Geodesic Curvature of Heart Sound Dynamics Separates Normal from Abnormal Cardiac Auscultation in 3,240 Phonocardiograms

Read full text Summarise with VIRAT
Authors: Kase Branham
Cited by: 0

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Papers I and II demonstrated that geodesic curvature on a flat torus T² detects heart failure from beat timing and cardiac dysfunction from echocardiographic motion. Here we extend the framework to a third substrate: heart sounds. Per-beat acoustic features—S1/S2 amplitude, systolic/diastolic energy, spectral content, and low-to-high frequency ratio—are extracted from 3,240 phonocardiograms (2,575 normal, 665 abnormal, PhysioNet CinC 2016 Challenge, 6 clinical sites) and mapped onto T² as consecutive beat-pairs.The energy×spectral cross-feature torus emerges as the strongest acoustic torus pairing (r = −0.338, p = 10⁻⁴¹), measuring how the joint relationship between beat loudness and pitch evolves across consecutive beats. This metric survives all duration controls: partial correlation controlling for recording length yields ρ = −0.130 (p = 10⁻¹³), and beat-count matching strengthens it from r = −0.338 to r = −0.344. Five of five tested torus metrics survive duration matching. A combined Basic+Torus classifier achieves AUC = 0.792, outperforming basic acoustic features alone (AUC = 0.749); adding recording duration contributes only 0.004 additional AUC, confirming that the torus signal is not duration-driven.The theoretically significant finding is a Gini direction reversal. In rhythm and imaging, healthy dynamics produce high Gini (curvature concentrated at physiological transitions). In acoustics, abnormal hearts produce higher Gini—15 of 16 metrics—because the murmur itself is a concentrating event that dominates each beat cycle. The torus framework defines a diagnostic geometry of cardiac dynamics that adapts to the structure of each substrate, rather than applying a fixed interpretive rule. Three substrates, one mathematical framework, three clinically distinct geometric signatures.

Fields of Study & Subjects

Phonocardiography and Auscultation Techniques Artificial Intelligence in Healthcare ECG Monitoring and Analysis

Publication details

Journal / source
Zenodo (CERN European Organization for Nuclear Research)
Publication date
30 March 2026
Language
EN