medical practitioners may collect
by the researchers since last two decades. New methods are only information they consider relevant. However, in ML
constantly being explored to develop better models for the based systems, such missing values are detrimental during
prediction of diseases now and then. It comes as no surprise prediction as learning may be biased to the available data.
that the models designed in the last few years have compre- To handle such scenarios, many researchers like Patil et al.
hensively outperformed previous research attempts across [160] for PIMA Indians, Almansour et al. [12] for the UCI
all the datasets considered in our survey. While this survey CKD dataset, and Choudhury and Pal [43] for ILPD opted
does chronicle significant inroads of AI and ML in disease for deleting the missing values prior to actual classification.
prediction and diagnosis, there are some significant research In some cases, the missing values are substituted with zero
gaps that future researchers should take up. Some of them [12] or the mean value of the feature for the rest of the data
have been enlisted and described hereafter in this section. like Chen et al. [39] in PIMA Indians among others to han-
dle such missing values. However, such crude techniques of
9.1 Exploring more DL Models missing value imputation or deletion may not be the right
choice for some disease predictions. To perform missing
Researchers primarily used ML (Pujianto et al. [167] in UCI value imputation, one might employ a clustering approach
CKD, Ilango and Ramaraj [99] in PIMA Indians, etc.) or where data samples can be clustered without considering the
Shallow NNs (Aljaaf et al. [11] in UCI CKD, Gavhane et al. attribute having missing values first and then the informa-
[64] in UCI Heart dataset, etc.) for image based classifica- tion from the same cluster data may be used to substitute the
tion systems. It is noteworthy that most of these approaches missing values. Additionally, some state-of-the-art methods
employed ML or shallow NNs as classifiers. DNNs con- as described in [126, 171, 191, 233] could be applied.
sisting of multiple hidden layers have not been explored
much as compared to shallow learners. DNNs were mainly
explored using transfer learning (work by Shafique et al.
[196] in ALL_IDB dataset) for designing pathological

13
Computer Based Diagnosis of Some Chronic Diseases: A Medical Journey of the Last Two Decades 5557

9.2.2 Exploring more Dimensionality Reduction Strategies 9.4 Handling Class Imbalance Problem

Many datasets have a large number of features but not all Class imbalance is very common when considering medi-
features are useful to generate the final output. Such redun- cal data since the number of infected subjects is very low
dant data only increases the computational complexity. Fea- as compared to the number of healthy subjects. However, a
ture engineering is a process to generate a reduced dataset very few researchers, as described during methodological
with new features which are produced from the existing description, have taken into account the problem of imbal-
raw dataset. However, we find that for data reduction, most anced class in the dataset (e.g., Han et al. [79] in BreakHis
researchers have focused on FS. Dimensionality reduction Dataset and Dutta et al. [54] in PIMA Indians) while using
using feature engineering has not been extensively explored ML based approach. Imbalanced datasets can often lead to
and is mostly restricted to only PCA like Ophir et al. [62] poor classification and misleading accuracy measures. Class
in Fred Hutchinson Cancer Research Center Dataset and imbalance could be rectified either using undersampling or
Karayılan and Kılıç [113] in UCI Heart Disease or LDA as oversampling or a hybrid of the two approaches in case of
in Mishra et al. [144]. Other dimensionality techniques like the diagnostic report based as well as image based classifica-
isomap embedding [184], locally linear embedding [157], tion problems. Besides, data augmentation for image based
supervised locally linear embedding algorithm [232] among classification problems could be a better choice.
others could be useful.
9.5 Generating Large‑scale Datasets
9.2.3 Designing new FS Strategies
It is observed from the methodological reviews of the exist-
There is also a dearth of variety in optimization algorithms ing techniques that there is a crisis of large-scale datasets
with most researchers using gradient descent. Very few of the various diseases. We find that the datasets used here
researchers used GA in WBC dataset (Huang et al. [98]), are restricted to a certain age bracket, gender, or location.
ILPD dataset [45] and PIMA Indians datasets [221], AMO Such homogeneity in the dataset may lead to the problem
Algorithm in UCI CKD Dataset [18] or PSO in UCI Heart of overfitting and the proposed model may not generalize
Disease [60], WBC dataset [234] etc. which yielded satisfac- well for real world scenarios. Hence, care must be taken to
tory results in respective research domains. It is imperative build datasets that include a heterogeneous sample space.
that researchers in future might extensively use one or more This is also a pressing need in the world of the Internet of
such optimization algorithms. To be specific, for small sized Things (IoT) where medical practitioners all over the world
data, a filter method or ensemble of filter methods could be can work as a team to deal with complicated diseases. Some
employed while for the large sized data, use of wrapper filter special categories of data could also be prepared for aged
methods (i.e., filter method followed by wrapper methods) or people with different co-morbidity or as in the current sce-
hybrid wrapper methods or wrapper method with some local nario COVID-19 patients having no prominent symptoms.
search could be another future scope for the researchers. Here, we describe some of the possible data requirements
that could be prepared in the future.
9.3 Fusing Handcrafted Features with Deep
Features 9.5.1 New Data Collection

For image based classification systems which typically need Breast Cancer: WBC dataset is the most widely used data-
feature extraction followed by the classification, researchers set in the literature on Breast Cancer detection. However,
mostly focused on shape, colour or texture based features. this dataset was last updated in 1991 which makes it out-
Some state-of-the-art feature extraction techniques like Scale dated. The number of instances is quite low with only 699
Invariant Feature Transform, Speeded Up Robust Features, instances. Moreover, some important attributes causing
Oriented FAST and Rotated BRIEF (ORB), Daisy and Breast Cancer like age, obesity, menopause, and family his-
Haralick (except [195, 222]) feature had not been explored tory of having Breast Cancer are also absent there. For image
much till now for such problems. Moreover, researchers had based classification, BACH and BreakHis are some of the
also overlooked the possibility of some intelligent hybrid prominent Breast Cancer histopathological image datasets
models combining handcrafted features with deep features widely used by the researchers. Various data augmentation
like [195]. Therefore, exploring the use of state-of-the-art techniques are used to cope with model overfitting due to
handcrafted features or hybrid models fusing handcrafted data scarcity issues in these datasets. However, there are
features with deep features might help in improving the pre- several problems associated with data augmentation like
sent state-of-the-art results obtained by image based CAD loss of discriminating features or information. The selec-
models. tion of a data augmentation approach should be performed

13
5558 S. Malakar et al.

judiciously based on the dataset characteristics. Having more model. A new dataset with subjects across locations, gen-
data is always the recommended panacea, as data augmenta- ders and age brackets is the need of the hour. Moreover, the
tion also brings up the possibility of the elimination of some dataset is too small with only 768 instances to generalize any
inherent properties from the image. method experimented on this dataset. Even within the exist-
Lung Cancer: The Irvine dataset (last update: 1992) ing dataset, the data is skewed with only 268 instances being
consists of only 32 instances and the MLCD dataset (last positive. In addition to these, the dataset was last updated in
update: August 2002) contains a total of 96 instances where 1990 which makes it obsolete in the current scenario.
86 primary lung adenocarcinomas samples and 10 non-neo- CKD: The dataset used for CKD prediction is the UCI
plastic lung samples. Such a low sample count and age-old CKD dataset. The dataset recorded patients from Tamil
dataset motivated researchers to provide the LCD dataset Nadu in India over a period of 2 months. Hence, like in
(last update: 2017) having 1000 instances. This data is much Diabetes, the dataset lacks diversity in terms of geography.
better as compared to the other two while the number of A dataset more inclusive of heterogeneity would lead to gen-
samples is taken into account. However, this dataset is not erating better models for the prediction of CKD. The dataset
enough in a practical scenario. Hence, more such dataset is has only 400 instances out of which 62% of the subjects had
required to be generated to improve the Lung Cancer detec- CKD. This points to the imbalance in the dataset. These
tion from diagnosis report and thereby assist the medical issues create an urge towards the generation of a new and
experts to make proper decisions. larger dataset covering the long period and large geographi-
Leukemia: Golub dataset is the most common and one cal locations.
of the early gene expression based datasets and this was Liver Disease: The LID (generated in 1990) and ILPD
a frequently used dataset in Leukemia related research. It (generated in 2012) were widely used diagnostic data-
originated in 1999, which makes it quite old. Moreover, sets for Liver Disease detection using classical ML based
this dataset has only ’ALL’ and ’AML’ data while data from approaches. Several challenges are associated with these
non-cancerous subjects are missing. Hence, preparing a datasets for clinical research. Both the datasets are quite
new dataset containing gene expression of ALL, AML and outdated and demographically localized. The ILPD has
non-cancerous subjects or adding gene expression data of both male and female subjects records while LID is limited
non-cancerous subjects to this dataset could open up a new to male subjects only. ILPD comprises 583 patient records
direction on Leukemia detection from gene expression data. and LID consists of scarcely 345 instances. DNNs are data-
There are only 72 instances which is a quite low number. eating approaches that require bountiful amounts of training
The histopathological image dataset used for the classifica- data. Hence, they failed to yield good results in ILPD and
tion of Leukemia is the ALL_IDB dataset which consists LID when compared to classical ML models.
of histopathological images collected as long back as 2005.
Hence the dataset has become outdated. With better micro- 9.5.2 Generating Synthetic Data
scopes and software available to capture the image under the
slides, a newer dataset would lead to better models. We have already mentioned that DL based models, in gen-
Heart: The Cleveland dataset is used in almost every eral, require a large amount of data for proper training of the
heart related research work which was developed in 1988. model. We have seen that the availability of sufficient and
This dataset has 303 instances and it is quite low from the useful data becomes a major concern for the researchers.
ML point of view. This dataset contains some important In the case of medical data, things are even more cumber-
attributes like smoking, Diabetes but ignores some poten- some as collection or preparation of medical data is most of
tially useful attributes. Moreover, this dataset was only col- the time very costly and often time-consuming. Besides the
lected from Cleveland city, USA (Ohio) and thus making assemblage, the data also needs to be annotated by medical
the dataset highly localized. It would be wiser to create a experts. In 2014, Goodfellow et al. [71] introduced Gen-
dataset that includes subjects from countries like Turkmeni- erative Adversarial Networks (GANs). GAN is a generative
stan, Kazakhstan, Kyrgyzstan, Mongolia, and Russia which model to generate new realistic data samples from the refer-
are highly affected by it. So we need more globalized and ence sample data that are not only similar to the examples
updated data for more applicability of the devised method but are indistinguishable as well. Over recent years, GAN
on that new dataset to real life. has been gaining a lot of attention in the medical fraternity
Diabetes: The dataset used in Diabetes includes only due to its impeccable features, namely, image synthesizing,
females above 21 years with the maximum age being 65. de-noising, segmentation, reconstruction, data simulation
All the subjects belonged to PIMA Indian heritage. Hence and classification. GAN and its extensions can extricate the
the sample space considered here is confined to narrow scarcity of labelled data problems by generating close to
geography and of a particular gender. Hence such a dataset realistic data from existing samples. However, synthetic data
is outdated and does not lead to the generation of a robust generated if not nearly identical to real-world data, can affect

