Convolutional Neural Network Based Heart Sounds Recognition on Edge Computing Platform

Cardiovascular diseases (CVDs) are often characterized by audio properties of heart sounds. The recognition of first and second heart sounds namely S-1 and S-2 is the first and the foremost step in detecting any cardiac abnormality. This work presents an implementation of optimal 1-Dimensional Convolutional Neural Network (1-D CNN) for S-1 and S-2 recognition on an edge computing platform. The optimal 1-D CNN model trained in the general purpose computing platform is deployed in the Raspberry Pi 3B based edge device for real time recognition of S-1 and S-2. In addition, to the best of our knowledge, this study is novice and aims for investigation of S-1 and S-2 recognition considering different portions of the segmented S-1 and S-2 signals obtained from the publicly available database as well as experimentally recorded phonocardiogram(PCG) signals. The achieved recognition accuracies are promising for both the databases with different decimation factors of data. The complete setup emulates the real time environment for heart sound recognition, thereby it paves a way for further analysis of the PCG signals for assisting physicians.