Intelligent Cardiac Auscultation for Murmur Detection via Parallel-Attentive Models with Uncertainty Estimation

TL;DR

This paper presents a novel parallel-attentive model with uncertainty estimation for heart murmur detection, improving interpretability and reliability in clinical decision-making, and achieving state-of-the-art accuracy on a public dataset.

Contribution

The study introduces a dual-branch model combining self-attention and convolutional networks, along with uncertainty estimation and calibration techniques, for enhanced murmur detection.

Findings

Achieved 79.8% weighted accuracy on CirCor Digiscope dataset.

Attained 65.1% F1 score, outperforming existing methods.

Enhanced model reliability through uncertainty estimation and calibration.

Abstract

Heart murmurs are a common manifestation of cardiovascular diseases and can provide crucial clues to early cardiac abnormalities. While most current research methods primarily focus on the accuracy of models, they often overlook other important aspects such as the interpretability of machine learning algorithms and the uncertainty of predictions. This paper introduces a heart murmur detection method based on a parallel-attentive model, which consists of two branches: One is based on a self-attention module and the other one is based on a convolutional network. Unlike traditional approaches, this structure is better equipped to handle long-term dependencies in sequential data, and thus effectively captures the local and global features of heart murmurs. Additionally, we acknowledge the significance of understanding the uncertainty of model predictions in the medical field for clinical decision-making. Therefore, we have incorporated an effective uncertainty estimation method based on Monte Carlo Dropout into our model. Furthermore, we have employed temperature scaling to calibrate the predictions of our probabilistic model, enhancing its reliability. In experiments conducted on the CirCor Digiscope dataset for heart murmur detection, our proposed method achieves a weighted accuracy of 79.8% and an F1 of 65.1%, representing state-of-the-art results.