13
Computer Based Diagnosis of Some Chronic Diseases: A Medical Journey of the Last Two Decades 5559

the quality of decision making. The only way to guarantee a years, researchers obtained considerable success in medi-
GAN generator is generating accurate, realistic outputs is to cal image segmentation using DL based approaches [86,
test its performance on well-understood, human annotated 235]. Thus, in the past few years, DL aided medical image
validation data using the discriminator. Although synthetic segmentation techniques have gained vast popularity in the
data generation has become easier over time, real-world research community. However, researchers of this domain
human annotated data remains a cornerstone of training encountered limited data both in terms of count and variety.
data for CADDs. To be specific, these data are scarcely annotated or weakly
annotated [217] and this makes the medical image segmen-
9.5.3 Preparing Multi‑modal Data tation problem a GT-hard problem [193]. To overcome this
issue, one may employ approaches like semi-supervised
For diseases like Breast Cancer, Leukemia and Lung Cancer learning [229], contrast learning [107], domain adapta-
we observe that researchers, in the past, used either diag- tion that is similar to zero-shot learning [29], and Graph
nostic reports or pathological images for screening these Neural Network (GNN) [179] aided learning for region and
diseases in most of the cases. Unfortunately, to the best of boundary aggregation [137] during the training process or
our knowledge, no method has used both. This might be due can generate synthetic data for training using a simulation-
to the non-availability of multi-modal patient data [128] in supervision approach [192, 193].
the public domain. By multi-modal data, we mean that both
diagnosis reports and pathological images for each subject in 9.7 Maintaining Ethical and Legal Aspects
the dataset. In this context, it is to be noted that recent IoT-
enabled smart health care systems are suggesting the use of AI in the medical domain started its journey in the early
multi-modal data for better patient care and more accurate 1970s when researchers were fascinated in this domain
disease diagnosis. Moreover, Sánchez-Cauce et al. [185] because of its huge real-life applications [159]. Till then,
showed that the use of patient information like age and sex the world has observed the development of several decision
with thermal images improves the overall performance of the support systems keeping AI at the backbone. Even in many
breast cancer detection system they designed. However, the cases, it has been observed that the AI-assisted methods are
lack of such data in the public domain limits the evolution outperforming human precision [150, 211]. However, the
of computer aided multi-modal disease screening systems performance of these methods is largely dependent on the
that can overcome challenges identified by Madabhushi training data and thus prejudiced samples in the training set
et al. [128]. Hence, the preparation of publicly available might hamper the end outcome. Hence, data verification and
multi-modal data can be another important future research validation by some medical experts while preparing such
direction. training samples are needed from ethical perspectives. Even
the consent of subjects, who are undergoing such experi-
9.6 Medical Image Segmentation ments, must be taken before conducting experiments and
releasing their data in public. Moreover, a notable loophole
A medical image segmentation technique aims at distin- in the existing AI-assisted methods is that the majority of
guishing the pixels of an organ or a lesion from the back- the algorithms fail to explain the reason behind their pre-
ground present in the pathological image. This is considered diction to the medical practitioners. Therefore, researchers
to be one of the most challenging tasks in medical image of a newly developed AI-assisted algorithm in the medical
analysis, which is a major component of AI-based diagno- domain should strongly make their algorithm explainable
sis models by researchers. A medical image segmentation on ethical grounds. Such initiatives would help the medical
based disease diagnosis system assists the experts by deliv- experts to spot any error present therein, and thereby would
ering critical information like the shape and volume of these increase the reliability of the algorithm. Furthermore, devel-
organs [86]. Even automatic segmentation of organs/lesions oped systems must mention which parts of a decision or
can assist the physician to monitor their growth and make what kinds of actions are taken by the AI based system. Any
surgical decisions [26]. In the early days of medical image hypothesis and consideration made by the AI-assisted algo-
segmentation methods, researchers relied on traditional rithm must be supported by some state-of-the-art medical
methods like edge detection, morphological operations, findings. The communication required for a designed sys-
skeletonization and various statistical properties. However, tem needs to be transparent i.e., the precautions that should
in the latter stage, handcrafted features based classification be taken and measurable damage due to malfunction are to
and clustering dominated the domain for a long period. At be mentioned clearly. Robustness of such systems is also
this stage, designing and extracting better features remained expected. Interested readers are suggested reading the article
the primary concern for developing such a method. In recent

13
5560 S. Malakar et al.

by Muller et al. [150] to know more about ethical aspects directions that need to be considered by researchers in order
related to practical AI assisted medical systems. to make the systems usable for medical professionals.
The legal aspect is another big concern when developing
and deploying any AI assisted systems. The present regula- Acknowledgements We are thankful to the Center for Microproces-
sor Applications for Training Education and Research (CMATER)
tions and acts (e.g., Artificial Intelligence Act23) related to research laboratory of the Computer Science and Engineering Depart-
legal consequences are mainly designed to protect funda- ment, Jadavpur University, Kolkata, India for providing infrastructural
mental rights, also known as human rights, as an erroneous support.
outcome from an AI-assisted medical system can cause seri-
ous physical and mental consequences of a subject. How- Funding The authors gratefully acknowledge financial support from
ANID PIA/APOYO AFB180003.
ever, the present regulations and acts may be insufficient
considering the evolving nature of AI in the medical system.
Even in many cases, for example, the Artificial Intelligence
Declarations
Act, the accountability and liability for the damage caused Conflict of interest All the authors declare that they have no conflict of
by an AI-assisted medical system is not addressed clearly interest or competing interests
[211]. Hence, in the future, more insightful discussions on
the current legal aspects such as regulations, acts and recom-
mendations can help increase the reliability and end users’ References
acceptability of the AI based medical systems.
1. Abdalrada AS, Yahya OH, Alaidi AHM, Hussein NA, Alrikabi
HT, Al-Quraishi TAQ (2019) A predictive model for liver disease
progression based on logistic regression algorithm. Periodicals
10 Conclusion of Engineering and Natural Sciences (PEN) 7(3):1255, https://​
doi.​org/​10.​21533/​pen.​v7i3.​667,
Chronic diseases generally persist for a longer period in 2. Abdeldaim AM, Sahlol AT, Elhoseny M, Hassanien AE (2017)
human organs and most importantly, there are no vaccines Computer-aided acute lymphoblastic leukemia diagnosis sys-
tem based on image analysis. In: Advances in Soft Comput-
to prevent these diseases. However, early detection of these ing and Machine Learning in Image Processing, Springer
diseases can save many human lives. Nowadays, technol- International Publishing, pp 131–147. https://​doi.​org/​10.​1007/​
ogy plays a significant role in healthcare systems, and there 978-3-​319-​63754-9_7,
are many computer-aided supporting systems available. 3. Abdul-Aziz AA, Desikan P, Prabhakaran D, Schroeder LF (2019)
Tackling the burden of cardiovascular diseases in India. Circula-
Owing to the huge social impact of this research field, many tion. https://​doi.​org/​10.​1161/​circo​utcom​es.​118.​005195
researchers from various domains like computer science, 4. Adil SH, Ebrahim M, Raza K, Ali SSA, Hashmani MA (2018)
biology, medicine and statistics have contributed to this Liver patient classification using logistic regression. In: 2018 4th
domain. To this end, the present survey makes a significant International Conference on Computer and Information Sciences
(ICCOINS), IEEE, pp 1–5. https://d​ oi.o​ rg/1​ 0.1​ 109/i​ ccoin​ s.2​ 018.​
effort to record the current trends in research on chronic 85105​81,
diseases like Breast Cancer, Leukemia, Lung Cancer, Heart 5. Ahmadi H, Gholamzadeh M, Shahmoradi L, Nilashi M, Rash-
Diseases, Diabetes, CKD and Liver. It covers methods vand P (2018) Diseases diagnosis using fuzzy logic methods: a
including missing value imputation, feature reduction, FS, systematic and meta-analysis review. Comput Methods Programs
Biomed 161:145–172
classifier combination, and ML and DL based approaches. 6. Ahmed S, Ghosh KK, Singh PK, Geem ZW, Sarkar R (2020)
The present survey puts more efforts on automatic disease Hybrid of Harmony Search Algorithm and Ring Theory-Based
prediction systems that used ML and DL based methods for Evolutionary Algorithm for Feature Selection. IEEE Access
the prediction of the said diseases. Moreover, this survey dis- 8:102629–102645
7. Akter O, Moni MA, Islam MM, Quinn JM, Kamal A (2021) Lung
cusses the methods that considered diagnostic report based cancer detection using enhanced segmentation accuracy. Appl
and images based inputs. Along with this, it mentions the Intell 51(6):3391–3404
advantages and limitations of the methods reported in the 8. Al-Milli N (2013) Backpropagation neural network for prediction
literature wherever possible. Additionally, this paper pre- of heart disease. J Theor Appl Inf Technol 56(1):131–135
9. Al-Tarawneh MS (2012) Lung cancer detection using image
sents various comparative studies of the methods proposed processing techniques. Leonardo Electron J Pract Technol
by different researchers. Though some significant develop- 11(21):147–58
ments have been made in the automation of AI-enabled deci- 10. Alam J, Alam S, Hossan A (2018) Multi-stage lung cancer detec-
sion support systems, there are still some prominent gaps tion and prediction using multi-class SVM classifie. In: 2018
International Conference on Computer, Communication, Chemi-
that hinder the systems from using them for practical pur- cal, Material and Electronic Engineering (IC4ME2), IEEE, pp
poses. Hence, this survey also tries to bring out some future 1–4. https://​doi.​org/​10.​1109/​ic4me2.​2018.​84655​93,
11. Aljaaf AJ, Al-Jumeily D, Haglan HM, Alloghani M, Baker T,
Hussain AJ, Mustafina J (2018) Early prediction of chronic
23
kidney disease using machine learning supported by predictive
https://​bit.​ly/​3AcDs​Ca

