Adaptive Neurofuzzy System for Tuberculosis

PDGC 2012 1569664171

Adaptive Neurofuzzy System for Tuberculosis A.Q. Ansari

Neeraj Kumar Gupta

Ekata

Department of Electrical Engg., Jamia Millia Islamia, New Delhi , India, [email protected]

Department of Electrical & Electronics Engg., Krishna Institute of Engg. & Tech., Ghaziabad, India, [email protected]

Department of Mathematics, Krishna Institute of Engg. & Tech., Ghaziabad, India [email protected]

system for tuberculosis and in the end section presents the conclusions and scope for future research.

Abstract: In this paper, a neurofuzzy system for tuberculosis (TB) is presented. This proposed work is rule-based fuzzy system which is form of intelligent technique and contain symptoms as its input variables in certain specified ranges & possible cures or referrals to doctors as its output. The adaptability of proposed work is depending upon the rule based algorithm which has decision-making ability and backpropagation learning of neurofuzzy system. Simulated results show the proposed work for automated diagnosis, which have performed by using the realistic causes of tuberculosis disease are effective.

II. NEUROFUZZY SYSTEM FOR TUBERCULOSIS Tuberculosis is a wasting disorder occurring mainly in children of two years and upwards, due to tuberculosis of the peritoneum and the mesenteric glands. This form was formerly known as tabes mesenterica. The onset is very insidious, and may extend over many months. Gradually the limbs and face become shrunken, and there are anaemia, listlessness, attacks of pyrexia, and sometimes abdominal cramps. The leading physical sign is the enlarged abdomen, which is generally tympanitic on percussion. There are three main types: (i) the ascitic, (ii) the fibro-caseous adhesive, and (iii) the loculated type, which is a combination of the first two. (i) In the ascitic variety, the patient complains of little pain; gastro-intestinal symptoms may be absent, but ascites is present. Ascites unaccompanied by anasarca in a young adult is usually due to tuberculosis . A sample of fluid may be withdrawn with a needle, and shows an excess of protein and of lymphocytes, and guinea-pig inoculations often confirm the diagnosis, (ii) In the adhesive variety, there is matting together of the peritoneum and intestines; this may be localised or generalised. Attacks of diarrhoea or constipation occur, perhaps with signs of intestinal obstruction. Pain and tenderness may be marked features, and localised thickenings and masses with a doughy feeling can be palpated. The rolled-up omentum often forms a tumour stretching across the upper abdomen below the edge of the liver. (iii) In the loculated variety, matting occurs with encysted fluid in the centre. The hectic fever so common in tuberculosis may be present, and sometimes the disease runs a more acute course with pyrexia, resembling typhoid fever, from which it can only be differentiated by the Widal reaction. (iv) Tuberculosis of the ileo-caecal group of lymph glands causes general ill-health, and may cause local pain and swelling often confused with appendicitis. In addition to the diseases just mentioned, tuberculosis may have to be distinguished from the distension of the bowels due to improper feeding, in which three is generally no pyrexia, no resistant masses, and disappearance on regulating the diet.

Keywords-Neurofuzzy System, Tuberculosis, Backpropagation

I. INTRODUCTION Intelligent decision-making is an important task in artificial intelligence. It has applications in several fields such as adaptive intelligent prediction and control systems [1], large bioinformatics data processing [2,3], mobile robots [4], adaptive speech recognition and language acquisition [5-7], adaptive temperature controller [8], visual monitoring system and multimode information processing and intelligent agent based system and adaptive agent on the web. The applications include character recognition, medical diagnosis, navigational guidance, stock market prediction, and financial analysis. Such applications can often be solved using either artificial neural networks (ANN) or fuzzy system, but complex applications may not be efficiently solvable by a single technique alone. Each technique has characteristics that make it suitable for particular applications. For example, ANN are good at recognizing patterns, classifying data, and predicting events. But they cannot explain how patterns are classified or recognized. Fuzzy systems are good at computing and explaining decisions but cannot adaptively change the rule base to allow for new environmental conditions. By integrating an ANN and a fuzzy system, the benefits of both techniques can be fully utilized. The design of hardware systems that learn and adapt to a dynamically changing environment is of research interest. In order to facilitate an easy and systematic understanding of the proposed work; section 2 spells Neurofuzzy system for tuberculosis ; section 3 presented the Backpropagation Algorithm for Neurofuzzy System, section 4 illustrates the Simulated results of adaptive neurofuzzy

