Real-Time Heart Murmur Classification using Attention Based Deep Learning Approach

In recent years, Cardiovascular diseases (CVDs) characterized by abnormal heart sounds and pathological heart murmurs rise serious concern across the globe. In this work, an automated detection system using attention-based one dimensional convolutional neural network(1-D CNN) framework for pathological heart murmurs is proposed. A coherent and efficient attention mechanism is developed for upholding the detailed feature mining with minimum time and space complexities in classical CNN model. In addition, the proposed model facilitates the direct application of raw phonocardiogram (PCG) signals and abstracts away the difficult task of feature extraction and feature selection. Adaptive synthetic sampling (ADASYN) approach is used for handling the imbalanced heart murmur data. The entire framework is trained on general purposes Operating system and tested on Raspberry PI based embedded platform to manifest the real-time murmur identification. The designed framework outperforms the conventional CNN classification model against the balanced and imbalanced datasets. Therefore, with the detailed refinement, the developed framework greatly aids the cardiologists in the proper management of cardiac disorders.