13
Computer Based Diagnosis of Some Chronic Diseases: A Medical Journey of the Last Two Decades 5561

analytics. In: 2018 IEEE Congress on Evolutionary Computation 26. Basak H, Kundu R, Sarkar R (2022) MFSNet: a multi focus seg-
(CEC), IEEE, pp 1–9. https://d​ oi.o​ rg/1​ 0.1​ 109/c​ ec.2​ 018.8​ 47787​ 6, mentation network for skin lesion segmentation. Pattern Recog-
12. Almansour NA, Syed HF, Khayat NR, Altheeb RK, Juri RE, nition, 108673
Alhiyafi J, Alrashed S, Olatunji SO (2019) Neural network and 27. Begum S, Chakraborty D, Sarkar R (2016) Identifying cancer
support vector machine for the prediction of chronic kidney biomarkers from leukemia data using feature selection and super-
disease: a comparative study. Comput Biol Med 109:101–111. vised learning. In: 2016 IEEE First International Conference on
https://​doi.​org/​10.​1016/j.​compb​iomed.​2019.​04.​017 Control, Measurement and Instrumentation (CMI), IEEE, pp
13. Alzubaidi MA, Otoom M, Jaradat H (2021) Comprehensive 249–253. https://​doi.​org/​10.​1109/​CMI.​2016.​74137​49
and comparative global and local feature extraction framework 28. Bhowal P, Sen S, Velasquez JD, Sarkar R (2022) Fuzzy ensemble
for lung cancer detection using ct scan images. IEEE Access of deep learning models using choquet fuzzy integral, coalition
9:158140–158154 game and information theory for breast cancer histology clas-
14. Amin MS, Chiam YK, Varathan KD (2019) Identification of sig- sification. Expert Syst Appl 190:116167
nificant features and data mining techniques in predicting heart 29. Bian C, Yuan C, Ma K, Yu S, Wei D, Zheng Y (2021) Domain
disease. Telematics Inform 36:82–93. https://​doi.​org/​10.​1016/j.​ adaptation meets zero-shot learning: an annotation-efficient
tele.​2018.​11.​007 approach to multi-modality medical image segmentation. IEEE
15. Amirgaliyev Y, Shamiluulu S, Serek A (2018) Analysis of Trans Med Imaging 41(5):1043–1056
chronic kidney disease dataset by applying machine learning 30. Brancati N, Frucci M, Riccio D (2018) Multi-classification of
methods. In: 2018 IEEE 12th International Conference on Appli- breast cancer histology images by using a fine-tuning strategy.
cation of Information and Communication Technologies (AICT), In: International conference image analysis and recognition.
IEEE, pp 1–4. https://​doi.​org/​10.​1109/​icaict.​2018.​87471​40, Springer, pp 771–778. https://d​ oi.o​ rg/1​ 0.1​ 007/9​ 78-3-3​ 19-9​ 3000-
16. Anifah L, Haryanto, Harimurti R, Permatasari Z, Rusimamto 8_​87
PW, Muhamad AR (2017) Cancer lungs detection on CT scan 31. Brooks JP (2011) Support vector machines with the ramp loss
image using artificial neural network backpropagation based gray and the hard margin loss. Oper Res 59(2):467–479
level coocurrence matrices feature. In: 2017 International Con- 32. Carrington AM, Fieguth PW, Qazi H, Holzinger A, Chen HH,
ference on Advanced Computer Science and Information Sys- Mayr F, Manuel DG (2020) A new concordant partial AUC
tems (ICACSIS), IEEE, pp 327–332. https://​doi.​org/​10.​1109/​ and partial c statistic for imbalanced data in the evaluation of
icacs​is.​2017.​83550​54, machine learning algorithms. BMC Med Inform Decis Mak
17. Aresta G, Araújo T, Kwok S, Chennamsetty SS, Safwan M, Alex 20(1):1–12
V, Marami B, Prastawa M, Chan M, Donovan M et al (2019) 33. Chakraborty B, Chaterjee A, Malakar S, Sarkar R (2022) An
Bach: grand challenge on breast cancer histology images. Med iterative approach to unsupervised outlier detection using ensem-
Image Anal 56:122–139 ble method and distance-based data filterings. Complex Intell
18. Arif-Ul-Islam, Ripon SH (2019) Rule induction and prediction of Syst. https://​doi.​org/​10.​1007/​s40747-​022-​00674-0
chronic kidney disease using boosting classifiers, ant-miner and 34. Chakraborty B, Chaterjee A, Malakar S, Sarkar R (2022) An iter-
j48 decision tree. In: 2019 International Conference on Electri- ative approach to unsupervised outlier detection using ensemble
cal, Computer and Communication Engineering (ECCE), IEEE, method and distance-based data filtering. Complex Intell Syst.
pp 1–6. https://​doi.​org/​10.​1109/​ecace.​2019.​86793​88, https://​doi.​org/​10.​1007/​s40747-​022-​00674-0
19. Ashiquzzaman A, Tushar AK, Islam MR, Shon D, Im K, Park 35. Charleonnan A, Fufaung T, Niyomwong T, Chokchueypattanakit
JH, Lim DS, Kim J (2017) Reduction of overfitting in diabetes W, Suwannawach S, Ninchawee N (2016) Predictive analytics
prediction using deep learning neural network. In: IT Conver- for chronic kidney disease using machine learning techniques.
gence and Security 2017, Springer Singapore, pp 35–43. https://​ In: 2016 Management and Innovation Technology International
doi.​org/​10.​1007/​978-​981-​10-​6451-7_5 Conference (MITicon), IEEE, pp MIT–80. https://​doi.​org/​10.​
20. Auxilia LA (2018) Accuracy prediction using machine learning 1109/​mitic​on.​2016.​80252​42
techniques for indian patient liver disease. In: 2018 2nd Inter- 36. Chatterjee B, Bhattacharyya T, Ghosh KK, Singh PK, Geem
national Conference on Trends in Electronics and Informatics ZW, Sarkar R (2020) Late acceptance hill climbing based
(ICOEI), IEEE, pp 45–50. https://​doi.​org/​10.​1109/​icoei.​2018.​ social ski driver algorithm for feature selection. IEEE Access
85536​82 8:75393–75408
21. Avci E, Karakus S, Ozmen O, Avci D (2018) Performance com- 37. Chatterjee S, Biswas S, Majee A, Sen S, Oliva D, Sarkar R
parison of some classifiers on chronic kidney disease data. In: (2021) Breast cancer detection from thermal images using a
2018 6th International Symposium on Digital Forensic and Secu- grunwald-letnikov-aided dragonfly algorithm-based deep feature
rity (ISDFS), IEEE, pp 1–4. https://​doi.​org/​10.​1109/​isdfs.​2018.​ selection method. Comput Biol Med 141:105027
83553​92 38. Chattopadhyay S, Dey A, Singh PK, Sarkar R (2022) Drda-net:
22. Babu MP, Ramjee M, Katta S, K S (2016) Implementation of Dense residual dual-shuffle attention network for breast cancer
partitional clustering on ILPD dataset to predict liver disorders. classification using histopathological images. Comput Biol Med
In: 2016 7th IEEE International Conference on Software Engi- 145:105437
neering and Service Science (ICSESS), IEEE, pp 1094–1097. 39. Chen W, Chen S, Zhang H, Wu T (2017) A hybrid prediction
https://​doi.​org/​10.​1109/​icsess.​2016.​78832​56 model for type 2 diabetes using k-means and decision tree. In:
23. Bache K, Lichman M (2013) Uci machine learning repository 2017 8th IEEE International Conference on Software Engineer-
[http://archive. ics. uci. edu/ml]. irvine, ca: University of Cali- ing and Service Science (ICSESS), IEEE, pp 386–390. https://​
fornia. School of Information and Computer Science 28 doi.​org/​10.​1109/​icsess.​2017.​83429​38
24. Baldomero-Naranjo M, Martinez-Merino LI, Rodriguez-Chia 40. Chennamsetty SS, Safwan M, Alex V (2018) Classification of
AM (2021) A robust svm-based approach with feature selection breast cancer histology image using ensemble of pre-trained
and outliers detection for classification problems. Expert Syst neural networks. In: International Conference Image Analysis
Appl 178:115017 and Recognition, Springer, pp 804–811. https://​doi.​org/​10.​1007/​
25. Baliarsingh SK, Muhammad K, Bakshi S (2021) SARA: a 978-3-​319-​93000-8_​91
memetic algorithm for high-dimensional biomedical data. Appl 41. Chetty N, Vaisla KS, Sudarsan SD (2015) Role of attributes
Soft Comput 101:107009 selection in classification of chronic kidney disease patients. In:

13
5562 S. Malakar et al.

