Classification of short unsegmented heart sound based on deep learning

Heart-related ailments are one of the primary causes of death worldwide. Hence the early investigation of a heart for such ailments is crucial. Recent approaches for automated analysis of the heart sounds require segmentation of Phonocardiograms (PCG) signal. However, segmentation of PCG adds up to the complexity and expanded computational difficulty in the algorithm. Thereby, the main aim of this paper is to eliminate the segmentation process and to measure the benefit for accurate and detailed classification of short unsegmented 5 second PCG recordings. A novel approach for the classification of heart sounds that had been provided by PhysioNet2016 challenge, based on the convolutional neural network using a pre-trained (AlexNet) model has been analyzed in this study. After pre-processing short 5 second PCG recordings accompanied by continuous wavelet transform (CWT) results to the generation of 2D scalogram images. The scalogram images have been used to train and test Convolutional neural network based on deep learning. The proposed design has obtained comparable performance compared to the state-of-the-art methods. Test results have demonstrated that the proposed technique presents excellent performance outcomes by reducing segmentation complexity.