International Journal of Engineering & Technology, 7 (2.

21) (2018) 440-442

International Journal of Engineering & Technology

Website: www.sciencepubco.com/index.php/IJET

Research paper

Analysis of diabetic retinopathy using naive

bayes classifier technique
V. Sudha1*, C. Karthikeyan2
1Research
Scholar, Anna University Chennai.
2Professor,
K.S.R College of Engineering, Thiruchengode.
*Corresponding author E-mail: [email protected]

Abstract

The vital issues of diabetes are Diabetic retinopathy (DR) and Retinal Vascular Disease which leads to the blindness. The DR disease
may be detected by the early regular screening. and the automatic detection of this disease is a great solution and which is more reliable
to identify the normality level in Fundus images (FI). The FI contains the texture discrimination capacity to differentiate the healthy
images. The Data mining technique are used for identifying the retinal features of DR disease. The Data mining technique contains two
stages. In first stage the features of DR disease extract from the Retinal Images (RI). The highlights for DR disease determination
incorporate blood vessels, optic nerve, neural tissue, neuroretinal edge, optic plate size, thickness and change and which are removed by
applying Data mining strategy. The result of the different information mining arrangement systems was looked at utilizing quick
excavator apparatus. Gullible bayes and Support Vector Machine classifiers are utilized to anticipate the early discovery of eye disease
diabetic retinopathy and observed that Naive bayes technique to be enhance the exactness of 89% precise.

Keywords: Diabetic Retinopathy (DR), Support Vector Machine, Fundus Images (FI), Naive bayes, Retinal Images(RI).

1. Introduction Diabetic retinopathy arises when excess glucose in the

bloodstream resulting from diabetes mellitus causes damage to the
blood vessels of the retina. Among US adults between the ages of
Diabetic retinopathy (DR) is a genuine eye disease that happens
20 and 74 years, diabetic retinopathy is the leading cause of
because of diabetes mellitus and it has developed as the most
blindness.Diabetes influences an expected 29.1 million individuals
widely recognized reason for visual deficiency in the present
in the United States. In a Centers for Disease Control and
world. In light of most recent reports by 2030 there is a pandemic
Prevention (CDC) assessment performed between 2005 and 2008,
ascent of 4.4% in the worldwide pervasiveness of diabetes.
4.2 million or 28.5% of people with diabetes aged 40 years or
Patient's sight can be influenced by diabetes which causes
older in that time period had diabetic retinopathy
waterfalls, glaucoma, and in particular, harm to blood vessels
inside the eye, a condition known as "diabetic retinopathy".
Powerful medications for DR are accessible however it requires 2. Related work
early analysis and the constant checking of diabetic patients.
Conclusion of DR is performed by the assessment of retinal In Classification method Variables with 50% or more of their
(fundus) pictures. Manual reviewing of these pictures to decide values missing were not included in the datasets used for machine
the seriousness of DR is somewhat moderate and asset requesting. learning. Missing data for the remaining variables (less than 50%
It happens when diabetes harms the little blood vessels inside the of values missing) were handled by using imputation techniques.
retina, the light touchy tissue at the back of the eye. This minor For the datasets analyzed, we performed feature subset selection.
blood vessel will spill blood and liquid on the retina shapes Since we used standard (single) classifiers in our previously
highlights, for example, miniaturized scale aneurysms, published study, for this study, we use ensembles, which combine
hemorrhages, hard exudates, cotton fleece spots or venous classifiers and may perform better than a single classifier
loops.[1] approach. We learned ensemble classifiers based on decision tree
learners designed to handle class imbalances such as
RUSBoost,30 which utilizes majority class under sampling. For
contrast, we also learned ensemble classifiers using
AdaBoost.M1,31 which uses adaptive boosting to combine the
weighted output of several weak learners to produce a boosted
classification output. On its own, AdaBoost.M1 has no special
accommodation for class imbalances. The ensemble classifiers
were learned on the full feature set as well as the feature subsets
obtained. We reserved 20% of each dataset for testing and then
performed 10-fold cross validation on the remaining 80% of the
dataset, selecting the best classifier from the cross-validation
Fig. 1.1. Diabetic retinopathy process for use on the reserved test set. For each classifier, we
measured sensitivity or the true positive rate (the total number of
Copyright © 2018 Authors. This is an open access article distributed under the Creative Commons Attribution License, which permits
unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
International Journal of Engineering & Technology 441