2015 International Conference on Computing, Communication ensemble classification methods in diabetes disease: a review.
and Security (ICCCS), IEEE, pp 1–6. https://​doi.​org/​10.​1109/​ In: 2019 41st Annual International Conference of the IEEE Engi-
cccs.​2015.​73741​93 neering in Medicine and Biology Society (EMBC), IEEE, pp
42. Chollet F (2017) Xception: Deep learning with depthwise sepa- 3956–3959
rable convolutions. In: Proceedings of the IEEE conference on 60. Feshki MG, Shijani OS (2016) Improving the heart disease diag-
computer vision and pattern recognition, pp 1251–1258 nosis by evolutionary algorithm of pso and feed forward neural
43. Choudhury SJ, Pal NR (2019) Imputation of missing data with network. In: 2016 Artificial Intelligence and Robotics (IRAN-
neural networks for classification. Knowl-Based Syst. https://d​ oi.​ OPEN), IEEE, pp 48–53. https://​doi.​org/​10.​1109/​RIOS.​2016.​
org/​10.​1016/j.​knosys.​2019.​07.​009 75294​89
44. Chouhan N, Khan A, Shah JZ, Hussnain M, Khan MW (2021) 61. Fu L et al (2015) The discriminate analysis and dimension reduc-
Deep convolutional neural network and emotional learning based tion methods of high dimension. Open J Soc Sci 3(03):7. https://​
breast cancer detection using digital mammography. Comput doi.​org/​10.​4236/​jss.​2015.​33002
Biol Med 132:104318 62. Gal O, Auslander N, Fan Y, Meerzaman D (2019) Pre-
45. Chua T, Tan W (2009) A new fuzzy rule-based initialization dicting complete remission of acute myeloid leukemia:
method for k-nearest neighbor classifier. In: 2009 IEEE Interna- machine learning applied to gene expression. Cancer Inform
tional Conference on Fuzzy Systems, IEEE, pp 415–420. https://​ 18:1176935119835544. https://​doi.​org/​10.​1177/​11769​35119​
doi.​org/​10.​1109/​fuzzy.​2009.​52772​15 835544
46. Daanouni O, Cherradi B, Tmiri A (2020) Diabetes diseases pre- 63. Gao X, Liu X (2018) A novel effective diagnosis model based
diction using supervised machine learning and neighbourhood on optimized least squares support machine for gene microarray.
components analysis. In: Proceedings of the 3rd International Appl Soft Comput 66:50–59. https://d​ oi.o​ rg/1​ 0.1​ 016/j.a​ soc.2​ 018.​
Conference on Networking, Information Systems & Security, 02.​009
ACM, pp 1–5. https://​doi.​org/​10.​1145/​33867​23.​33878​87 64. Gavhane A, Kokkula G, Pandya I, Devadkar K (2018) Predic-
47. Danjuma KJ (2015) Performance evaluation of machine learning tion of heart disease using machine learning. In: 2018 Second
algorithms in post-operative life expectancy in the lung cancer International Conference on Electronics, Communication and
patients. arXiv preprint arXiv:​1504.​04646 Aerospace Technology (ICECA), IEEE, pp 1275–1278. https://​
48. Das R, Turkoglu I, Sengur A (2009) Effective diagnosis of heart doi.​org/​10.​1109/​ICECA.​2018.​84749​22
disease through neural networks ensembles. Expert Syst Appl 65. Georgescu MI, Ionescu RT, Popescu M (2019) Local learning
36(4):7675–7680. https://​doi.​org/​10.​1016/j.​eswa.​2008.​09.​013 with deep and handcrafted features for facial expression recogni-
49. Dey C, Bose R, Ghosh KK, Malakar S, Sarkar R (2021) Lagoa: tion. IEEE Access 7:64827–64836
learning automata based grasshopper optimization algorithm for 66. Ghosh KK, Singh PK, Hong J, Geem ZW, Sarkar R (2020)
feature selection in disease datasets. J Ambient Intell Hum Com- Binary social mimic optimization algorithm with x-shaped trans-
put 13(6):3175–3194 fer function for feature selection. IEEE Access. https://​doi.​org/​
50. Dey S, Bhattacharya R, Malakar S, Mirjalili S, Sarkar R (2021) 10.​1109/​ACCESS.​2020.​29966​11
Choquet fuzzy integral-based classifier ensemble technique for 67. Ghosh KK, Guha R, Bera SK, Kumar N, Sarkar R (2021)
covid-19 detection. Comput Biology Med 135:104585 S-shaped versus V-shaped transfer functions for binary Manta
51. Dey S, Roychoudhury R, Malakar S, Sarkar R (2022) An opti- ray foraging optimization in feature selection problem. Neural
mized fuzzy ensemble of convolutional neural networks for Comput Appl 33(17):11027–11041
detecting tuberculosis from chest x-ray images. Appl Soft Com- 68. Gogi VJ, MN V (2018) Prognosis of liver disease: using machine
put 114:108094 learning algorithms. In: 2018 International Conference on Recent
52. Dey S, Roychoudhury R, Malakar S, Sarkar R (2022b) Screen- Innovations in Electrical, Electronics & Communication Engi-
ing of breast cancer from thermogram images by edge detection neering (ICRIEECE), IEEE, pp 875–879. https:// ​doi.​org/​10.​
aided deep transfer learning model. Multimedia Tools and Appli- 1109/​icrie​ece44​171.​2018.​90084​82
cations pp 1–19 69. Gokulnath CB, Shantharajah S (2019) An optimized feature
53. Doppalapudi S, Qiu RG, Badr Y (2021) Lung cancer survival selection based on genetic approach and support vector machine
period prediction and understanding: deep learning approaches. for heart disease. Clust Comput 22(6):14777–14787. https://​doi.​
Int J Med Inform 148:104371 org/​10.​1007/​s10586-​018-​2416-4
54. Dutta D, Paul D, Ghosh P (2018) Analysing feature importances 70. Golub TR, Slonim DK, Tamayo P, Huard C, Gaasenbeek M,
for diabetes prediction using machine learning. In: 2018 IEEE Mesirov JP, Coller H, Loh ML, Downing JR, Caligiuri MA
9th Annual Information Technology, Electronics and Mobile et al (1999) Molecular classification of cancer: class discovery
Communication Conference (IEMCON), IEEE, pp 924–928, and class prediction by gene expression monitoring. Science
https://​doi.​org/​10.​1109/​iemcon.​2018.​86148​71 286(5439):531–537. https://​doi.​org/​10.​1126/​scien​ce.​286.​5439.​
55. Dwivedi AK (2018) Artificial neural network model for effec- 531
tive cancer classification using microarray gene expression data. 71. Goodfellow IJ, Pouget-Abadie J, Mirza M, Xu B, Warde-Farley
Neural Comput Appl 29(12):1545–1554. https://d​ oi.o​ rg/1​ 0.1​ 007/​ D, Ozair S, Courville A, Bengio Y (2014) Generative adversarial
s00521-​016-​2701-1 networks. 1406.2661
56. Dwivedi AK (2018) Performance evaluation of different 72. Guha R, Ghosh M, Chakrabarti A, Sarkar R, Mirjalili S (2020)
machine learning techniques for prediction of heart disease. Introducing clustering based population in Binary Gravitational
Neural Comput Appl 29(10):685–693. https://​doi.​org/​10.​1007/​ Search Algorithm for Feature Selection. Appl Soft Comput
s00521-​016-​2604-1 93:106341
57. Fatima M, Pasha M et al (2017) Survey of machine learning algo- 73. Guha R, Ghosh M, Mutsuddi S, Sarkar R, Mirjalili S (2020)
rithms for disease diagnostic. J Intell Learn Syst Appl 9(01):1 Embedded chaotic whale survival algorithm for filter-wrapper
58. Feig SA, Yaffe MJ (1995) Digital mammography, computer- feature selection. Soft Comput 24(17):12821–12843
aided diagnosis, and telemammography. Radiol Clin N Am 74. de Guia JD, Concepcion RS, Bandala AA, Dadios EP (2019) Per-
33(6):1205 formance comparison of classification algorithms for diagnosing
59. Fernández-Alemán JL, Carrillo-de Gea JM, Hosni M, Idri A, chronic kidney disease. In: 2019 IEEE 11th International Confer-
García-Mateos G (2019) Homogeneous and heterogeneous ence on Humanoid, Nanotechnology, Information Technology,

