Convolutional Neural Network Approach for Heart Murmur Sound Detection in Auscultation Signals Using Wavelet Transform Based Features

The heart auscultation signal contains strong beats representing cardiac valve closures and the murmur sounds (if present). These key-components of the signal differ in time and frequency, therefore continuous wavelet transform (CWT) was proposed for features formation. The result of CWT of randomly taken excerpt of the signal is two-dimensional array. It contains bold areas of high value estimates representing the strong beats and some areas of moderate values representing murmur sounds in case they are present. Strong beat representations in these arrays give the time marks for the eventual representations of sought murmur sounds. Therefore, we did not do the signal segmentation, but we calculate CWT results of sliding-overlapping windows along the whole signal instead. For final analysis we use CWT-results per recording, having lowest, non-zero entropy. Therefore, we get rid of noisy or corrupted signal parts. The convolutional neural network does the final classification. We used the same convolutional neural network and CWT features to classify patient's clinical outcomes. Algorithm was tested on the George B. Moody PhysioNet Challenge 2022 hidden test set. "LSMU" team's murmur classifier received a weighted accuracy score of 0.671 (ranked 23th out of 40 teams) and Challenge cost score of 15402 (ranked 35th out of 39 teams).