Murmur Identification Using Supervised Constrastive Learning

Ľubomír Antoni1, Erik Bruoth1, Alexander Szabari1, Gabriela Vozáriková1, Peter Bugata2, Peter Bugata Jr.2, Dávid Gajdoš2, Dávid Hudák2, Vladimíra Kmečová2, Monika Staňková2
1Pavol Jozef Šafárik University, 2VSL Software, a.s.


As part of the George B. Moody PhysioNet Challenge 2022, we developed a computational approach to identify abnormal cardiac function from phonocardiograms that combines deep learning and traditional machine learning methods. We adopted a supervised contrastive learning and a deep convolutional neural network to obtain an embedding of the phonocardiogram slice onto a unit hypersphere in low-dimensional space. Thus, we applied the obtained latent factors to classify patients using a Random Forest model. The murmur detection classifier created by our team CeZIS received a weighted accuracy score of 0.796 (ranked 3rd out of 305 submissions) and Challenge cost score of 9479 (ranked 22nd out of 305 submissions) on the hidden validation set.