Exploring a Segmentation-Classification Deep Learning-based Heart Murmurs Detector

Abstract

This work presents the advances of the UZ-ULPGC team in the Heart Murmur Detection from Phonocardiogram Recordings: The George B. Moody PhysioNet Challenge 2022. As the 2016 challenge proved the success of the combination of a segmentation algorithm and a classifier, a deep learning-based murmur detector is developed using the sequence segmentation-classification. A U-Net-based segmentation model is used to extract each cardiac cycle from the PCG with state-of-the-art accuracy. Three deep models are tested for the classification: a model based on four independent 1D-convolutional feature extractors; its variation enabling combination of the features; and an autoencoder. Furthermore, to enable unique patient diagnostic, a decision model gathering all the patient-related cardiac cycles information is added. All classifiers show limited performance, probably due to the heavy class imbalance of the data at the cardiac cycle level and the minimal preprocessing chosen in the architecture. Note that our models have not been tested in the hidden challenge data and therefore we are not ranked. Hence, a 10-fold cross-validation over the training set is used to evaluate their performance, with the best model getting a weighted accuracy score in the presence task of 0.58 ± 0.10 and 10 735 ± 2208 in Challenge cost score for the outcome.