Genetic-Neuro-Fuzzy Inferential Model For

Tuberculosis Detection
Mrs.R.Beaulah Jeyavathana, *1, Dr.R.Balasubramanian, 2, M.Celin Supriya 3
1
Department of Computer Science and Engineering, Mepco Schlenk Engineering College, Sivakasi, India
2
Department of CSE, Manonmaniam sundaranar University, Tirunelveli, India
3
Department of Computer Science and Engineering, Mepco Schlenk Engineering College, Sivakasi, India
Nigeria and South Africa. An estimated one million children
Abstract: Tuberculosis is one of the dangerous infectious disease
became ill with TB and 170000 children died of TB
that can be characterized by the growth of tubercles in the tissues.
This disease mainly affects the lungs and also the other parts of our (excluding children with HIV).
body. The orthodox diagnosis methods available for Tuberculosis
diagnosis were been faced with a number of challenges which can, if
The World Health Organization (WHO) estimated in
measure not taken, increase the spread rate; hence, there is a need for 2006, that each year, more than 8 million new cases of TB
aid in diagnosis of the disease. This study proposes a technique for occur and approximately 3 million persons die from the
intelligent diagnosis of TB using Genetic-Neuro-Fuzzy Inferential disease and estimated that between 19% and 43% of the
method to provide a decision support platform that can assist medical world’s population will be infected with Mycobacterium
practitioner in administering accurate, timely, and cost-effective Tuberculosis. Within the past decade it has become clear that
diagnosis of Tuberculosis. The medical record of 100 TB patients the spread of HIV infection and the immigration of persons
aged 15 to 75 were used to evaluate the performance of the multi-
from areas of high incidence have resulted in increased
technique decision support system. 70% of the dataset was used
numbers of TB cases. It has always occurred
training data, 15% was used for validation while the remaining 15%
was used to observe the performance of the proposed system.
disproportionately among disadvantaged populations such as
Index Terms—Tuberculosis, Fuzzy Logic approach, Neural the homeless, malnourished, and overcrowded. Today, several
Network, Genetic Algorithm, Decision Support system. methods for the diagnosis of TB have been proposed.
Tuberculin Test, Radiological Examination, and Sputum
I.INTRODUCTION Smear Microscopy are common conventional approaches
however in the last 10 years, several molecular methods have
Tuberculosis is the infectious bacterial disease that is caused been developed for direct detection, identification and
by the organism called Mycobacterium tuberculosis and that susceptibility testing of mycobacteria.
may affect any tissues of the body but it mainly affects the
lungs. The TB is one of airborne pathogen and that can spread Orthodox methods of diagnosing TB are primarily
through air or by coughing or sneezing from one person to through physical examination and laboratory tests. The former
another. TB affects all age groups in all parts of the world. involves asking patients certain questions for prognosis
Mostly it affects young adults and also the peoples who are all purposes while tests are carried out to affirm physical
in the developing countries. Active lung TB are cough with examination. Diagnosis can be stopped if medical practitioner
sputum and blood at a time, weight loss, chest pains, fever, is totally convinced after physical examination however, this
weakness and night sweats. Tuberculosis bacteria are present is not advised. This orthodox method is currently faced with a
in sputum samples are identified under a microscope. It number of challenges such as lack of medical facilities in most
detects only half the number of Tuberculosis cases and cannot medical centers and as a result, inhibiting the management of
detect drug-resistance. TB in developing countries Orthodox methods of diagnosing
TB are primarily through physical examination and laboratory
In 2015, around 11 million people fell ill with TB tests. The former involves asking patients certain questions for
and 2 million peoples were died from the disease. Over 95% prognosis purposes while tests are carried out to affirm
of the deaths in TB occur in low and middle-income countries. physical examination. Diagnosis can be stopped if medical
Around Six countries account for 60% of the total, with India practitioner is totally convinced after physical examination
leading the count, followed by Pakistan, China, Indonesia, however, this is not advised. This orthodox method is
currently faced with a number of challenges such as lack of detect early signs of cancer and tuberculosis was also
medical facilities in most medical centers and as a result, explained in this work [3].
inhibiting the management of TB in developing countries. Wenbo Li, Yan Kang (2015) using a new adaptive
The strength of IT in providing an effective and VOI selection method. Twenty-two features were extracted to
efficient solution to real life problems has been explored to aid distinguish nodules, vascular endpoint or vascular cross
scientific discoveries and advancement of different fields of structure, designed an optimal feature combination selection
medicine. Hence, to reduce the morbidity and mortality rates frame based on improved genetic algorithm and support vector
in human as a result of TB, there is need to incorporate IT into machine. The improved GA algorithm to select the optimal
its diagnostic approach. This study, therefore, proposes a feature combination from the feature pool to establish SVM
decision support model for intelligent diagnosis of TB using classifier [4].
Genetic-Neuro-Fuzzy Inferential technique. The model is Asogbon MG (2016), proposed enhanced Neuro-