13
Computer Based Diagnosis of Some Chronic Diseases: A Medical Journey of the Last Two Decades 5563

Communication and Control, Environment, and Management ( Conference of the IEEE Engineering in Medicine and Biology
HNICEM ), IEEE, pp 1–7. https://d​ oi.o​ rg/1​ 0.1​ 109/h​ nicem
​ 48295.​ Society (EMBC), IEEE, pp 1367–1370
2019.​90735​68 92. Hosni M, García-Mateos G, Carrillo-de Gea JM, Idri A, Fernán-
75. Gultepe Y (2021) Performance of Lung Cancer Prediction Meth- dez-Alemán JL (2020) A mapping study of ensemble classifica-
ods Using Different Classification Algorithms. CMC-COMPUT- tion methods in lung cancer decision support systems. Med Biol
ERS MATERIALS & CONTINUA 67(2) Eng Comput 58(10):2177–2193
76. Gunarathne W, Perera K, Kahandawaarachchi K (2017) Perfor- 93. Houssein EH, Emam MM, Ali AA, Suganthan PN (2021)
mance evaluation on machine learning classification techniques Deep and machine learning techniques for medical imaging-
for disease classification and forecasting through data analytics based breast cancer: a comprehensive review. Expert Syst Appl
for chronic kidney disease (CKD). In: 2017 IEEE 17th Interna- 167:114161
tional Conference on Bioinformatics and Bioengineering (BIBE), 94. Hsieh SH, Wang Z, Cheng PH, Lee IS, Hsieh SL, Lai F (2010)
IEEE, pp 291–296. https://​doi.​org/​10.​1109/​bibe.​2017.​00-​39 Leukemia cancer classification based on support vector machine.
77. Gupta K, Chawla N (2020) Analysis of histopathological images In: 2010 8th IEEE International Conference on Industrial Infor-
for prediction of breast cancer using traditional classifiers with matics, IEEE, pp 819–824. https://d​ oi.o​ rg/1​ 0.1​ 109/I​ NDIN.2​ 010.​
pre-trained CNN. Procedia Comput Sci 167:878–889. https://d​ oi.​ 55496​38
org/​10.​1016/j.​procs.​2020.​03.​427 95. Hsu CH, Chen X, Lin W, Jiang C, Zhang Y, Hao Z, Chung YC
78. Gupta S, Gupta MK (2021) A comparative analysis of deep (2021) Effective multiple cancer disease diagnosis frameworks
learning approaches for predicting breast cancer surviv- for improved healthcare using machine learning. Measurement
ability. Arch Comput Methods Eng. https://​doi.​org/​10.​1007/​ 175:109145
s11831-​021-​09679-​317 96. Huang G, Liu Z, Van Der Maaten L, Weinberger KQ (2017a)
79. Han Z, Wei B, Zheng Y, Yin Y, Li K, Li S (2017) Breast cancer Densely connected convolutional networks. In: Proceedings of
multi-classification from histopathological images with struc- the IEEE Conference on Computer Vision and Pattern Recogni-
tured deep learning model. Sci Rep 7(1):1–10 tion, pp 4700–4708
80. Haque MR, Islam MM, Iqbal H, Reza MS, Hasan MK (2018) 97. Huang GB, Zhu QY, Siew CK (2006) Extreme learning machine:
Performance evaluation of random forests and artificial neural theory and applications. Neurocomputing 70(1–3):489–501.
networks for the classification of liver disorder. In: 2018 Inter- https://​doi.​org/​10.​1016/j.​neucom.​2005.​12.​126
national Conference on Computer, Communication, Chemical, 98. Huang MW, Chen CW, Lin WC, Ke SW, Tsai CF (2017) Svm
Material and Electronic Engineering (IC4ME2), IEEE, pp 1–5. and svm ensembles in breast cancer prediction. PLoS ONE
https://​doi.​org/​10.​1109/​ic4me2.​2018.​84656​58 12(1):e0161501. https://​doi.​org/​10.​1371/​journ​al.​pone.​01615​01
81. Harimoorthy K, Thangavelu M (2021) Multi-disease pre- 99. Ilango BS, Ramaraj N (2010) A hybrid prediction model with
diction model using improved svm-radial bias technique in f-score feature selection for type II diabetes databases. In: Pro-
healthcare monitoring system. J Ambient Intell Hum Comput ceedings of the 1st Amrita ACM-W Celebration on Women in
12(3):3715–3723 Computing in India—A2CWiC ’10, ACM Press, pp 1–4. https://​
82. Hasan MK, Alam MA, Das D, Hossain E, Hasan M (2020) Dia- doi.​org/​10.​1145/​18583​78.​18583​91
betes prediction using ensembling of different machine learning 100. Ismaeel S, Miri A, Chourishi D (2015) Using the extreme learn-
classifiers. IEEE Access 8:76516–76531. https://d​ oi.o​ rg/1​ 0.1​ 109/​ ing machine (elm) technique for heart disease diagnosis. In: 2015
access.​2020.​29898​57 IEEE Canada International Humanitarian Technology Confer-
83. Hashem EM, Mabrouk MS (2014) A study of support vector ence (IHTC2015), IEEE, pp 1–3
machine algorithm for liver disease diagnosis. Am J Intell Syst 101. Jabbar M, Deekshatulu B, Chandra P (2015) Computational intel-
4:9–14 ligence technique for early diagnosis of heart disease. In: 2015
84. Hashi EK, Zaman MSU, Hasan MR (2017) An expert clinical IEEE International Conference on Engineering and Technology
decision support system to predict disease using classification (ICETECH), IEEE, pp 1–6. https://​doi.​org/​10.​1109/​ICETE​CH.​
techniques. In: 2017 International Conference on Electrical, 2015.​72750​01,
Computer and Communication Engineering (ECCE), IEEE, pp 102. Jahangir M, Afzal H, Ahmed M, Khurshid K, Nawaz R (2017)
396–400. https://​doi.​org/​10.​1109/​ecace.​2017.​79129​37 An expert system for diabetes prediction using auto tuned multi-
85. He K, Zhang X, Ren S, Sun J (2016) Deep residual learning for layer perceptron. In: 2017 Intelligent Systems Conference (Intel-
image recognition. In: Proceedings of the IEEE Conference on liSys), IEEE, pp 722–728. https://​doi.​org/​10.​1109/​intel​lisys.​
Computer Vision and Pattern Recognition, pp 770–778 2017.​83242​09,
86. Hesamian MH, Jia W, He X, Kennedy P (2019) Deep learning 103. Jain D, Singh V (2018) Feature selection and classification sys-
techniques for medical image segmentation: achievements and tems for chronic disease prediction: a review. Egypt Inform J
challenges. J Digit Imaging 32(4):582–596 19(3):179–189
87. Hinton GE, Salakhutdinov RR (2006) Reducing the dimension- 104. Jakimovski G, Davcev D (2018) Lung cancer medical image rec-
ality of data with neural networks. Science 313(5786):504–507. ognition using deep neural networks. In: 2018 Thirteenth Interna-
https://​doi.​org/​10.​1126/​scien​ce.​11276​47 tional Conference on Digital Information Management (ICDIM),
88. Holte RC (1993) Very simple classification rules perform well on IEEE, pp 1–5. https://​doi.​org/​10.​1109/​icdim.​2018.​88471​36
most commonly used datasets. Mach Learn 11(1):63–90. https://​ 105. Jannesari M, Habibzadeh M, Aboulkheyr H, Khosravi P, Ele-
doi.​org/​10.​1023/A:​10226​31118​932 mento O, Totonchi M, Hajirasouliha I (2018) Breast cancer his-
89. Holzinger A (2016) Machine Learning for Health Informatics. topathological image classification: a deep learning approach.
In: Machine learning for health informatics, Springer, pp 1–24 In: 2018 IEEE International Conference on Bioinformatics and
90. Hosni M, Abnane I, Idri A, de Gea JMC, Alemán JLF (2019) Biomedicine (BIBM), IEEE, p 242818. https://​doi.​org/​10.​1109/​
Reviewing ensemble classification methods in breast cancer. bibm.​2018.​86213​07
Comput Methods Programs Biomed 177:89–112 106. Jayaraman V, Sultana HP (2019) Artificial gravitational cuckoo
91. Hosni M, Carrillo-de Gea JM, Idri A, Fernández-Alemán JL, search algorithm along with particle bee optimized associative
García-Berná JA (2019b) Using ensemble classification meth- memory neural network for feature selection in heart disease
ods in lung cancer disease. In: 2019 41st Annual International classification. J Ambient Intell Hum Comput. https://​doi.​org/​10.​
1007/​s12652-​019-​01193-6

13
5564 S. Malakar et al.

