
Research Article


Diagnosis of Heart Disease using Fuzzy Resolution Mechanism

A.V. Senthil Kumar


ABSTRACT

The aim of this study is to combine the neural networks (ANNs) and Fuzzy Logic
(FL) to make a powerful tool to diagnosis heart disease. By combining the Fuzzy
inference system and neural network, the input values are passed through the
input layer (by input membership function) and the output could be seen in output
layer (by output membership functions). Training involves iterative adjustment
of parameters of the adaptive neurofuzzy inference system using a hybrid learning
procedure to diagnosis the heart disease. This mechanism presents five layer,
each layer has its own nodes. Layer 1 had the input variables with membership
function. Tnorm operator that perform the AND operator can be used in layer
2. The sum of all rules firing strengths are assigned in layer 3. The nodes
in layer 4 are adaptive and perform the consequent of the rules. Single node
computes the overall output in layer 5. The proposed method is tested with Cleveland
heart disease dataset. The ANFIS approach is implemented using MATLAB. The proposed
mechanism can work more effectively for diagnosis of heart disease and also
improves the accuracy. The result of the proposed methods is compared with earlier
method using accuracy as metrics. 




Received:
August 04, 2011; Accepted: October 18, 2011;
Published: January 05, 2012 

INTRODUCTION Adaptive neuro fuzzy inference similar to that of a neural network which maps inputs through input membership functions and associated parameters and then through output membership functions and associated parameters to outputs, can be used to interpret the input/output result for diagnosis of heart disease. The learning process can be refined with the parameters associated with the membership functions.
Roan et al. (1993) used conventional crisp sets,
the concepts of fuzzy sets provides more robust representations of the model
of realworld objects. Jang (1993) employed a technique
called adaptive neurofuzzy inference system (ANFIS). It employs a NN approach
to the design of a fuzzy inference system. Kosko (1994)
used Learning and adaptation of the Neural Network makes this fuzzy system more
systematic and less reliant on knowledge of experts. It has been shown that
under proper conditions, ANFIS can be used as a universal approximator. Serpen
et al. (1997) developed probabilistic potential function neural network
algorithm. Haykin (1999), the application of artificial
intelligence approaches such as Neural Network (NN) and Fuzzy Logic (FL) do
not require an explicit mathematical model and are suitable for nonlinear physiological
systems. Moreover, they offer several advantages such as nonlinear inputoutput
mapping, adaptivity and fault tolerance. One of the most commonly used classifier
techniques is artificial neural networks. The reason for being commonly used
is to present some properties such as learning from examples and exhibiting
some capability for generalization beyond the training data (Mukhopadhyay
et al., 2002). A Lot of research has also been done about Cleveland
heart disease database. A new learning model called granular support vector
machines for data classification problems. The accuracy rates were 83.04 and
84.04% for SVM and GSVM, respectively (Tang et al.,
2004). Kahramanli and Allahverdi (2008) developed
a hybrid system for diabetes and heart diseases using artificial neural network
and fuzzy neural network. Liu et al. (2010) derived
new ANFIS for parameter prediction with numeric and categorical inputs. Forouzanfar
et al. (2010) oscillometric estimated blood pressure using adaptive
neuro fuzzy inference system.
Heart disease is a large health problem. It is the leading cause of death for
both men and women. Each year more than 500,000 people die of heart attacks
caused by chronic heart failure. The increasing number of death due to heart
disease worldwide has drawn the attention to adopt a technology such as Adaptive
NeuroFuzzy Inference System for research. Experimental results indicate that
the proposed fuzzy expert system can work more effectively than other methods
can (Tang et al., 2004; Kahramanli
and Allahverdi, 2008; Senthil Kumar, 2011).
DIAGNOSIS TOOL FOR HEART DISEASE
Cleveland heart disease dataset: This database has one of the highest
known heart disease data. The experimental Cleveland dataset is retrieved from
the Internet (http://archive.ics.uci.edu/ml/)
and it contains the collected personal data (Fig. 1). Table
1 lists the attributes of Cleveland dataset.
Fuzzification: The conversion from crisp to fuzzy input is known as
fuzzification (Fasanghari and Montazer, 2010). If the
form of uncertainty happens to arise because of imprecision, ambiguity or vagueness,
the variable is probably fuzzy and can be represented by a membership function.
Architecture of the fuzzy resolution mechanism for heart disease: The
fuzzy resolution mechanism can take fuzzy inputs but the output produced are
always a fuzzy sets. With the crisp inputs and outputs, fuzzy resolution mechanism
implements mapping from its input variable to output variable through a number
of fuzzy if then rules.
Table 1: 
Attributes of cleveland dataset 


