Ischemia prediction via ECG using MLP and RBF predictors with ANFIS classifiers

Abstract

In this paper, a new algorithm is presented in using Multi Layer Perceptron (MLP) and Radial Base Function (RBF) to predict Ischemia diseases by Electrocardiogram (ECG) signals. The process would be very difficult due to non-stationary and nonlinear characteristics of ECG signals. MLP and RBF algorithms are well known in predicting the problems. However, they have not been used for real time prediction through signals, especially bio signals such as ECG. Pre-processing is necessary for ECG signal in order to detect QRS complex. Regarding the extract influential features in Ischemia disease, the baseline wandering and noise suppression are done. MLP and RBF, the predictors, are employed to foresee the further next beats in ECG signals. The validity of predictor accuracy is evaluated by Root Mean Square Error (RMSE) criterion. After the prediction stage, The predicted beats are classified by Adaptive Neuro-Fuzzy network (ANFIS) classifier as ischemic and normal. MLP and RBF are tested for their abilities in order to predict Ischemic Heart Disease (IHD) upon ECG signals. The performances of classified beats are evaluated based on computed Sensitivity (Se) and Specificity (Sp). In this study several ECG signals recorded by European Society of Cardiology for ST-T database are used. By applying prediction methods (Direct and Recursive Predictions) 48 steps can be predicted ahead in ECG signal. Then the predicted beats are classified as Ischemic or normal beats. Therefore, the ischemic beats can be predicted in 48 steps ahead. By comparing the results obtained in this study, the MLP and RBF networks are evaluated for their capabilities in predicting Ischemia. According to this comparison, MLP shows better results and the results of ANFIS as a classifier has been satisfactory enough in classification of Ischemic beats. Therefore, these results can be used for early diagnosis of Ischemic Heart Disease (IHD).