Development of an ANFIS Neural Network System for Early Diagnosis of Critical Health Condition of a Patient

Abstract

Aim of this work is to design and training of a ANFIS neural network based early diagnosis of a patient through a smart instrument. The data collected from the patient is used to train the Neural Network. The rules are designed on the basis of the collected data. The data collected from the patients are weight, BMI, Glucose, Creatinine, Systolic BP, and Diastolic BP. The rules for system are given according to the characteristics of the data obtained from patients. The output is mainly classified into three classes that are severe, moderately critical, and normal. Output of Neural Network is connected to LED to display the corresponding outputs. The advantage of this system is we can implement this device as portable. Hence we can easily monitor the patient's condition at anywhere. And for the normal type of application we need a personal computer to compute the inputs and outputs. Because here we are using FPGA the problem can be easily avoided and also PC will need an uninterrupted power supply. For this device just battery power is enough to work. The FPGA will take a very small power only hence we can easily use the device a long time without charging again and again. These are some of the major advantages of Smart device. The main disadvantage of using fuzzy logic is the reconfiguration problem. When we change the input parameters we should also change the rules. That is one of the important disadvantages of the fuzzy logic. This will also affect the cascaded structure of the neural network. Hence the performance of the system will effect sufficiently. The changing of rules can do only by the technical experts. This also will increase the complexity of the system. The problem can be easily solved by using ANFIS instead of fuzzy controllers. By using this we can avoid the re configuration complexity problem. Because ANFIS generates rules automatically from the training data. Hence no need for giving rules externally. Hence we can change the input data as per the user wish. Training complexity can be reduced by this method.