aimed at providing a decision support platform that can aid Fuzzy system based on genetic algorithm for medical
medical practitioners in administering accurate, timely, and diagnosis proposed the use of Genetic Algorithm (GA)
cost-effective diagnosis of TB in developing countries. technique to automatically evolve optimum connection
weights needed to efficiently train a built ANFIS model used
II.RELATED WORK for Typhoid fever diagnosis. The GA module computes the
best set of connection weights, stores them, and later supplies
A novel approach to identify tuberculosis bacteria them to the corresponding hidden layer nodes for training the
based on shape and colour was proposed by M.Forero et al. ANFIS [5].
(2004). Designed algorithm technique was based on combined An artificial intelligent approach for estimation of
use of invariant shape features together of bacilli with simple disease and resource utilization was discussed by
thresholding operation on chromatic channels. This E.Papageorgiou et al. (2009). Fuzzy cognitive map-based
methodology is based on segmentation followed by an tool was used to represent medical diagnosis system
identification procedure, for which 110 samples of bacilli was concentrated on relating to lungs infections. Due to easy
analyzed. Usefulness of K-means clustering algorithm graphical representation approach, the proposed method
techniques was applied to predict classification, accuracy, and makes wide use of computer consultation system. The
sensitivity versus specificity was evaluated using ROC presented system would offer a solution for requirements
analysis procedure. Further, the author suggested exploring a imposed by the target application, disease symptoms, signs
colour-based edge segmentation technique using derivative and laboratory tests [6].
operators to all chromatic channels and by using Bayesian Usefulness of fuzzy logic approach to decision
decision theory [1]. support system in medicine was discussed by U. Dev et al.
N. Walia et al. (2015) had presented a systematic (2011). This approach was based on the diagnosis of a patient
approach for design and identification of tuberculosis using suffering heart failure treated with beta blockers. The
fuzzy based decision support system. Their framework briefly developed system is a prototype warning system for clinical
explains relation between different input attributes and its problems which is based on the assumption that can be
symptoms. Author concluded that fuzzy basis dependent analyzed using simple rules. The planned technique generates
expert systems can be used during diagnosis. Further,author basic rules using fuzzy logic based on expert experience [7].
suggested that designed system can be extended for A spectrum of soft computing decision-making
construction of other chronic obstructive diseases using hybrid model to solve a real life complex problem related with
neuro systems [2]. medical science was explored by P.Srivasta et al. (2013). The
An integrated approach for automated detection of designed network was tested with ECG analysis and the
early lung cancer and tuberculosis-based X-ray image analysis satisfactory factor was measured under a domain of
was demonstrated by K. Lee (2006). Various symptoms of the considered inputs [8].
disease and finding nodules were focused during this paper. A novel Intuitionistic fuzzy cognitive map (iFCM)
The proposed technique uses watershed segmentation based on theory of Intuitionistic fuzzy sets was explained was
approach to isolate a lung X-ray image, and then apply a small M. Arts et al. (2013). This model offers checking and
scanning window to determine whether any pixel is a part of a classification techniques to predict human decision model.
disease nodule or not. Additionally, various methods used to The proposed system has an extension of FCM to the co-
evaluate degree of hesitation; experts may suffer while
defining a relation between concepts of FCM. The author evaluate the performance of the multi-technique decision
demonstrates the effectiveness of FCM with numeric support system. 70% of the dataset was used training data,
reproducible expels on a process of control and decision 15% was used for validation while the remaining 15% was
support. The simulation studies describe the performance of used to observe the performance of the proposed system.
iFCM for medical decision support platform and the results
obtained were significantly better than obtained with 4.2. DATABASE
conventional FCM model. In this work, fuzzy rules are Structured database presents quantitative data about
applied to determine the stage of tuberculosis facts and the established rules in the field of medicine
focusing on diagnosis of TB. The facts comprise of signs and
III.SYSTEM DESIGN symptoms of TB, while rules are patterns to draw deductions
based on available information. Unstructured database is
heuristic in nature and hence gathered by experience, good
Input Data practices, guesses, and judgments. The database comprises of
Patient-Bio-Data, Disease-Physical-Signs, Disease-Symptoms,
Medical-History, Physical Examination, results of diagnostic
tests and Patient Diagnosis.
Fuzzy Logic
4.3. FUZZY LOGIC
The diagnosis process harnesses the strength of fuzzy
logic component in the following operational sequence:
Neural Network Back
propagation 4.3.1. Fuzzification of input variables:
vector
Fuzzification is a process that determines the degree
of membership to the fuzzy set based on fuzzy membership
Genetic Algorithm function. The first step is to
• Create a fuzzy set of the parameters. The parameters
will be described with five linguistic variables (very
mild, mild, moderate, severe and very severe).
• The degree of membership for a fuzzy system is of
the range [0 1]. A range of the fuzzy value using the
linguistic variables will be determined by the expert.
