ARTIFICIAL NEURAL NETWORK BASED ABNORMALITIES CLASSIFICATION USING HEART RATE VARIABILITY SIGNAL

Abstract

Since the late 1980s, the study of heart rate variability as a measure of autonomic modulation of the cardiovascular system has grown with pace. For this reason, the analysis of their variability has gained growing importance both for the clinical evaluation and physiological studies. This noninvasive measure provides important information about the cardiovascular function in healthy and diseased subjects. Different time domain and frequency domain measures are used for the quantification of aifonomic involvement in the cardiovascular system. In this thesis, Once the RR intervals were extracted from ECG signal, missing beats and ectopic beats were taken care of using Cheung's algorithm and visual inspection. The. beat-to-beat variability, normally called heart rate variability (HRV) analysis was made using Autoregressive (AR) modeling. In AR modeling, the input time series should be equi-spaced. Since the time interval between two R peaks is not uniform, interpolation and resampling algorithm was implemented to convert it to equi spaced RR' time series. In AR modeling, the current sample was estimated as the weighted linear sum of previous -values plus white noise. The order of the model (the number of previous values to be added) was selected by using standard order selection criteria called Akaik's information criteria (AIC) and final prediction error (FPE), where we observed no noticeable difference between the two in our study. The validity of the selected model was tested by using Anderson's test.