Implementation of an Irregular Lung Sound Diagnostic System

Abstract

This paper presents the implementation of a diagnostic system for respiratory diseases via lung sound. The system consists of an electronic stethoscope device, smartphone, and a server. The smartphone captures lung sound through the electronic stethoscope device and sends data to the server for diagnosis. We propose a three-stage processing algorithm for lung sound classification. The first stage is feature extraction, in which lung sound recordings are processed by using Gammatone-filter bands. Secondly, all the extracted files are divided into training and testing groups; then, data augmentation and mixing up steps are executed. At last, these data are fed in VGG-12 architecture which is attached with an attention mechanism named Convolution Block Attention Module (CBAM). The results show that the integration of the attention mechanism into the VGG-12 network brings a marked improvement in ICBHI score at least 6% compared to VGG-7 and VGG-12. The proposed algorithm is successfully imported to the diagnostic system for respiratory diseases.