Fig. 1: 
Architecture of the diagnosis of heart disease 
Cleveland dataset is taken with fourteen attributes, i.e., age, sex, cp, trestbps,
chol, fbs, restecg, thalach, exang, oldpeak, slope, ca, thal are selected as
the input fuzzy variables and num as output fuzzy variable are adopted for fuzzy
resolution mechanism. For fuzzy resolution mechanism the ANFIS approach learns
the rules and membership functions from Cleveland dataset.
The neuroadaptive learning techniques provide a method for the fuzzy modeling procedure to learn about heart disease from cleveland data set, in order to compute the membership function parameters. An adaptive network is network of nodes and directional links. Associated with the network is a learning rulehybrid method. Networks are learning a relationship between inputs and outputs. The architecture of the fuzzy resolution mechanism using ANFIS is shown in Fig. 2. The circular nodes represent nodes that are fixed whereas, the square nodes are nodes that have parameters to be learnt. During training, all of the training dataset would be present to network and it tries by learning.
Layer 1: The node function in Layer 1 is the membership of the fuzzy
set associated with the corresponding input. The first order Sugeno fuzzy model
provides the following rule based structure by Sadighi and
Kim (2011).
If Age is young and sex is female and cp is low and trestbps is medium and chol is high and fbs is low and restecg is high and thalach is low and exang is high and oldpeak is medium and slope is high and ca is high and thal is normal then f1 = age.x+sex.y+cp.z+trestbps.a+chol.b+fbs.c+ restecg.d+thalach.e+exang.f+oldpeak.g+slope.h+ca.i+thal.j+t.
where, high, medium, low, young, normal are fuzzy sets for the input, {Age,
sex, cp, trestbps, chol, fbs, restecg, thalach, exang, oldpeak, slope, ca, thal}
is the consequent parameter set and f1 is the output. The membership functions
where parameterized using the generalized Gaussian function (Alves
et al., 2011):
where, c and σ represent the membership function center and width respectively in order to determine coordinates of Gaussian membership function.