e.g. mild 0.1 ≤ x ≤ 0.3.
4.3.2. Establishment of fuzzy rule base: The rule base for
Fig. 1, System Design TB diagnosis is characterized by a set of IF–THEN rules in
which the antecedents (IF parts) and consequents (THEN
Fig.1, shows the system design of the proposed method. This
parts) involve linguistic variables. The rules can be formulated
is an inferential technique proposed to integrate GA, NN and
with assistance of experts in the management of TB, or on
FL components to provide a self-learning and adaptive system
consultation to existing standard literature. A rule can only fire
for handling uncertain and imprecise data for diagnosis of
if any of its precedence parameters such as very mild, mild,
tuberculosis. The inference engine consists of reasoning
moderate, severe, and very severe evaluates to TRUE,
algorithm driven by the production rules based on Mamdani’s
otherwise it does not fire.
Inference Mechanism.
4.3.3. Fuzzy Inference:
IV. PROPOSED SYSTEM The inference engine controls how the rules are applied
towards facts. This is the part of rule-based expert system that
makes inferences. It decides which rules are satisfied by facts
4.1. MATERIAL AND METHODS and controls overall execution. Also, it matches the facts
In this system, dataset are collected for 100 TB against the rules to see what rules are applicable. The system
patients from the goverment hospital of tirunelveli. The will make use of forward chaining reasoning; it would make
medical record of 100 TB patients aged 15 to 75 were used to use of the facts given by the patient to diagnose the problem.
Fuzzy inference is the process of mapping from a given input An individual chromosome consists of 24 genes and
to an output using the theory of fuzzy sets. Rules are used in each gene represents the connection weight of a diagnosis
the knowledge-base by the fuzzy inference engine to derive variable in a length of 1 bit.
conclusion based on the rules. One feasible solution is to generate an initial
4.3.4 Defuzzification: population holding a set of possible solutions from random
This involves changing fuzzy output back into chromosomes. A chromosome is represented as a vector C =
numerical values for system action. The output from the (CA,…,CX) of binary decision variables Ci=0,2,3 encoded in
inference engine is translated into crisp output which is more binary representation as string consisting {0, 1} genes.
precise than the fuzzy output . A gene Ci =1 if the ith variable is included in a
solution set of a diagnostic process otherwise 0. Fitness
function is used to optimize each chromosome by evaluating
4.4 NEURAL NETWORK
the genes that constitute the chromosome using their fitness
Neural Network has the capability of capturing
value.
domain knowledge from available indicators and can readily
As evolutionary algorithm continues through its
handle both continuous and discrete data. NN is used to train
cycle, fitness value of each chromosome keeps improving till
and test the designed fuzzy system to optimize the
it reaches an optimum value when it can no longer improve.
performance of the overall system. Each diagnosis variable
Finally, it shows chromosomes of some candidates
has a weight Wi which shows its contribution in the diagnosis
and their fitness values. A number of constraints have been
process.
considered in carrying out appropriate management of disease
The raw information obtained from patients is fed
in medical diagnosis, therefore fitness evaluation of
into NN via input layer and participation of each category of
chromosome must be done with proper constraint validation.
variables is determined at a hidden layer of the network using:
n
Constraints can be termed as objectives that must be
CATi   Ai * W Ai achieved in which some render most of the solutions from the
i search space. The fitness evaluation of an individual F(i) is
CATi is ith category of variable, n is count of variables in done as:
CATi, and Ai is the ith diagnosis variable with weight WAi n
Fi  (1  Wi * Ci ( p )) 1
Result of the output layer represents an overall output of i 0
diagnosis by the NN component. The output result is given by, where n is the number of diagnosis variables, Wi is the weight
n associated with ith variable and Ci(p) is the number of
OutputNN   CATi * WCATi violations for ith constraint at solution p.
i
This fitness function has a range of [0, 1] and an
where WCATi is the connection weight of CATi
optimal solution occurs when we have 0 violations thus
n
4.5 GENETIC ALGORITHM
W
i 0
i * C i ( p ) which results in Fi=1. Chromosomes with
Actually, NN provides a structure for combining the higher fitness values are selected as parents for mating in
diagnostic parameters which could serve as a platform for the
order to produce outstanding candidates and maximize the
inference engine, but a specific issue with NN is lack of fitness function. The probability of choosing an individual for
definite way of determining the connection weights for hidden genetic operation is proportional to its fitness. This process is
layers when dealing with a particular problem. repeated until an optimal connection weight is achieved.
A number of medical diagnosis had been assisted by
neuro-fuzzy systems though such systems had been built
based on trial and errors, this increases computation cost. In V.SIMULATION BASED RESULTS AND
this study, genetic optimization is performed to choose DISCUSSIONS
optimal values from a group of diagnostic parameters which
serve as input. There are 23 diagnostic parameters in the NN
Membership function plot: The potential symptoms are
but the task is to decide which parameters are taken as input in taken as input and membership functions are defined
order to minimize complexity. according to the level of severity of the disease.
 Performance Analysis

