Transfer Learning in Heart Sound Classification using Mel Spectrogram

Congenital heart illnesses impact roughly 1% of newborns, and they are a significant cause of morbidity and mortality in a variety of serious situations, including progressive heart failure. Phonocardiogram (PCG) studies can reveal crucial clinical information about heart malfunction caused by congenital and acquired heart disease. One of the 23th PhysioNet/Computing in Cardiology Challenge 2022 tasks is to develop computer tools for detecting the presence or absence of murmurs from multiple heart sound recordings from multiple auscultation locations. Mel spectrograms were generated from up to 30 seconds per recording and reshaped at input of pre-trained AlexNet. The last three layers of AlexNet were modified to suit the task as multilabel classification. The database was split into 80% for training and 20% for validation. The database appeared imbalanced, so the class with small number of data entries was oversampled proportionally before training. The prepossessing and classifier were implemented in Matlab R2022a. Team Leicester Fox's final score in the official phase achieved challenge scores of 0.536 for murmur detection (ranked at 32/40) and 13844 for outcome prediction (ranked at 26/39). Transfer learning and neural networks approaches showed potential application for murmurs detection using PCG.