Fig. 2: 
Architecture of the fuzzy resolution mechanism using ANFIS 
Layer 2: Tnorm operator that perform the AND operator can be used (Liu
et al., 2010). Every node in this layer is a fixed node labeled Prod.
The output is the product of all the incoming values. In layer 2, multiplies
the inputs from the nodes in layer 1 and generates the firing strength of the
rules. The output of this layer is given by:
where, w_{i} is the firing strength of rule i. Layer 3: Regularize all the rules firing strengths and node function in Layer 3. The i^{th} node calculates the ration of the i^{th} rules firing strength to the sum of all rules firing strengths:
where, the output are called normalized firing strengths is of this layer. Layer 4: The nodes in this layer are adaptive and perform the consequent of the rules:
where,
is a normalized firing strength from layer 3 and {Age_{i}, sex_{i},
cp_{i}, trestbps_{i}, chol_{i}, fbs_{i}, restecg_{i},
thalach_{i}, exang_{i}, oldpeak_{i}, slope_{i},
ca_{i}, thal_{i}, t_{i}} are the parameter set of this
node. Parameters in this layer are referred to as consequent parameters.
Layer 5: There is a single node that computes the overall output:
The input vector is fed through the network layer by layer. We now consider how the ANFIS learns the premise and consequent parameters for the membership functions and the rules.
ALGORITHM FOR FUZZY RESOLUTION MECHANISM
Input: Input the fuzzy set for Age, sex, cp, trestbps, chol, fbs, restecg,
thalach, exang, oldpeak, slope, ca and thal.
Output: Output the fuzzy set for num(angiographic disease status). Method: Begin: • 
Step 1: Input the crisp values for Age, sex.cp, trestbps,
chol, fbs, restecg, thalach, exang, oldpeak, slope, ca and thal 
• 
Step 2: Set first order sugeno fuzzy model, common
rule set with fuzzy ifthen rules 
• 
Step 3: ANFIS is executed by Sugeno method 
• 
Step 4: Layer 1every node is an adaptive node with node function 
where, x is input to node and Age, sex.cp, trestbps, chol, fbs, restecg, thalach, exang, oldpeak, slope, ca and thal is a linguistic label associated with this node.
• 
Step 5: In layer 2, multiplies the inputs from the
nodes in layer 1 and generates the firing strength of the rules. Tnorm
operator that perform the AND operator is used 
• 
Step 6: Layer 3 contains fixed nodes. The i^{th}
node calculates the ration of the i^{th} rules firing strength to
the sum of all rules firing strengths 
• 
Step 7: In Layer 4, the nodes in this layer are adaptive
and perform the consequent of the rules 
• 
Step 8: There is a single node here that computes the
overall output 
• 
Step 9: Present the knowledge in the form of human
natural language 
End.
Defuzzification: Defuzzification process is conducted to convert aggregation
result into crisp value for num output. In this process the single number represent
the outcome of the fuzzy set evaluation. The final combined fuzzy conclusion
is converted into a crisp value by using the weighted average method (Alves
et al., 2011).
EXPERIMENTAL RESULTS The proposed Fuzzy Resolution Mechanism for heart disease was implemented with the MATLAB. The experimental environment was constructed to evaluate the performance of the proposed approach with Cleveland data set. The first experiment shows sets of results in Fig. 3 and 4, indicating that the proposed approach automatically supports the analysis of the data. The Decision can be taken from the about the status of angiographic disease. Neural network was developed with the training data, with rule, parameter and membership function.

Fig. 3: 
ANFIS modeling framework 

Fig. 4: 
Result obtained from MATLAB 
Table 2: 
Comparison of proposed method accuracy with earlier methods 

EVALUATION OF SYSTEM PERFORMANCE
The second experiment evaluates the performance of the system. Accuracy is
the common performance metrics used in medical diagnosis tasks. The measure
of the ability of the classifier to produce accurate diagnosis is determined
by accuracy. So that accuracy Loo and Rao (2005) is
given by Eq. 2:
The final experiment compares the accuracy of the proposed method with results
of studies involving the Cleveland heart disease dataset (Tang
et al., 2004; Kahramanli and Allahverdi, 2008;
Senthil Kumar, 2011). Comparing these methods, as listed
in Table 2, reveals that the proposed method achieves the
first highest accuracy values based on the proposed Fuzzy Resolution Mechanism.
CONCLUSIONS AND FUTURE RESEARCH Fuzzy Resolution Mechanism is used to diagnosis the heart disease. The experimental, Cleveland heart disease dataset, is initially processed and the crisp values are converted into fuzzy values in the stage of fuzzification. The Fuzzy Resolution Mechanism undergoes five layers to execute rules, to make a decision on the possibility of individuals suffering from heart disease. Defuzzification process is conducted to convert the result into crisp value for angiographic disease status. Experimental results indicate that the proposed method can analyze data more efficiently than other methods. Future works should test the Fuzzy Resolution Mechanism approach used herein for other similar tasks or other related data sets to evaluate its capability to produce a similar accuracy.