Given a True Positive value (TP) that represents the number

of patients with tuberculosis as agreed by both model and
human expert, a True Negative value (TN) indicating the
number of patients where agreement could not be reached by
both model and human expert, and TNR as the total number of
records; the sensitivity and accuracy of this model are:

Sensitivity= TP/TNR * 100%

Accuracy= TP+TN/TNR * 100%

Rule Viewer for Pulmonary TB: On the basis of the if-then
rules defined, a set of values are obtained and processed using The sensitivity and accuracy of the proposed model are 68%
rule viewer. The Rule Viewer presents the on-sight view of and 82%.
the fuzzy inference systems process. The Rule Viewer also
depicts how the shape of certain membership functions VI. CONCLUSION
influences the final result. Each rule is a row of plots, and each
column is a variable. The system output obtained using The use of soft computing techniques in medical diagnosis
weighted average defuzzification process. All output cannot be overemphasized as they have greatly imparted the
membership functions should be of similar type whether linear processes in medical diagnosis and aided an increase in
or a constant. diagnosis accuracy. This work, demonstrated how an
aggregation of such technique can assist in the diagnosis of
TB.

VII.REFERENCES

[1] C.A. Robert, J.E. Buikstra, The Bio-Archaeology of

Tuberculosis: A Global Perspective on a Re-Emerging
Disease, University Press of Florida, Gainesville, Fl, 2009.
[2] David M. Morens, Gregory K. Folkers, Anthony S. Fauci,
The challenge of emerging and re-emerging infectious
diseases, Nature 430 (2004) 242–249.
[3] World Health Organization, WHO Country Cooperation
Strategy: Federal Republic, 2002.
Table 1 Parameters used in finding optimal solution [4] World Health Organization, Global Tuberculosis Control,
S.No Parameters Value WHO Report 2008, Geneva, 2008.
1 Number of Generations 20 [5] P. Sudre, G.T. Dam, A. Kochi, Tuberculosis: a global
2 Number of individuals 40 overview of the situation today, Bull. World Health Organ. 70
3 Crossover probability 0.55 (1992) 149– 159.
4 Mutation probability 0.35 [6] American Thoracic Society, Diagnostic standards and
classification of tuberculosis in adults and children, Am. J.
In this simulation, the GA used selected values given in Respir. Crit. Care. Med. 161 (2000) 1376–1395.
Table1 to find optimal solution and once a complete set of [7] Syed Tasleem, Gulraiz Ali, Farzana Mahdi SikandarHasan,
optimal value for the parameters is reached, the GA processes Molecular diagnosis of tuberculosis: a new primer design,
stops. High mutation probability used in the parameter settings Iran. J. Clin. Infect. Dis. 4 (2) (2009) 105–108.
is a brain behind bringing new individuals at each stage hence, [8] Young S. Kim, Donald E. Shelton, Gregg Barak, studying
to avoid a scenario whereby best combination will only be juror expectations for scientific evidence: a new model for
from initial individuals.
looking at the CSI myth, J. Am. Judges Assoc. 47 (1–2)
(2011) 8–18.
[9] S.A. Oke, A literature review on artificial intelligence, Int.
J. Inf. Manage. Sci. 19 (4) (2008) 535–570.
[10] V.P. Masnikosa, The fundamental problem of an artificial
intelligence realisation, Cybernetes 27 (1) (1998).
[11] W.E. Halal, Artificial intelligence is almost here, On the
Horizon, The Strategic Planning Resource for Education
Professionals, vol. 11(2), 2003.
[12] Y. Peng, X. Zhang, Integrative data mining in systems
biology: from text to network mining, Artif. Intell. Med. 41
(2) (2007) 83– 86.
[13] X. Zhou, B. Liu, Z. Wu, Y. Feng, Integrative mining of
traditional chinese medicine literature and medline for