Heart Murmur Detection and Clinical Outcome Prediction Using Multilayer Perceptron Classifier

Abstract

Abnormal sound waves, or heart murmurs in Phonocardiogram (PCG) recordings, are potential indicators of congenital and acquired heart disease in pediatric populations. Detection of these murmurs, and therefore early diagnosis, is usually performed by cardiology specialists. In the 2022 PhysioNet/Computing in Cardiology Challenge, under the team name AKSJ_97BSc, we proposed a Multilayer Perceptron-based model that automatically classifies patient heart murmur status into three categories of present, unknown, or absent based on their metadata and PCG recording. The model also further divides the patients into two outcomes, indicating whether the patient's clinical outcome diagnosed by the medical expert is normal or abnormal. The model was ranked 231 out of 305 submissions with a 0.491 challenge score in the murmur classification category and 161 out of 305 submissions with a 11330.062 challenge score in the outcome classification category.