REFERENCES 
Alves, M.C., E.A. Pozza, J.C.B. Costa, L.G. de Carvalho and L.S. Alves, 2011. Adaptive neurofuzzy inference systems for epidemiological analysis of soybean rust. Environ. Modell. Software, 26: 10891096. CrossRef  Direct Link 
Dich, W., K. Grudzinski and G.H.F. Diercksen, 1998. Minimal distance neural methods. Proceedings of the World Congress of Computational Intelligence, May 49, 1998, Anchorage, Alaska, USA., pp: 12991304.
Duch, W. and K. Grudzinski, 1998. A framework for similaritybased methods. Proceedins of the 2nd Polish Conference on Theory and Applications of Artificial Intelligence, September 2830, 1998, Lodz, Poland, pp: 3360.
Duch, W., R. Adamczak, K. Grąbczewski and G. Zal, 1998. A hybrid method for extraction of logical rules from data. Proceedings of the 2nd Polish Conference on Theory and Applications of Artificial Intelligence, September 2830, 1998, Lodz, Poland, pp: 6182.
Fasanghari, M. and G.A. Montazer, 2010. Design and implementation of fuzzy expert system for Tehran stock exchange portfolio recommendation. Expert Syst. Appl., 37: 61386147. CrossRef  Direct Link 
Forouzanfar, M., H.R. Dajani, V.Z. Groza, M. Bolic and S. Rajan, 2010. Adaptive neurofuzzy inference system for oscillometric blood pressure estimation. Proceedings of the IEEE International Workshop on Medical Measurements and Applications, April 30May 1, 2010, Ottawa, Canada, pp: 125129.
Haykin, S., 1999. Neural Networks: A Comprehensive Foundation. 2nd Edn., PrenticeHall International Inc., New Jersey, USA., ISBN13: 9780139083853, Pages: 842.
Jang, J.S.R., 1993. ANFIS: Adaptivenetworkbased fuzzy inference system. IEEE Trans. Syst. Man Cybernet., 23: 665685. CrossRef 
Jankowski, N. and V. Kadirkamanathan, 1997. Statistical control of RBFlike networks for classification. Proceedings of the 7th International Conference on Artificial Neural Networks, October, 1997, Lausanne, Switzerland, pp: 385390.
Kahramanli, H. and N. Allahverdi, 2008. Design of a hybrid system for the Diabetes and heart diseases. Expert Syst. Appl., 35: 8289. CrossRef  Direct Link 
Kosko, B., 1994. Fuzzy systems as universal approximators. IEEE Trans. Comput., 43: 13291333. CrossRef 
Liu, M., M. Dong and C. Wu, 2010. A new ANFIS for parameter prediction with numeric and categorical inputs. IEEE Trans. Autom. Sci. Eng., 7: 645653. CrossRef 
Loo, C.K. and M.V.C. Rao, 2005. Accurate and reliable diagnosis and classification using probabilistic ensemble simplified fuzzy ARTMAP. IEEE Trans. Knowledge Data Eng., 17: 15891593. CrossRef 
Mukhopadhyay, S., C. Tang, J. Huang, M. Yu and M. Palakal, 2002. A comparative study of genetic sequence classification algorithms, neural networks for signal processing. Proceedings of the 12th IEEE Workshop on Neural Networks for Signal Processing, September 46, 2002, Martigny, Switzerland, pp: 5766.
Roan, S.M., C.C. Chiang and H.C. Fu, 1993. Fuzzy RCE neural network, fuzzy systems. Proc. IEEE Int. Conf. Fuzzy Syst., 1: 629634. CrossRef 
Sadighi, A. and W.J. Kim, 2011. Adaptiveneurofuzzybased sensorless control of a smartmaterial actuator. IEEE/ASME Trans. Mechatron., 16: 371379. CrossRef  Direct Link 
Senthil Kumar, A.V., 2011. Adaptive neurofuzzy inference system for heart disease diagnosis. Proceedings of the International Conference on Information System, Computer Engineering and Application, November 2829, 2011, Singapore .
Serpen, G., H. Jiang and L.G. Allred, 1997. Performance analysis of probabilistic potential function neural network classifier. Proc. Artif. Neural Networks Eng. Conf., 7: 471476.
Ster, B. and A. Dobnikar, 1996. Neural networks in medical diagnosis: Comparison with other methods. Proc. Int. Conf. EANN, 1: 427430. Direct Link 
Tang, Y., B. Jin, Y. Sun and Y.Q. Zhang, 2004. Granular support vector machines for medical binary classification problems. Proceedings of the IEEE Symposium on Computational Intelligence in Bioinformatics and Computational Biology, October, 78, 2004, USA., pp: 7378.