978-1-4673-2924-8/12/$31.00 ©2012 IEEE 1

Tuberculosis of the mesenteric glands iis usually due to the ingestion of infected milk: it may occuur at almost any age, but is rare under two. Males are affe fected more than females if the mucous membrane of the aliimentary canal is diseased, the risk of infection is greater[9,110]. The military type with ascites may be of systemic originn. Other varieties arise by extension from tuberculous enteritis,, salpingitis, ileocaecal or mesenteric gland tuberculosis. A neurofuzzy functional block diagram m for tuberculosis is shown in fig.1.

(RNTCP) since 1992. The intern national community and Indian government authorities are needed to expand more programmes. A FOR III. BACKPROPAGATION ALGORITHM NEUROFUZZY SYSTEM S Architecture of the neurofuzzy integ grated system contains six layer as shown in figure 2. A new kind of error backpropagation algorithm to adjust the membership functions of each variable and optimize fuzzy rules is proposed.

Fig1 . Block diagram of Neurofuzzy system for tuberculosis Fig 2. Architecture of Neurofuzzzy integrated system

These are the following factor can incrreases the risk of acquiring tuberculosis. These factors include: A. Age: Tuberculosis is a wasting disorder ooccurring mainly in children of two years and upwards. B. Weakened immune system: These are folllowing diseases and medications may be affected the imm mune system: • Diabetes • End-stage kidney disease • Chemotherapy for cancer treatment • AIDS • Drugs to prevent rejection of transpllanted organs • Crohn's disease and psoriasis • Malnutrition C. International Connections: People who travel or live in countries that have high rates of tuberculosis increases the risk of acquiring same disease. D. Economic Status: The economic stattus of a person directly affects his living conditions, ssurroundings and environment. E. Alcohol intake: It has been observeed that risk of tuberculosis increases who consumee more than a specified amount of alcohol per day. Alcohol consumption distorted the immune systeem and a chance of acquiring Tuberculosis is increased.

Backpropagation (BP) learn ning to the parameter identification of neurofuzzy mo odels whose antecedent membership function is Gausssian type. Layer-wise backpropagation description is follo owed as: Layer Six: Only the error signal needs n to be computed and propagated which is calculated as:

δ i( 6 ) = − where

∂D = T (t ) − O (t ) ∂a ( 6 )

δ i( 6 )

(1)

is the error signal.

Layer Five: The update rule for a ji is:

−

∂D ∂D ∂a ( 6 ) ∂a (5 ) = − (6) ∂a ji ∂a ∂a ( 5) ∂a ji

(2)

and

From the last four decades, tuberculosis control activities have existed in India but quality of diagnossis and treatment of TB in the govt. hospital and private ssectors has been variant. World Health Organization (WHO O) recommended directly observed treatment, short-course (DO OTS) strategy for a Revised National Tuberculosis Conttrol Programme

2

∂a ( 6) =1 ∂a (5)

(3)

∂a ( 5) = x j ∑ p i(5 ) ∂a ji i

(4)

where the summation is over the number of links from

if there are multiple outputs, then the error signal becomes:

layer five for the i th node. Hence the parameter a ji is

δ i ( 4 ) = ∑k δ k

updated by: ∑

1

Layer Three: As in layer four , only the error signal need to be computed in this layer.