cases delegated having diabetic retinopathy partitioned by the class earlier probabilities might be evaluated by P(Ci) =
aggregate number of cases really including retinopathy), si/s, where si is the quantity of preparing tests of class Ci,
specificity or the genuine negative rate, the AUC, which speaks to and s is the aggregate number of preparing tests.
the exchange off between the genuine positive rate/affectability
and the false positive rate or specificity, and precision (the Cross approval
aggregate number of effectively characterized cases separated by Cross-Validation (CV) is the standard information digging
the aggregate number of cases). [2] strategy for assessing execution of order method. Chiefly
it's utilized to assess the blunder rate of a learning
3. Algorithm for diabetic retinopathy procedure. In CV a dataset is parceled in n folds, where
each is utilized for testing and the rest of utilized for
preparing. The method of testing and preparing is
Portrayal of dataset rehashed n times so each segment of crease is utilized
The dataset was gotten at the Eye Clinic of the Sakarya University once to test.
Educational and Research Hospital. The dataset comprises of 385 In a stratified 10-overlap Cross-Validation the information
records. In dataset each record comprises of 9 highlights. is isolated haphazardly into 10 sections in which the class
These are, in particular, Glycated Hemoglobin (HbA1C), is spoken to in roughly an indistinguishable extents from
Hemoglobin (HGB), URE, High-Density Lipoprotein (HDL), in full dataset. Each part is held out thus and the learning
Low-Density Lipoprotein (LDL), Diabetes Duration, Triglyceride, plan prepared on the staying nine-tenths; at that point its
Creatine and Glucose. Since Naïve Bayes calculation does not mistake rate is ascertained on the holdout set. The learning
allow consistent information compose, every one of the qualities methodology is executed an aggregate of 10 times on
in the dataset are dealt with as straight out. In Table 1, the finding various preparing sets, lastly the 10 blunder rates are
segments demonstrate the all-out qualities for the relating found the middle value of to yield a general mistake
highlights. The conclusion segment is recognized as unsurprising evaluate.
element with esteem "1" for patients with diabetic retinopathy and Perplexity grid
esteem "0" for patients with non-diabetic retinopathy. All the Perplexity grid is a representation device which is
unmitigated highlights in the dataset were chosen by specialists generally used to introduce the precision of the classifiers
and assessment was made in view of these highlights. in arrangement (Han and Kamber, 2006). It is utilized to
demonstrate the connections amongst results and
Table 1: Demonstrates the Clinical Feature of the Patients in the Dataset anticipated classes.

The passages in disarray grid have the accompanying

implications with regards to our investigation:

• a is the quantity of right expectations that an example

is negative,

• b is the quantity of mistaken forecasts that an example

is certain,

Gullible bayes • c is the quantity of mistaken forecasts that an example

is negative,
The Bayesian Classification speaks to a directed learning and
additionally a factual technique for characterization. Accept a • d is the quantity of right expectations that an example
basic probabilistic model and it enables us to catch vulnerability is certain
about the model principledly by deciding probabilities of the Table 2: Confusion Matrix
results. It can take care of symptomatic and prescient issues. Predicted
Innocent Bayes calculation depends on Bayesian Theorem. Negative Positive
1. Each information test is spoken to by a n dimensional
component vector, X = (X1, X2,… , Xn), portraying estimations Negative a b
Actual
made on the example from n properties, individually A1, A2, Positive c d
An.
2. Suppose that there are m classes, C1, C2,… , Cm.
Given an obscure information test, X (i.e., having no class
name), the classifier will foresee that X has a place with the For Example
class having the most noteworthy back likelihood, adapted if
and just if:

P(Ci|X) > P(Cj|X) for all i<= j <= m and j != I

In this way we amplify P(Ci|X). The class Ci for which

P(Ci|X) is amplified is known as the most extreme
posteriori theory. By Bayes hypothesis,

P(Ci|X) = (P(X|Ci)P(Ci))/P(X)

3. As P(X) is consistent for all classes, just P(X|Ci)P(Ci)

should be amplified. In the event that the class earlier
probabilities are not known, at that point it is usually
expected that the classes are similarly likely, i.e. P(C1) =
P(C2) = … = P(Cm), and we would in this way boost
P(X|Ci). Else, we augment P(X|Ci)P(Ci). Note that the
442 International Journal of Engineering & Technology

