Phonocardiogram (PCG) Signals Based Classification of Heart Abnormalities

Cardiovascular disease (CVD) is a serious illness that affects over the world. Early detection and prevention of CVD is thought to help reduce CVD mortality rates. There are many advanced technologies available for detecting CVD related symptoms, however they are only available in urban regions. Healthcare institutions in rural areas would only be equipped with minimal diagnostic devices, and the primary investigate tool would be the stethoscopes. Therefore, this study focusses on the use of phonocardiogram (PCG) data as a means of detecting heart abnormalities through machine learning approach. This study would concentrate on the use of phonocardiogram signals to detect abnormalities such as murmur, extrasystole, which are frequently cited in literature as early indicator for numerous CVD related illness. A sixth order Butterworth filter with a frequency range of 25Hz - 900Hz was used to remove undesired signal followed by a zerophased digital filter. The signal was then segmented using average Shannon energy. Both morphological features as well as non-linear features derived from Hilbert-Huang Transform (HHT) was used to analyze the PCG signals. Analysis of Variance (ANOVA) was used to select the statistically significant features for classification. Support Vector Machine (SVM), Ensemble and K Nearest- Neighbor (KNN) classifiers were used to quantify the efficacy of the proposed methods in detecting heart abnormalities. Overall accuracy of 70.2 % was achieved using the proposed features and Ensemble classifier. The experimental outcome shows that the use of PCG data is a promising field for development of non-invasive early screening system for CVD and could potentially help to uplift the general healthcare of rural populations..