Find the locally optimal learning rate by determining the value of η that minimizes

∂D

equations:

∂a

, using the following

dQik =ηψik +η2vik =ηφij , da da

dRij

δ i ( 3) = −

∂D ∂a (3)

= −∑

(6)

j

where

φ ij = δ i O j ,ψ ik = δ i x k + δ k xi and vik = δ iδ k

∂a j

( 4)

∂a i

( 3)

=

η=−

∂D ∂Rij

φ ij + Σ ik

Σ ij

∂D ∂Qik

∂D ∂Qik

ψ ik

(7)

(3)

∂D ∂a ( 4 )

∂a j

− aj

⎡ (3) ⎤ ⎢∑ a i ⎥ ⎣ i =1 ⎦ c

=

−aj

(11)

( 3)

( 3)

(12)

, j=i

2

( 3)

⎡ c (3) ⎤ ⎢∑ ai ⎥ ⎦ ⎣ i =1

δ i (3) = ∑ δ j ( 4 )

2

∂a j

( 4)

∂a i

(3)

j

,j≠i

(13)

(14)

Layer Two: With the choice of the Gaussian membership function, the operation performed in this layer is using:

(8) where cij and ( 5)

where

∂a i

( 3) i

( 4)

so, we have:

Layer Four: Only the error is need to be calculated in his layer:

=−

(4)

∂a j

i

ν ik

The lowest generalization error for the symmetric phase characterized by lack of differentiation between the nodes is achieved by gradually reducing the learning rate towards zero[11-17]. However, decaying the learning rate in the symmetric phase will prevent the system from escaping the symmetric fixed point, thus resulting in sub-optimal solution.

δ i ( 4) = −

∂a j

( 4)

∂D ∂a j

∑a

and obtained value of η is

Σ ij

(10)

where the summation is performed over the consequents of a rule node.

(5)

x 0 =1.

where

(4)

∂D ∂a ∂a (5) ∂a ( 4 )

σ ij are

center and width of the Gaussian

membership function. The update rule of

cij

( 2)

is derived as in the following:

( 5)

∂a = ∑ a ji x j ∂a ( 4 ) j

(9)

−

3

∂D ∂cij

( 2)

∂D = − (3) ∂a

(2)

∂a ( 3 ) ∂a k ∑k ∂a ( 2) ∂c ( 2) k ij

(15)

•

where ∂a (3) ∂a k

( 2)

=

a ( 3) ak

•

( 2)

(16)

if termnode j is connected to rule node k

•

0

So, the update rule of

Cij

( 2)

(t + 1) = Cij

The update rule of

−

∂D ∂σ ij

(2)

(2)

C ij

(2)

is:

(t ) − η

•

∂a ( 3) ( 2) ∂Cij

(17) •

σ ij ( 2) is derived as: (2)

∂D ∂a ( 3 ) ∂ak = − (3) ∑ (2) (2) ∂a ∂σ ij k ∂a k

Adaptive Neurofuzzy system can be used identification of the disease as their performance monitor by rule based adaptive system since their adaptation mechanisms rely on an estimate for values of the disease symptoms parameters. A. Simulations results for Tuberculosis a. A Fuzzy Inference System using Sugeno Method A Sugeno model is developed in MATLAB Fuzzy tool by defining the inputs and output for the system. We have taken 5 inputs for the system which are actually the risk factors: Age, Weakened immune system, International Connections, Economic Status, Alcohol consumption.

(18)

if termnode j is connected to rule node k

(19)

0

The update rule of 2

1

σ ij ( 2 ) is :

2 2

If (age is young_ones) and (immune system is weak) and (economic_status is high) and (international_connections is no) then (output is low). If (age is adults) and (immune system is weak) and (economic_status is low) and (alcohol_consumption is low) and (international_connections is yes) then (output is high). If (age is adults) and (immune system is strong) and (economic_status is low) and (alcohol_consumption is low) and (international_connections is no) then (output is low). If (age is oldage) and (immune system is stong) and (economic_status is low) and (alcohol_consumption is low) and (international_connections is no) then (output is low). If (age is oldage) and (immune system is weak) and (economic_status is high) and (alcohol_consumption is high) and (international_connections is yes) then (output is high).