4. Performance and evaluation [6] Rocha A, Carvalho T, Jelinek HF, Goldenstein S & Wainer J,
“Points of Interest and Visual Dictionaries for Automatic Retinal
Lesion Detection”, IEEE Transactions on Biomedical Engineering,
To acquire and assess the test consequences of Naïve Bayes Vol.59, (2012), pp.2244 – 2253.
classifiers, 10-crease cross approval strategy was utilized. [7] Giancardo L, Meriaudeau F, Karnowski T, Tobin K & Grisan E,
Consequently the dataset is haphazardly isolated into preparing set “Textureless macula swelling detection with multiple retinal
and testing set 10 times. Table 3, demonstrates the detail fundus images”, IEEE Trans Biomed Eng., Vol.58, (2011),
consequences of 10-overlap cross approval. Number of lines pp.795-799
[8] Hayrettin E & Menduh Ç, “Prediction and Diagnosis of Diabetic
section speaks to testing set of each overlay though Sum of Retinopathy using Data Mining Technique”, Journal of science
Diagonal segment speaks to the aggregate number of forecasts that and technology Vol. 4, No.3.
were right. As we said before the exactness is figured utilizing [9] Agurto C, Murray V, Barriga E, Murillo S & Pattichis M,
condition (1). “Multiscale Am-Fm methods for diabetic retinopathy lesion
The aftereffect of the precision that is acquired is great utilizing detection”, IEEE Transactions on Medical Imaging, Vol.29,
Naïve Bayes calculation in the genuine dataset. The precision rate (2010), pp.502-512.
is 89%. [8] [10] Kumar A, Gaur AK & Srivastava M, “A Segment based Technique
for detecting Exudate from Retinal Fundus image”, Procedia
Table 3: Detail Results of 10 Fold Cross Validation Technology, Vol.6, (2012), pp.1-9.
[11] Sopharak A, Dailey MN, Uyyanonvara B, Barman S & Williamson
Fold Sum of Diagonal Number of rows Accuracy
T, “Machine learning approach to automatic Exudates detection in
1 34 39 87.18%
retinal images from diabetics patients”, Journal of Modern Optics.,
2 31 39 89.49% Vol.57, (2008), pp.124-135.
3 36 39 93.31% [12] Yazid H, Arof H & Isa HM, “Exudates segmentation using inverse
4 35 39 90.74% surface adaptive thresholding”, Measurement, Vol.45, (2012),
5 33 39 84.62% pp.1599-1608.
6 33 39 84.62% [13] Osare A, Shadgar B & Markham R, “A Computational
7 38 39 97.44% Intelligence-Based Approach for De- tection of Exudates in
8 38 39 98.44% Diabetic Retinopathy Images”, IEEE Transactions on Information
9 35 39 89.74% Technology in Biomedicine, Vol.13, (2009), pp.535-545.
10 31 35 88.57% [14] Kumari CJ & Maruthi R, “Detection of Hard Exudates in Color
Accuracy: 89% Fundus Images of the Human Retina”, Procedia Engineering,
Vol.30, (2012), pp.297-302.
[15] Saleh MD & Eswaran C, “An automated decision-support system
5. Conclusion for non-proliferative diabetic retinopathy disease based on MAs
and Has Detection”, Computer Methods and Programs in
This investigation obviously demonstrates that the outcomes are Biomedicine, Vol.108, (2012), pp.186-196.
[16] Lazar I & Hajdu A, “Retinal Microaneurysm Detection Through
promising for the use of the information mining methods into
Local Rotating Cross-Section Profile Analysis”, IEEE
forecasts of issue in therapeutic databases. In this paper, a choice Transactions On Medical Imaging, Vol. 32, (2013), pp.400-407.
emotionally supportive network was intended for diabetic [17] Antal B & Hajdu A, “An Ensemble Based System for
retinopathy. The framework can be filled in as preparing Microaneurysm Detection and Diabetic Retinopathy Grading”,
instrument for medicinal understudies. Likewise, it will help IEEE Transactions On Biomedical Engineering, Vol.59, (2012),
hand for specialists. The framework can be additionally pp.1720-1726.
improved and extended; it can consolidate other restorative [18] Deepak KS & Sivaswamy J, “Automatic Assessment of Macular
highlights other than in the Table 1, likewise it can join other Edema from Colour Retinal Images”, IEEE Transactions Medical
Imaging, Vol. 31, (2012), pp.766-776.
information mining strategies. Persistent information can be
utilized rather than simply clear cut information.

Future work
Several research fields are remained open in the field of diabetic
retinopathy management and temporal diseases. Some future
research works directly visualized by the advances of this thesis
and from the experience in working in this field are listed below.

References
[1] Gupta S & Karandikar, A.M. Diagnosis of diabetic retinopathy
using machine learning. Journal of Research and
Development, Vol.3, No.3, (2015), pp.1-6.
[2] Sopharak A, Dailey M.N, Uyyanonvara B, Barman S, Williamson
T, New K.T & Moe Y.A. Machine learning approach to automatic
exudate detection in retinal images from diabetic patients. Journal
of Modern optics, Vol.57, No.2, (2010), pp.124-135.
[3] Gandhi M & Dhanasekaran R, “Diagnosis of Diabetic Retinopathy
Using Morphological Process and SVM Classifier”, IEEE
International conference on Communication and Signal
Processing, India, (2013), pp.873-877.
[4] Li T, Meindert N, Reinhardt JM, Garvin MK & Abramoff MD,
“Splat Feature Classification with Application to Retinal
Hemorrhage Detection in Fundus Images”, IEEE Transactions on
Medical Imaging, Vol.32, (2013), pp.364-375.
[5] Ramani RG, Geetha R, Lakshmi B & Jacob SG, “Data Mining
Method of Evaluating Classifier Pre- diction Accuracy in Retinal
Data”, IEEE International Conference on Computational
Intelligence and Computing Research, (2012), pp.1-4.