ANN BASED CLASSIFICATION OF HRV SIGNALS

Abstract

Measurement of heart rate variability (HRV) is a non-.invasive technique that can be used to classify different diseases. It has been proven to be very useful in humans for both research and clinical studies concerned with cardiovascular diseases. In the world, the cardiac dysfunction is the main reason of the sudden natural death. The autonomic nervous system affects the cardiovascular functioning due to which heart rate variability arises. So it can also help to study the behavior of autonomic nervous system for the diseases. We get the different features from the HRV data, which is useful to identify the diseases. The HRV data is analyzed in time domain and frequency domain. The parametric (AR modeling) method is used here for extracting parameters in frequency domain. Data is pre-processed (correction of ectopic beats, even sampling) for getting better result. Covariance method and akaik's information criterion (AIC) is used for calculating model parameters and selecting better order of AR model respectively. Here HRV data of 10 minutes is taken from physiobank database of, Atrial Fibrillation, Supraventricular Arrhythmia, normal humans, CHF (Congestive heart failure) and arrhythmia. After the signal processing in time domain and frequency domain we get 11 features. These features provide a range by which we can detect the diseases. These features are used for classifying diseases by using back propagation feed forward artificial neural network. We have a data set of 255 cases of arrhythmia, Atrial Fibrillation, Supraventricular Arrhythmia, CHF and normal humans. We have 10 features for the data classification. One feature is separate out from the total number of features as it have a high correlation with other features. In the all artificial neural networks, supervised learning technique is used. We developed four networks in this dissertation for classification of diseases which are 1) between normal sinus rhythm and diseases 2) between normal sinus rhythm and diseases 3) between normal sinus rhythm and CHF 4) between normal sinus rhythm and Arrhythmia. These networks provide us a good result based on our performance criteria (Sensitivity, Specificity, Positive predictive value and Accuracy).