107. Jog A, Carass A, Roy S, Pham DL, Prince JL (2015) MR image 122. Kumari M, Singh V (2018) Breast cancer prediction system. Pro-
synthesis by contrast learning on neighborhood ensembles. Med cedia Comput Sci 132:371–376. https://​doi.​org/​10.​1016/j.​procs.​
Image Anal 24(1):63–76 2018.​05.​197
108. Johari AA, Wahab MHA, Mustapha A (2019) Two-class clas- 123. Labbé M, Martínez-Merino LI, Rodríguez-Chía AM (2019)
sification: Comparative experiments for chronic kidney disease. Mixed integer linear programming for feature selection in sup-
In: 2019 4th International Conference on Information Systems port vector machine. Discret Appl Math 261:276–304
and Computer Networks (ISCON), IEEE, pp 789–792. https://​ 124. Lakshmi J, Jayanthi K, Sathya M (2017) Investigating ilpd for
doi.​org/​10.​1109/​iscon​47742.​2019.​90363​06 most significant features. Int J Mech Eng Technol 8:741–749
109. Kamel SR, YaghoubZadeh R, Kheirabadi M (2019) Improv- 125. Li D, Zhang G, Wu Z, Yi L (2010) An edge embedded marker-
ing the performance of support-vector machine by selecting the based watershed algorithm for high spatial resolution remote
best features by gray wolf algorithm to increase the accuracy of sensing image segmentation. IEEE Trans Image Process
diagnosis of breast cancer. J Big Data 6(1):90. https://d​ oi.o​ rg/1​ 0.​ 19(10):2781–2787. https://​doi.​org/​10.​1109/​tip.​2010.​20495​28
1186/​s40537-​019-​0247-7 126. Lin WC, Tsai CF (2020) Missing value imputation: a review
110. Kanchan BD, Kishor MM (2016) Study of machine learning and analysis of the literature (2006–2017). Artif Intell Rev
algorithms for special disease prediction using principal of com- 53(2):1487–1509
ponent analysis. In: 2016 International Conference on Global 127. Lv C, Chen C, Chen Q, Zhai H, Zhao L, Guo Y, Wang N (2019)
Trends in Signal Processing, Information Computing and Com- Multiple pregnancies and the risk of diabetes mellitus in post-
munication (ICGTSPICC), IEEE, pp 5–10. https://​doi.​org/​10.​ menopausal women. Menopause 26(9):1010–1015
1109/​ICGTS​PICC.​2016.​79552​60 128. Madabhushi A, Agner S, Basavanhally A, Doyle S, Lee G (2011)
111. Kannadasan K, Edla DR, Kuppili V (2019) Type 2 diabetes data Computer-aided prognosis: predicting patient and disease out-
classification using stacked autoencoders in deep neural net- come via quantitative fusion of multi-scale, multi-modal data.
works. Clin Epidemiol Global Health 7(4):530–535. https://​doi.​ Comput Med Imaging Graph 35(7–8):506–514
org/​10.​1016/j.​cegh.​2018.​12.​004 129. Mafarja MM, Mirjalili S (2017) Hybrid whale optimization algo-
112. Kant S, Ansari IA (2015) An improved k means clustering rithm with simulated annealing for feature selection. Neurocom-
with Atkinson index to classify liver patient dataset. Int J Syst puting 260:302–312. https://​doi.​org/​10.​1016/j.​neucom.​2017.​04.​
Assurance Eng Manag 7(S1):222–228. https://​doi.​org/​10.​1007/​ 053
s13198-​015-​0365-3 130. Maji S, Arora S (2019) Decision tree algorithms for prediction of
113. Karayılan T, Kılıç Ö (2017) Prediction of heart disease using heart disease. In: Information and Communication Technology
neural network. In: 2017 International Conference on Computer for Competitive Strategies, Springer, pp 447–454. https://d​ oi.o​ rg/​
Science and Engineering (UBMK), IEEE, pp 719–723. https://​ 10.​1007/​978-​981-​13-​0586-3_​45
doi.​org/​10.​1109/​UBMK.​2017.​80935​12 131. Malakar S, Sharma P, Singh PK, Das M, Sarkar R, Nasipuri M
114. Kavitha K, Gopinath A, Gopi M (2017) Applying improved svm (2017) A holistic approach for handwritten Hindi word recogni-
classifier for leukemia cancer classification using fcbf. In: 2017 tion. Int J Comput Vis Image Process 7(1):59–78
International Conference on Advances in Computing, Communi- 132. Maleki N, Zeinali Y, Niaki STA (2021) A k-nn method for lung
cations and Informatics (ICACCI), IEEE, pp 61–66. https://​doi.​ cancer prognosis with the use of a genetic algorithm for feature
org/​10.​1109/​ICACCI.​2017.​81258​17 selection. Expert Syst Appl 164:113981
115. Khandekar R, Shastry P, Jaishankar S, Faust O, Sampathila N 133. Mandal M, Singh PK, Ijaz MF, Shafi J, Sarkar R (2021) A tri-
(2021) Automated blast cell detection for Acute Lymphoblastic stage wrapper-filter feature selection framework for disease clas-
Leukemia diagnosis. Biomed Signal Process Control 68:102690 sification. Sensors 21(16):5571
116. Kirkpatrick S, Gelatt CD, Vecchi MP (1983) Optimization by 134. Manhas J, Gupta RK, Roy PP (2021) A review on automated
simulated annealing. Science 220(4598):671–680. https://​doi.​ cancer detection in medical images using machine learning and
org/​10.​1126/​scien​ce.​220.​4598.​671 deep learning based computational techniques: Challenges and
117. Kulkarni A, Panditrao A (2014) Classification of lung cancer opportunities. Arch Comput Methods Eng 1:1–41
stages on CT scan images using image processing. In: 2014 IEEE 135. Mansourypoor F, Asadi S (2017) Development of a reinforce-
International Conference on Advanced Communications, Control ment learning-based evolutionary fuzzy rule-based system for
and Computing Technologies, IEEE, pp 1384–1388. https://​doi.​ diabetes diagnosis. Comput Biol Med 91:337–352. https://​doi.​
org/​10.​1109/​icacc​ct.​2014.​70193​27 org/​10.​1016/j.​compb​iomed.​2017.​10.​024
118. Kulkarni UV, Shinde SV (2013) Hybrid fuzzy classifier based 136. Medhekar DS, Bote MP, Deshmukh SD (2013) Heart disease pre-
on feature-wise membership given by artificial neural network. diction system using naive bayes. Int J Enhanced Res Sci Technol
In: 2013 Fourth International Conference on Computing, Com- Eng 2(3)
munications and Networking Technologies (ICCCNT), IEEE, pp 137. Meng Y, Zhang H, Zhao Y, Yang X, Qiao Y, MacCormick IJ,
1–7. https://​doi.​org/​10.​1109/​icccnt.​2013.​67265​49 Huang X, Zheng Y (2021) Graph-based region and boundary
119. Kumar A, Sahu N (2017) Categorization of liver disease using aggregation for biomedical image segmentation. IEEE Trans
classification techniques. Int J Res Appl Sci Eng Technol Med Imaging 41(3):69–701
5:826–828 138. Miao KH, Miao JH, Miao GJ (2016) Diagnosing coronary heart
120. Kumar S, Katyal S (2018) Effective analysis and diagnosis of disease using ensemble machine learning. International Journal
liver disorder by data mining. In: 2018 International Conference of Advanced Computer Science and Applications 7(10):30–39,
on Inventive Research in Computing Applications (ICIRCA), https://​doi.​org/​10.​14569/​IJACSA.​2016.​071004
IEEE, pp 1047–1051. https://​doi.​org/​10.​1109/​icirca.​2018.​85968​ 139. Mirjalili S (2015) The ant lion optimizer. Adv Eng Softw 83:80–
17 98. https://​doi.​org/​10.​1016/j.​adven​gsoft.​2015.​01.​010
121. Kumar V, Mishra BK, Mazzara M, Thanh DN, Verma A (2020) 140. Mirjalili S, Lewis A (2016) The whale optimization algorithm.
Prediction of malignant and benign breast cancer: A data mining Adv Eng Softw 95:51–67. https://​doi.​org/​10.​1016/j.​adven​gsoft.​
approach in healthcare applications. In: Advances in Data Sci- 2016.​01.​008
ence and Management, Springer, pp 435–442. https://d​ oi.o​ rg/1​ 0.​ 141. Mirjalili S, Mirjalili SM, Lewis A (2014) Grey wolf optimizer.
1007/​978-​981-​15-​0978-0_​43 Adv Eng Softw 69:46–61. https://​doi.​org/​10.​1016/j.​adven​gsoft.​
2013.​12.​007

13
Computer Based Diagnosis of Some Chronic Diseases: A Medical Journey of the Last Two Decades 5565

142. Mishra S, Sharma L, Majhi B, Sa PK (2016) Microscopic image 160. Patil B, Joshi R, Toshniwal D (2010) Hybrid prediction model
classification using DCT for the detection of acute lymphoblastic for type-2 diabetic patients. Expert Syst Appl 37(12):8102–8108.
leukemia (ALL). In: Advances in Intelligent Systems and Com- https://​doi.​org/​10.​1016/j.​eswa.​2010.​05.​078
puting, Springer Singapore, pp 171–180. https://d​ oi.o​ rg/1​ 0.1​ 007/​ 161. Patra R (2020) Prediction of lung cancer using machine learning
978-​981-​10-​2104-6_​16 classifier. In: International Conference on Computing Science,
143. Mishra S, Majhi B, Sa PK, Sharma L (2017) Gray level co- Communication and Security, Springer, pp 132–142. https://​doi.​
occurrence matrix and random forest based acute lymphoblastic org/​10.​1007/​978-​981-​15-​6648-6_​11
leukemia detection. Biomed Signal Process Control 33:272–280. 162. Paul A, Pramanik R, Malakar S, Sarkar R (2021) An ensemble
https://​doi.​org/​10.​1016/j.​bspc.​2016.​11.​021 of deep transfer learning models for handwritten music sym-
144. Mishra S, Majhi B, Sa PK (2019) Texture feature based clas- bol recognition. Neural Comput Appl. https://​doi.​org/​10.​1007/​
sification on microscopic blood smear for acute lymphoblastic s00521-​021-​06629-
leukemia detection. Biomed Signal Process Control 47:303–311. 163. Polakowski W, Cournoyer D, Rogers S, DeSimio M, Ruck D,
https://​doi.​org/​10.​1016/j.​bspc.​2018.​08.​012 Hoffmeister J, Raines R (1997) Computer-aided breast cancer
145. Mohammed ZF, Abdulla AA (2021) An efficient cad system for detection and diagnosis of masses using difference of Gaussians
all cell identification from microscopic blood images. Multimed and derivative-based feature saliency. IEEE Trans Med Imaging
Tools Appl 80(4):6355–6368 16(6):811–819. https://​doi.​org/​10.​1109/​42.​650877
146. Mohan S, Thirumalai C, Srivastava G (2019) Effective heart dis- 164. Polat H, Mehr HD, Cetin A (2017) Diagnosis of chronic kidney
ease prediction using hybrid machine learning techniques. IEEE disease based on support vector machine by feature selection
Access 7:81542–81554. https://​doi.​org/​10.​1109/​ACCESS.​2019.​ methods. J Med Syst. https://d​ oi.o​ rg/1​ 0.1​ 007/s​ 10916-0​ 17-0​ 703-x
29237​07 165. Prakash PS, Rajkumar N (2020) Improved local fisher discrimi-
147. Mondal R, Malakar S, Barney Smith EH, Sarkar R (2022) nant analysis based dimensionality reduction for cancer disease
Handwritten English word recognition using a deep learn- prediction. J Ambient Intell Hum Comput. https://​doi.​org/​10.​
ing based object detection architecture. Multimed Tools Appl 1007/​s12652-​020-​02542-6
81(1):975–1000 166. Preetha R, Jinny SV (2020) Early diagnose breast cancer
148. Moon AM, Singal AG, Tapper EB (2019) Contemporary epide- with pca-lda based fer and neuro-fuzzy classification sys-
miology of chronic liver disease and cirrhosis. Clin Gastroenterol tem. J Ambient Intell Hum Comput. https://​doi.​org/​10.​1007/​
Hepatol. https://​doi.​org/​10.​1016/j.​cgh.​2019.​07.​060 s12652-​020-​02395-z
149. Moorthy U, Gandhi UD (2020) A novel optimal feature selection 167. Pujianto U, Ramadhani NA, Wibawa AP (2018) Support vec-
technique for medical data classification using anova based whale tor machine with purified k-means clusters for chronic kidney
optimization. J Ambient Intell Hum Comput. https://​doi.​org/​10.​ disease detection. In: 2018 2nd East Indonesia Conference on
1007/​s12652-​020-​02592-w Computer and Information Technology (EIConCIT), IEEE, pp
150. Muller H, Mayrhofer MT, Van Veen EB, Holzinger A (2021) 56–60. https://​doi.​org/​10.​1109/​eicon​cit.​2018.​88785​11
The ten commandments of ethical medical AI. Computer 168. Purwar A, Singh SK (2015) Hybrid prediction model with
54(07):119–123 missing value imputation for medical data. Expert Syst Appl
151. Murty NR, Babu MP (2017) A critical study of classification 42(13):5621–5631. https://​doi.​org/​10.​1016/j.​eswa.​2015.​02.​050
algorithms for lungcancer disease detection and diagnosis. Int J 169. Radhika P, Nair RA, Veena G (2019) A comparative study of
Comput Intell Res 13(5):1041–1048 lung cancer detection using machine learning algorithms. In:
152. Nadkarni NS, Borkar S (2019) Detection of lung cancer in CT 2019 IEEE International Conference on Electrical, Computer and
images using image processing. In: 2019 3rd International Con- Communication Technologies (ICECCT), IEEE, pp 1–4. https://​
ference on Trends in Electronics and Informatics (ICOEI), IEEE, doi.​org/​10.​1109/​ICECCT.​2019.​88690​01
pp 863–866. https://​doi.​org/​10.​1109/​icoei.​2019.​88625​77 170. Rahman A, Hasan MM (2018) Automatic detection of white
153. Nasser IM, Abu-Naser SS (2019) Lung cancer detection using blood cells from microscopic images for malignancy classifi-
artificial neural network. Int J Eng Inf Syst 3(3):17–23 cation of acute lymphoblastic leukemia. In: 2018 International
154. NirmalaDevi M, Appavu S, Swathi UV (2013) An amalgam Conference on Innovation in Engineering and Technology
KNN to predict diabetes mellitus. In: 2013 IEEE International (ICIET), IEEE, pp 1–6. https://​doi.​org/​10.​1109/​ciet.​2018.​86609​
Conference ON Emerging Trends in Computing, Communication 14
and Nanotechnology (ICECCN), IEEE, pp 691–695. https://​doi.​ 171. Raja P, Thangavel K (2020) Missing value imputation using
org/​10.​1109/​ice-​ccn.​2013.​65285​91 unsupervised machine learning techniques. Soft Comput
155. Ogurtsova K, da Rocha Fernandes J, Huang Y, Linnenkamp U, 24(6):4361–4392
Guariguata L, Cho N, Cavan D, Shaw J, Makaroff L (2017) IDF 172. Rakhlin A, Shvets A, Iglovikov V, Kalinin AA (2018) Deep
diabetes atlas: global estimates for the prevalence of diabetes for convolutional neural networks for breast cancer histology image
2015 and 2040. Diabetes Res Clin Pract 128:40–50. https://​doi.​ analysis. In: Lecture Notes in Computer Science, Springer Inter-
org/​10.​1016/j.​diabr​es.​2017.​03.​024 national Publishing, pp 737–744. https://​doi.​org/​10.​1007/​978-3-​
156. Pan WT (2012) A new fruit fly optimization algorithm: taking 319-​93000-8_​83
the financial distress model as an example. Knowl-Based Syst 173. Rakshit S, Manna S, Biswas S, Kundu R, Gupta P, Maitra S,
26:69–74. https://​doi.​org/​10.​1016/j.​knosys.​2011.​07.​001 Barman S (2017) Prediction of diabetes type-II using a two-class
157. Pan Y, Ge SS, Al Mamun A (2009) Weighted locally lin- neural network. In: Communications in Computer and Informa-
ear embedding for dimension reduction. Pattern Recogn tion Science. Springer, Singapore, pp 65–71. https://​doi.​org/​10.​
42(5):798–811 1007/​978-​981-​10-​6430-2_6
158. Patel J, TejalUpadhyay D, Patel S (2015) Heart disease prediction 174. Ramana BV, Boddu RSK (2019) Performance comparison of
using machine learning and data mining technique. Heart Dis classification algorithms on medical datasets. In: 2019 IEEE 9th
7(1):129–137 Annual Computing and Communication Workshop and Confer-
159. Patel VL, Shortliffe EH, Stefanelli M, Szolovits P, Berthold MR, ence (CCWC), IEEE, pp 0140–0145. https://​doi.​org/​10.​1109/​
Bellazzi R, Abu-Hanna A (2009) The coming of age of artificial ccwc.​2019.​86664​97
intelligence in medicine. Artif Intell Med 46(1):5–17 175. Rashmi R, Prasad K, Udupa CBK (2022) Breast histopatho-
logical image analysis using image processing techniques for

