PCA-based heart sound feature generation for a ventricular septal defect discrimination

Abstract

In this study, a simple system based on PCA-based heart sound feature extraction is proposed for discriminating ventricular septal defects (VSDs), which are generally divided into three types: small VSDs (SVSDs), moderate VSDs (MVSDs) and large VSDs (LVSDs). The three stages corresponding to the discrimination system implementation are summarized as follows. In stage 1, the heart sound is collected by a stethoscope and is preprocessed using the wavelet decomposition (WD). In stage 2, time domain features [T12, T11] are first extracted from a time domain envelope Et, which is extracted from the heart sound signal Xt filtered using the WD method, and frequency domain features [Fg, Fw] are subsequently extracted from a frequency domain envelope Ef for one period heart sound, which is automatically segmented from heart sounds based on the short time modified Hilbert transform (STMHT). Finally, the time-frequency feature matrix (TFFM), expressed as TFFM = [T12, T11, Fg, Fw], is generated. In stage 3, the PCA-based diagnostic features y and y2 from SVSD, MVSD, LVSD and normal sounds TFFM expressed as the mean and standard deviation are [−2.41±0.49, 2.16±0.45], [−1.87±0.35, 0.22±0.33], [−1.63 ± 0.56, −2.11 ± 0.68] and [1.11 ± 0.43, 0.09 ± 0.43], respectively. Therefore, for a given heart sound, heart sound features y1 and y2 are generated through three stages to discriminate from different types of VSD sounds and other sounds. Moreover, to validate the usefulness of the proposed diagnostic system, mitral stenosis (MS) and aortic regurgitation (AR) sounds are used as examples for detection analysis using the scatter diagram of [y1, y2].