Deep determination of cardiac condition from phonocardiograms

Article Type

Research Article

Publication Title

Neural Computing and Applications

Abstract

Every year, a significant volume of the global population is affected by coronary artery disease (CAD). On the other hand, heart murmur, an unusual whooshing sound caused by turbulent blood flow through the heart, often remains undetected at the initial stages. However, attention to such issues often gets delayed due to non-availability of medical experts and the situation is more serious in villages or remote areas. Robust automatic systems for analyzing results of cost-effective relevant diagnostic tests, like phonocardiography, which is capable of high-fidelity recording of heart sounds and murmurs, may be a useful alternative, particularly for low-income groups of the population of developing and underdeveloped countries. Here, we describe our recent study of a novel approach based on a deep architecture, incorporating pyramid dilated convolution and multi-headed attention, to predict the cardiac status related to CAD and murmurs. Utilizing image representations of phonogram samples, the proposed strategy has been tested on CirCor DigiScope and 2016 PhysioNet/CinC Challenge, two large manually annotated datasets of heart sound samples from four major auscultation locations. The detection results obtained on the test set demonstrate significant improvement over the state-of-the-art. Results of comparative as well as ablation studies establish the effectiveness of the proposed strategy.

First Page

26099

Last Page

26123

DOI

10.1007/s00521-025-11617-4

Publication Date

11-1-2025

Link to Full Text

Share

COinS