13
5566 S. Malakar et al.

diagnostic purposes: a methodological review. J Med Syst patterns of vesicles in living cells. In: Proceedings of the IEEE/
46(1):1–24 CVF Conference on Computer Vision and Pattern Recognition,
176. Rawat J, Singh A, Bhadauria HS, Virmani J, Devgun JS (2017) pp 14014–14023
Classification of acute lymphoblastic Leukaemia using hybrid 193. Sekh AA, Opstad IS, Godtliebsen G, Birgisdottir ÅB, Ahluwalia
hierarchical classifiers. Multimed Tools Appl 76(18):19057– BS, Agarwal K, Prasad DK (2021) Physics-based machine learn-
19085. https://​doi.​org/​10.​1007/​s11042-​017-​4478-3 ing for subcellular segmentation in living cells. Nat Mach Intell
177. Rigla M, García-Sáez G, Pons B, Hernando ME (2018) Artificial 3(12):1071–1080
intelligence methodologies and their application to diabetes. J 194. Sen SK (2017) Predicting and diagnosing of heart disease using
Diabetes Sci Technol 12(2):303–310 machine learning algorithms. Int J Eng Comput Sci. https://​doi.​
178. Roy K, Banik D, Bhattacharjee D, Nasipuri M (2019) Patch- org/​10.​18535/​ijecs/​v6i6.​14
based system for classification of breast histology images using 195. Sethy PK, Behera SK (2022) Automatic classification with
deep learning. Comput Med Imaging Graphics 71:90–103 concatenation of deep and handcrafted features of histological
179. Roy S, Sarkar D, Malakar S, Sarkar R (2021) Offline signature images for breast carcinoma diagnosis. Multimed Tools Appl.
verification system: a graph neural network based approach. https://​doi.​org/​10.​1007/​s11042-​021-​11756-5
J Ambient Intell Hum Comput. https:// ​ d oi. ​ o rg/ ​ 1 0. ​ 1 007/​ 196. Shafique S, Tehsin S (2018) Acute lymphoblastic leukemia
s12652-​021-​03592-0 detection and classification of its subtypes using pretrained
180. Safiyari A, Javidan R (2017) Predicting lung cancer survivability deep convolutional neural networks. Technol Cancer Res Treat
using ensemble learning methods. In: 2017 Intelligent Systems 17:153303381880278. https://​d oi.​o rg/​1 0.​1 177/​1 5330​3 3818​
Conference (IntelliSys), IEEE, pp 684–688. https://​doi.​org/​10.​ 802789
1109/​Intel​liSys.​2017.​83243​68 197. Shafique S, Tehsin S, Anas S, Masud F (2019) Computer-assisted
181. Sahlol AT, Kollmannsberger P, Ewees AA (2020) Efficient clas- acute lymphoblastic leukemia detection and diagnosis. In: 2019
sification of white blood cell leukemia with improved swarm 2nd International Conference on Communication, Computing
optimization of deep features. Sci Rep 10(1):1–11 and Digital systems (C-CODE), IEEE, pp 184–189. https://​doi.​
182. Saji SA, Balachandran K (2015) Performance analysis of training org/​10.​1109/c-​code.​2019.​86809​72
algorithms of multilayer perceptrons in diabetes prediction. In: 198. Shakeel PM, Burhanuddin MA, Desa MI (2019) Lung cancer
2015 International Conference on Advances in Computer Engi- detection from ct image using improved profuse clustering and
neering and Applications, IEEE, pp 201–206. https://​doi.​org/​10.​ deep learning instantaneously trained neural networks. Measure-
1109/​icacea.​2015.​71646​95 ment 145:702–712
183. Salaken SM, Khosravi A, Khatami A, Nahavandi S, Hosen MA 199. Shallu Mehra R (2018) Breast cancer histology images classifi-
(2017) Lung cancer classification using deep learned features on cation: training from scratch or transfer learning? ICT Express
low population dataset. In: 2017 IEEE 30th Canadian Conference 4(4):247–254. https://​doi.​org/​10.​1016/j.​icte.​2018.​10.​007
on Electrical and Computer Engineering (CCECE), IEEE, pp 200. Shao YE, Hou CD, Chiu CC (2014) Hybrid intelligent mod-
1–5. https://​doi.​org/​10.​1109/​CCECE.​2017.​79467​00 eling schemes for heart disease classification. Appl Soft Comput
184. Samko O, Marshall AD, Rosin PL (2006) Selection of the opti- 14:47–52. https://​doi.​org/​10.​1016/j.​asoc.​2013.​09.​020
mal parameter value for the isomap algorithm. Pattern Recogn 201. Sharma A, Rani R (2021) A systematic review of applications
Lett 27(9):968–979 of machine learning in cancer prediction and diagnosis. Arch
185. Sánchez-Cauce R, Pérez-Martín J, Luque M (2021) Multi-input Comput Methods Eng 28(7):4875–4896
convolutional neural network for breast cancer detection using 202. Shaw SS, Ahmed S, Malakar S, Garcia-Hernandez L, Abraham
thermal images and clinical data. Comput Methods Programs A, Sarkar R (2021) Hybridization of ring theory-based evolu-
Biomed 204:106045 tionary algorithm and particle swarm optimization to solve class
186. Sandhiya S, Palani U (2020) An effective disease predic- imbalance problem. Complex Intell Syst 7(4):2069–2091
tion system using incremental feature selection and temporal 203. Shaw SS, Ahmed S, Malakar S, Sarkar R (2021b) An ensemble
convolutional neural network. J Ambient Intell Hum Comput approach for handling class imbalanced disease datasets. In: Pro-
11(11):5547–5560 ceedings of International Conference on Machine Intelligence
187. Santhakumar D, Logeswari S (2020) Efficient attribute selection and Data Science Applications, Springer, pp 345–355
technique for leukaemia prediction using microarray gene data. 204. Shen L, Chen H, Yu Z, Kang W, Zhang B, Li H, Yang B, Liu D
Soft Comput. https://​doi.​org/​10.​1007/​s00500-​020-​04793-z (2016) Evolving support vector machines using fruit fly optimi-
188. Santhakumar D, Logeswari S (2021) Hybrid ant lion mutated ant zation for medical data classification. Knowl-Based Syst 96:61–
colony optimizer technique for leukemia prediction using micro- 75. https://​doi.​org/​10.​1016/j.​knosys.​2016.​01.​002
array gene data. J Ambient Intell Hum Comput 12(2):2965–2973 205. Simonyan K, Zisserman A (2014) Very deep convolutional net-
189. Sanyal R, Kar D, Sarkar R (2021) Carcinoma type classification works for large-scale image recognition. arXiv preprint arXiv:​
from high-resolution breast microscopy images using a hybrid 1409.​1556
ensemble of deep convolutional features and gradient boosting 206. Sisodia D, Sisodia DS (2018) Prediction of diabetes using clas-
trees classifiers. IEEE/ACM Trans Comput Biol Bioinform. sification algorithms. Procedia Comput Sci 132:1578–1585.
https://​doi.​org/​10.​1109/​TCBB.​2021.​30710​22 https://​doi.​org/​10.​1016/j.​procs.​2018.​05.​122
190. Sedighi Z, Ebrahimpour-Komleh H, Mousavirad SJ (2015) Fea- 207. Sisodia DS, Verma A (2017) Prediction performance of indi-
ture selection effects on kidney desease analysis. In: 2015 Inter- vidual and ensemble learners for chronic kidney disease. In: 2017
national Congress on Technology, Communication and Knowl- International Conference on Inventive Computing and Informat-
edge (ICTCK), IEEE, pp 455–459. https://d​ oi.o​ rg/1​ 0.1​ 109/i​ ctck.​ ics (ICICI), IEEE, pp 1027–1031. https://​doi.​org/​10.​1109/​icici.​
2015.​75827​12 2017.​83652​95
191. Sefidian AM, Daneshpour N (2019) Missing value imputation 208. Sonawane JS, Patil D (2014) Prediction of heart disease using
using a novel grey based fuzzy c-means, mutual information multilayer perceptron neural network. In: International Confer-
based feature selection, and regression model. Expert Syst Appl ence on Information Communication and Embedded Systems
115:68–94 (ICICES2014), IEEE, pp 1–6. https://​doi.​org/​10.​1109/​ICICES.​
192. Sekh AA, Opstad IS, Birgisdottir AB, Myrmel T, Ahluwalia 2014.​70338​60
BS, Agarwal K, Prasad DK (2020) Learning nanoscale motion

