Deep Learning-Based Classification of PCG Signals Using Chaogram Transform and CNN-LSTM Network

The localization and classification of Phonocardiogram (PCG) signals have been very important in the medical field. It is highly challenging to identify heart diseases using the PCG signal without proper feature extraction and classification. In this paper, a deep learning-based classification using a combination of CNN-LSTM network with Chaogram transform-based feature extraction is proposed. PCG signal converted as Chaogram image will serve as the input to CNN-LSTM network. Following training the network model is tested and the obtained findings were compared with other existing methods. An overall accuracy of 89.68% is obtained by the proposed model. By taking future consideration such as proper data augmentation, precise calculation of embedding dimension and using parallel architecture of CNN-LSTM the efficiency of the proposed system can be increased further.