Heart Sound Classification using Residual Neural Network and Convolution Block Attention Module

Listening to the heart sound with digital or manual stethoscopes has become one of the practical ways to identify heart diseases in recent years. It's still difficult because of its manual approach and the fact that only experienced healthcare practitioners can use it to diagnose anomalies. The automatic extraction of heart sound features to aid in classification has been explored, however there is still potential for improvement. This paper proposes a residual neural network integrated with a convolutional block attention module (CBAM) for heart sound analysis, using generated Mel-spectrograms as input for our network. We tested our model using the Pascal Heart Sound Challenge dataset, and it performed favorably to other cutting-edge models.