# Beyond Heart Murmur Detection: Automatic Murmur Grading from Phonocardiogram

**Authors:** Andoni Elola, Elisabete Aramendi, Jorge Oliveira, Francesco Renna, Miguel T. Coimbra, Matthew A. Reyna, Reza Sameni, Gari D. Clifford, Ali Bahrami Rad  
**arXiv:** [2209.13385](https://arxiv.org/abs/2209.13385)  
**Date:** 2022-09-27  
**Summary page:** https://www.summarizepaper.com/en/paper/2209.13385v1/

## Key points

- The study aims to develop an algorithmic approach for automatic murmur grading from phonocardiograms (PCGs) of a large population of pediatric patients from a low-resource rural area.
- Murmurs are abnormal heart sounds that can be identified by experts through cardiac auscultation, and the murmur grade is strongly correlated with the patient's clinical condition.
- The proposed method uses Mel spectrogram representation of each PCG recording and an ensemble of 15 convolutional residual neural networks with channel-wise attention mechanisms to classify each PCG recording.
- The final murmur grade for each patient is derived based on the proposed decision rule and considering all estimated labels for available recordings.
- The method was cross-validated on a dataset consisting of 3456 PCG recordings from 1007 patients using a stratified ten-fold cross-validation and tested on a hidden test set comprised of 1538 PCG recordings from 442 patients.
- The overall cross-validation performances for patient-level murmur gradings are 86.3% in terms of unweighted average sensitivities and 81.6% in terms of F1-scores.
- On the test set, the algorithm achieved an unweighted average sensitivity score of 80.4% and an F1-score of 75.8%.
- The sensitivities (and F1 scores) for absent, soft, and loud murmurs were 90.7% (93.6%), 75.8% (66.8%), and 92.3% (84.2%) respectively; indicating that the proposed method provides characterization beyond detection that may enhance classification accuracy in resource limited settings where expert screening costs are relatively high.

## Summary

This study aims to develop an algorithmic approach for automatic murmur grading from phonocardiograms (PCGs) of a large population of pediatric patients from a low-resource rural area. Murmurs are abnormal heart sounds that can be identified by experts through cardiac auscultation, and the murmur grade, which is a quantitative measure of the murmur intensity, is strongly correlated with the patient's clinical condition. The proposed method uses Mel spectrogram representation of each PCG recording and an ensemble of 15 convolutional residual neural networks with channel-wise attention mechanisms to classify each PCG recording. The final murmur grade for each patient is derived based on the proposed decision rule and considering all estimated labels for available recordings. The method was cross-validated on a dataset consisting of 3456 PCG recordings from 1007 patients using a stratified ten-fold cross-validation and tested on a hidden test set comprised of 1538 PCG recordings from 442 patients. The overall cross-validation performances for patient-level murmur gradings are 86.3% in terms of unweighted average sensitivities and 81.6% in terms of F1-scores. On the test set, the algorithm achieved an unweighted average sensitivity score of 80.4% and an F1-score of 75.8%. The sensitivities (and F1 scores) for absent, soft, and loud murmurs were 90.7% (93.6%), 75.8% (66.8%), and 92.3% (84.2%) respectively; indicating that the proposed method provides characterization beyond detection that may enhance classification accuracy in resource limited settings where expert screening costs are relatively high.