A Lightweight Robust Approach for Automatic Heart Murmurs and Clinical Outcomes Classification from Phonocardiogram Recordings

Hui Lu1, Julia Yip2, Tobias Steigleder2, Stefan Grießhammer3, Naga Venkata Sai Jitin Jami4, Bjoern Eskofier4, Christoph Ostgathe2, Alexander Koelpin5
1Brandenburg University of Technology, 2Palliative Medicine, Universitätsklinikum Erlangen, 3Palliative Medicine, Universitätsklinikum Erlangen,, 4Friedrich-Alexander-Universität Erlangen-Nürnberg, 5Institute of High-Frequency Technology, Hamburg University of Technology


Cardiac auscultation provides an efficient and cost-effective way for cardiac disease pre-screening. The George B. Moody PhysioNet Challenge 2022 aimed to detect heart murmurs and clinical outcomes with heart sound recordings from multiple auscultation locations. Our team HearHeart proposed a lightweight convolutional neural network (CNN) to detect murmurs and a random forest model to classify clinical outcomes. 128 Mel-spectrogram features and wide features like the socio-demographic data and statistical features are extracted. Different techniques are employed to migrate the data imbalance and model the overfitting problem. We used two data augmentation methods, noise injection and spectrogram augmentation in the time and frequency domain to increase the training samples and avoid overfitting during training. Besides, weighted loss functions are applied to both tasks to deal with data imbalance. In the end, we ensembled the models from cross-validation and used voting for the final classification. We achieved a murmur score of 0.791, a clinical outcome score of 11731.64 on 5-fold cross-validation and a murmur score of 0.751, and a clinical outcome score of 11459 on the hidden test data.