13
Computer Based Diagnosis of Some Chronic Diseases: A Medical Journey of the Last Two Decades 5567

209. Sridhar K, Baskar S, Shakeel PM, Dhulipala VS (2019) Devel- 10th International Conference on Cloud Computing, Data Sci-
oping brain abnormality recognize system using multi-objective ence & Engineering (Confluence), IEEE, pp 346–350. https://​
pattern producing neural network. J Ambient Intell Humaniz doi.​org/​10.​1109/​confl​uence​47617.​2020.​90581​78
Comput 10(8):3287–3295 224. Wei L, Su R, Wang B, Li X, Zou Q, Gao X (2019) Integration of
210. Srivenkatesh M (2019) Performance evolution of different deep feature representations and handcrafted features to improve
machine learning algorithms for prediction of liver disease. the prediction of n6-methyladenosine sites. Neurocomputing
International Journal of Innovative Technology and Exploring 324:3–9
Engineering 9(2), 1115–1122, https://​doi.​org/​10.​35940/​ijitee.​ 225. Wibawa HA, Malik I, Bahtiar N (2018) Evaluation of kernel-
l3619.​129219 based extreme learning machine performance for prediction of
211. Stöger K, Schneeberger D, Holzinger A (2021) Medical artifi- chronic kidney disease. In: 2018 2nd International Conference
cial intelligence: the European legal perspective. Commun ACM on Informatics and Computational Sciences (ICICoS), IEEE, pp
64(11):34–36 1–4. https://​doi.​org/​10.​1109/​icicos.​2018.​86217​62
212. Sudharshan P, Petitjean C, Spanhol F, Oliveira LE, Heutte L, 226. Wibawa MS, Maysanjaya IMD, Putra IMAW (2017) Boosted
Honeine P (2019) Multiple instance learning for histopathologi- classifier and features selection for enhancing chronic kidney
cal breast cancer image classification. Expert Syst Appl 117:103– disease diagnose. In: 2017 5th International Conference on Cyber
111. https://​doi.​org/​10.​1016/j.​eswa.​2018.​09.​049 and IT Service Management (CITSM), IEEE, pp 1–6. https://d​ oi.​
213. Sumi TA, Hossain MS, Andersson K (2022) Automated Acute org/​10.​1109/​citsm.​2017.​80892​45
Lymphocytic Leukemia (ALL) Detection Using Microscopic 227. Wu H, Yang S, Huang Z, He J, Wang X (2018) Type 2 diabetes
Images: An Efficient CAD Approach. In: Proceedings of Trends mellitus prediction model based on data mining. Inform Med
in Electronics and Health Informatics, Springer, pp 363–376 Unlocked 10:100–107. https://​doi.​org/​10.​1016/j.​imu.​2017.​12.​
214. Szegedy C, Liu W, Jia Y, Sermanet P, Reed S, Anguelov D, Erhan 006
D, Vanhoucke V, Rabinovich A (2015) Going deeper with con- 228. Yan R, Ren F, Wang Z, Wang L, Zhang T, Liu Y, Rao X, Zheng
volutions. In: Proceedings of the IEEE Conference on Computer C, Zhang F (2020) Breast cancer histopathological image classi-
Vision and Pattern Recognition, pp 1–9 fication using a hybrid deep neural network. Methods 173:52–60.
215. Szegedy C, Vanhoucke V, Ioffe S, Shlens J, Wojna Z (2016) https://​doi.​org/​10.​1016/j.​ymeth.​2019.​06.​014
Rethinking the inception architecture for computer vision. In: 229. Yang D, Xu Z, Li W, Myronenko A, Roth HR, Harmon S, Xu
Proceedings of the IEEE Conference on Computer Vision and S, Turkbey B, Turkbey E, Wang X et al (2021) Federated semi-
Pattern Recognition, pp 2818–2826 supervised learning for COVID region segmentation in chest CT
216. Taher F, Werghi N, Al-Ahmad H (2015) Rule based classification using multi-national data from China, Italy. Japan. Med Image
of sputum images for early lung cancer detection. In: 2015 IEEE Anal 70:101992
International Conference on Electronics, Circuits, and Systems 230. Yang Z, Ran L, Zhang S, Xia Y, Zhang Y (2019) EMS-net:
(ICECS), IEEE, pp 29–32. https://​doi.​org/​10.​1109/​icecs.​2015.​ ensemble of multiscale convolutional neural networks for clas-
74402​41 sification of breast cancer histology images. Neurocomputing
217. Tajbakhsh N, Jeyaseelan L, Li Q, Chiang JN, Wu Z, Ding X 366:46–53. https://​doi.​org/​10.​1016/j.​neucom.​2019.​07.​080
(2020) Embracing imperfect datasets: a review of deep learn- 231. Zhang Q, Kong X (2020) Design of automatic lung nodule detec-
ing solutions for medical image segmentation. Med Image Anal tion system based on multi-scene deep learning framework. IEEE
63:101693 Access 8:90380–90389
218. Tekade R, Rajeswari K (2018) Lung cancer detection and clas- 232. Sq Zhang (2009) Enhanced supervised locally linear embedding.
sification using deep learning. In: 2018 Fourth International Con- Pattern Recogn Lett 30(13):1208–1218
ference on Computing Communication Control and Automation 233. Zhang Y, Zhou B, Cai X, Guo W, Ding X, Yuan X (2021) Miss-
(ICCUBEA), IEEE, pp 1–5. https://​doi.​org/​10.​1109/​iccub​ea.​ ing value imputation in multivariate time series with end-to-end
2018.​86973​52 generative adversarial networks. Inf Sci 551:67–82
219. Ting KM, Zheng Z (2003) A study of adaboost with Naive 234. Zheng B, Yoon SW, Lam SS (2014) Breast cancer diagnosis
Bayesian classifiers: weakness and improvement. Comput Intell based on feature extraction using a hybrid of k-means and sup-
19(2):186–200 port vector machine algorithms. Expert Syst Appl 41(4):1476–
220. Tiwari M, Chakrabarti P, Chakrabarti T (2018) Performance 1482. https://​doi.​org/​10.​1016/j.​eswa.​2013.​08.​044
analysis and error evaluation towards the liver cancer diagnosis 235. Zhou T, Ruan S, Canu S (2019) A review: Deep learning for
using lazy classifiers for ILPD. In: Soft Computing Systems, medical image segmentation using multi-modality fusion. Array
Springer Singapore, pp 161–168. https://​doi.​org/​10.​1007/​978-​ 3:100004
981-​13-​1936-5_​19 236. Zhou Y, Lu Y, Pei Z (2021) Accurate diagnosis of early lung
221. Vaishali R, Sasikala R, Ramasubbareddy S, Remya S, Nalluri cancer based on the convolutional neural network model of the
S (2017) Genetic algorithm based feature selection and MOE embedded medical system. Microprocess Microsyst 81:103754
fuzzy classification algorithm on pima indians diabetes dataset. 237. Zhu C, Idemudia CU, Feng W (2019) Improved logistic regres-
In: 2017 International Conference on Computing Networking sion model for diabetes prediction by integrating PCA and
and Informatics (ICCNI), IEEE, pp 1–5. https://​doi.​org/​10.​1109/​ k-means techniques. Inform Med Unlocked 17:100179. https://​
iccni.​2017.​81238​15 doi.​org/​10.​1016/j.​imu.​2019.​100179
222. Vas M, Dessai A (2017) Lung cancer detection system using
lung CT image processing. In: 2017 International Conference Publisher's Note Springer Nature remains neutral with regard to
on Computing, Communication, Control and Automation jurisdictional claims in published maps and institutional affiliations.
(ICCUBEA), IEEE, pp 1–5. https://​doi.​org/​10.​1109/​iccub​ea.​
2017.​84638​51
223. Vashisth S, Dhall I, Saraswat S (2020) Chronic kidney disease
(CKD) diagnosis using multi-layer perceptron classifier. In: 2020

13