Statistical signal processing approach to segment primary components from pathological phonocardiogram

Abstract

Cardiac disorders has become pretty common in the current world. Despite of the availability of many advanced techniques like electrocardiography (ECG), Echocardiography and Carotid pulse, listening to the heart sounds has become one of the orthodox approach which is being performed from long ago, often named as auscultation methodology or Phonocardiogram. This methodology is the primary tool for the health care physicians to screen the patients for heart pathology. However, to master, it needs a lot of experience and knowledge. Yet the non-availability of advanced techniques at every door step and its cost made this orthodox approach to survive. The proposed study is to make the health care physicians to diagnose the pathology using phonocardiography in an effective manner. The study uses the statistics of the signal information in the form of variance. The proposed technique uses filtering and decimation as preprocessing method to limit the low frequency noises/disturbances and to concentrate only on the components of interest (i.e. S1 and S2). The preprocessed signal is wavelet analyzed and synthesized followed by principal component analysis to extract necessary features which resembles the information of S1 and S2. A proposed splitting algorithm is processed to the featured signal to separate the phonocardiogram signal into series of cardiac cycles and energy envelope is calculated for the same featured signal. By using the information of the cardiac cycles and energy envelopes, segmentation of S1 and S2 from pathological phonocardiogram is performed. The results show that the proposed technique does not rely on any time-frequency parameter which effects the performance of the study. Hence a novel technique based on statistical analysis has been proposed to detect the primary components (S1 and S2) from pathological phonocardiogram with less computational effort and better accuracy.