Weighted Cross-Entropy Loss Based Respiratory Diseases Classification

I. Respiratory sounds contain critical diagnostic information for the early identification of severe lung diseases. The COVID-19 pandemic has amplified the demand for contactless healthcare solutions enabled by electronic stethoscopes. Consequently, advanced deep-learning models have been developed to aid in diagnosing lung conditions. However, these efforts face challenges due to the limited availability of medical data. In this research, we show that pre-trained models on extensive visual and audio datasets can be effectively adapted for respiratory sound classification. We have employed labelaware concatenation in this task effectively to address class imbalance. We further applied a customized loss function to effectively distinguish between all classes. Additionally, to address the class imbalance present in the ICBHI 2017 [1] dataset, data augmentation techniques…