(20)

IV. SIMULATED RESULT OF ADAPTIVE NEUROFUZZY SYSTEM FOR TUBERCULOSIS Fig 3. Fuzzy Inference System for Tuberculosis

In this section, illustrates the adaptive neurofuzzy system for tuberculosis. The successful attempt has been made to neurofuzzy integrated system for coronary heart disease [18]. Our contribution to this field is a humble effort and to explore this important technology in other diseases. Now depending upon the input values to our fuzzy inference engine, define the position of the input in a fuzzy subset as defined by the membership functions. If the output shows that a person is on the danger side it can give a warning and suggest the preventive measures as does a doctor in real life.

b.

Membership Function of ‘Immune system’: The risk factors are taken as inputs to the system and membership functions for each input have been defined. The Membership Function of ‘Immune system’ is shown in fig. 4.

For tuberculosis, laid down 50 rules of the form, some of these are listed here. Fig 4. Membership Function of ‘Immune system’

4

c.

risk of tuberculosis with inputs immune system and consumption is shown in fig.8.

Rule Viewer of Tuberculosis: The fuzziness of a fuzzy membership permits us to handle the problem of disease prognosis. Combining the various research data about the diseases and have laid down linguistic fuzzy rules. The rule viewer of tuberculosis is shown in fig.5.

Fig 8 .Risk of TB with immune system and alcohol consumption Fig 5. Rule Viewer of Tuberculosis

d.

g. Neurofuzzy architecture for tuberculosis: This process produced inference chains which are represented in the structure shown in fig.9. Input membership and rule in the fig.9 represented a fuzzy rule based structure. During the design process, the number of input variables to each model was minimized in order to reduce the number of required rules. The adaptive neurofuzzy architecture for TB is shown in fig.9.

Training data: The research data about the diseases required the training of neurofuzzy system. Training yielded 70 sets of input-output data. This data was then used to tune the membership functions, using the algorithm given in section III. The training data used to train the adaptive neurofuzzy system is shown in fig 6.

Fig 6. Training data used to train the neurofuzzy system

e.

Training error: Error is only 0.0052404 after complete the training by using proposed method as shown in fig 7.

Fig 9. Neurofuzzy architecture for TB

V. CONCLUSION AND FUTURE SCOPE In this paper neurofuzzy system for tuberculosis has been proposed. Rule-based fuzzy system has contain symptoms as its input variables in certain specified ranges & possible cures or referrals to doctors as its output. To minimize the output error, a backpropagation algorithm for neurofuzzy system has been proposed. The proposed algorithm has the adaptability of a back propagation based neural network and the decision-making ability of a rule based system. The proposed work for automated diagnosis, which have performed by using the realistic causes of tuberculosis disease are effective. A good level of agreement with doctor's opinions using neurofuzzy systems has been achieved. The proposed research has wide application in many fields, like noisy speech recognitions, noisy image filtering, medical science, intelligent agents, nonlinear adaptive control and performance analysis of dynamical systems. In

Fig 7. Training error( ERROR: 0.0052404)

f. Performance Measures: The performance is measured by the level of satisfaction of doctor's opinions. A Neurofuzzy performance measure good levels of agreement with doctor’s opinions are described by the performance fuzzy sets is used in proposed system. The design of the rule base was performed by optimizing the intermediate linguistic variables. Then the input variables necessary to infer these optimizing intermediate values were identified. The

5

remote areas, this proposed research work is under consideration where there is no easy availability of a doctor. Our contribution to this field is a humble effort and to explore this important technology in other diseases.

[17] Ying Sun; Yubin Xu; Lin Ma; “The Implementation of Fuzzy RBF Neural Network on Indoor Location” Pacific-Asia Conference on Knowledge Engineering and Software Engineering KESE, 2009 , pp. 90 – 93. [18] A.Q.Ansari, N. K. Gupta, “Automated Diagnosis of Coronary Heart Disease Using Neuro-Fuzzy Integrated System” World Congress on Information and Communication Technology, Co-Organized by Machine Intelligence Research Labs (MIR Labs) USA and University of Mumbai,pp.1383-1388,2011, DOI 978-1-4673-0125-1 ©2011 IEEE

