Computer Aided Detection of Normal and Abnormal Heart Sound using PCG

A PCG (phonocardiogram) is a method of plotting of heart sounds and murmurs during a cardiac cycle, with the help of machine called phonocardiograph. A PCG can be visually represented. PCG recordings comprise of bio-acoustic statistics indicating the functional condition of the heart. Intelligent and automated analysis of the PCG is therefore very important not only in detection of cardiac diseases but also in monitoring the effect of certain cardiac drugs on the condition of the heart. PCG analysis includes segmentation of the PCG signal, feature extraction from the segmented signal and then classification. We used Kaggle data sets [10] and have extracted feature sets of different domains i.e. Time domain, frequency domain and statistical domain. We used 8 features of 118 recordings and train our different classifiers (Bagged Tree, subspace Discriminant, Subspace KNN, LDA, Quadratic SVM and Fine Tree) to obtain and compare accuracy and results. We use only two classes for classification i.e. normal and abnormal. Out of these 6 classifiers Bagged tree gave highest accuracy of 80.5%.