REFERENCES [1] J. S. Albus, “A new approach to manipulator control: The cerebellar model articulation controller (CMAC)”, American Society of Mechanical Engineers Trans. Journal of Dynamic Systems, Measurement, and Control vol. 97, pp. 220–227 Sept. 1975. [2] S. Amari and N. Kasabov, Brain-Like Computing and Intelligent Information Systems, New York: Springer-Verlag, 1998. [3] M. Arbib, The Handbook of Brain Theory and Neural Networks, Cambridge, MA: MIT Press, 1995. [4] T. Fukuda, Y. Komata, and T. Arakawa, “Recurrent neural networks with self-adaptive GAs for biped locomotion robot”, in Proc. Int. Conf. Neural Networks ICNN. Piscataway, NJ: IEEE Press, pp. 1710-1715, June 1997. [5] C. Furlanello, D. Giuliani, and E. Trentin, “Connectionist speaker normalization with generalized resource allocation network”, in Advances in neural information processing systems (NIPS), D. Toretzky et al., Eds.Cambridge, MA: MIT Press, pp.1704–1707, 1995. [6] S. Amari and N. Kasabov, “A framework for intelligent conscious machines utilizing fuzzy neural networks and spatial temporal maps and a case study of multilingual speech recognition”, in Brain-Like Computing and Intelligent Information Systems, Eds. New York: Springer- Verlag, pp.106–126, 1998. [7] Kasabov et al., A methodology for speech data analysis and a framework for adaptive speech recognition using fuzzy neural networks and self organizing maps, in Neuro Fuzzy Techniques for Intelligent Information Systems, N. Kasabov and R. Kozma, Eds. New York: Springer- Verlag, 1999. [8] P.P Bhogle, B.M.Patre, L.M Waghmare,V.M. Panchade, “Neuro Fuzzy Temperature Controller”, International Conference in Mechatronics and Automation pp. 3344-3348, Aug. 2007. [9] M.Sánchez ,S.Uremovich, P.Acrogliano “Mining tuberculosis data” Medical Information Science Reference, New York, 2009,pp 332–349. [10] A. Bakar,F. Febriyani “ Rough neural network model for tuberculosis patient categorization” In: Proceedings of the international conference on electrical engineering and informatics, 2007, pp 765–768. [11] D. Theodoridis, M. Christodoulou, Y. Boutalis, “Direct adaptive control of unknown multi-variable nonlinear systems with robustness analysis using a new neuro-fuzzy representation and a novel approach of parameter hopping”, 17th Mediterranean Conference on Control and Automation, 2009, pp. 558 – 563. [12] A.Q.Ansari, Neeraj Gupta, “Backpropagation Algorithm for Neurofuzzy Filter” International Journal of Computational Cognition (IJCC), vol. 9, no.3, pp. 41-47, Sept. 2011. [13] Chia-Feng Juang; Ren-Bo Huang; Yang-Yin Lin, “A Recurrent SelfEvolving Interval Type-2 Fuzzy Neural Network for Dynamic System Processing”, IEEE Transactions on Fuzzy Systems, Vol. 17 , Issue 5 , 2009, pp. 1092 – 1105. [14] Chun-Ru Dong; Xi-Zhao Wang; Xiao-Dong Dai; “ Learning the Weights of Weighted Fuzzy If-Then Rules Via Training T-S Norms Neural Network”, International Conference on Machine Learning and Cybernetics, 2006 , pp. 2920 – 2924. [15] Liu Fang; “ Designing the Self-Adaptive Fuzzy Neural Networks”, International Joint Conference on Bioinformatics, Systems Biology, and Intelligent Computing, IJCBS, 2009, pp. 537 – 540. [16] Oh Sung-Kwun, W. Pedrycz, Byoung-Jun Park, “Multilayer Hybrid Fuzzy Neural Networks: Synthesis via Technologies of Advanced Computational Intelligence”, IEEE Transactions on Circuits and Systems I, Vol. 53, Issue 3, 2006, pp. 688 – 703.

6