NMCSE: Noise-Robust Multi-Modal Coupling Signal Estimation Method via Optimal Transport for Cardiovascular Disease Detection

Peihong Zhang, Zhixin Li, Yuxuan Liu, Rui Sang, Yiqiang Cai, Yizhou Tan, Shengchen Li · 2025

Read NMCSE: Noise-Robust Multi-Modal Coupling Signal Estimation Method via Optimal Transport for Cardiovascular Disease Detection by Peihong Zhang, Zhixin Li, Yuxuan Liu, Rui Sang, Yiqiang Cai, Yizhou Tan, Shengchen Li (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2505-18174.txt.

Introduction

Peihong Zhang and co-authors present a 2025 computing paper on detecting heart disease from paired signals. The heart's electrical trace and its sound trace each tell part of the story. A hidden coupling signal links the two, marking how an electrical spark becomes a mechanical beat.

Older ways to estimate that coupling signal break down under noise, which is common in real clinics. The new method, NMCSE, recasts the task as matching two distributions through optimal transport. Aligning both size and timing at once keeps noise from blowing up. Fed into a feature network, the coupling signal sharpened detection of cardiovascular disease. The team designed two tests that target distinct noise sources met in real-world settings.

Catalog record

CREATORPeihong Zhang, Zhixin Li, Yuxuan Liu, Rui Sang, Yiqiang Cai, Yizhou Tan, Shengchen Li
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE33 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2505-18174

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

NMCSE: Noise-Robust Multi-Modal Coupling Signal Estimation Method via Optimal Transport for Cardiovascular Disease Detection Peihong Zhang, Zhixin Li, Yuxuan Liu, Rui Sang, Yiqiang Cai, Yizhou Tan, Shengchen Li Abstract Electrocardiogram (ECG) and Phonocardiogram (PCG) signals are linked by a latent coupling signal representing the electrical-to-mechanical cardiac transformation. While valuable for cardiovascular disease (CVD) detection, this coupling signal is traditionally estimated using deconvolution methods that amplify noise, limiting clinical utility. In this paper, we propose Noise-Robust Multi-Modal Coupling Signal Estimation (NMCSE), which reformulates the problem as distribution matching via optimal transport theory. By jointly optimizing amplitude and temporal alignment, NMCSE mitigates noise amplification without additional preprocessing. Integrated with our Temporal-Spatial Feature Extraction network, NMCSE enables robust multi-modal CVD detection. Experiments on the PhysioNet 2016 dataset with realistic hospital noise demonstrate that NMCSE reduces estimation errors by approximately 30% in Mean Squared Error while maintaining higher Pearson Correlation Coefficients across all tested signal-to-noise ratios. Our approach achieves 97.38% accuracy and 0.98 AUC in CVD detection, outperforming state-of-the-art methods and demonstrating robust performance for real-world clinical applications. 1 Introduction Cardiovascular disease (CVD), the leading global cause of morbidity and mortality [ 2 ] , can be effectively managed through early detection and precise diagnosis [ 10 ] . Electrocardiogram (ECG) and Phonocardiogram (PCG) are widely utilized non-invasive diagnostic modalities, with ECG providing insights into cardiac electrical activity [ 5 ] and PCG capturing heart sounds indicative of mechanical function [ 20 ] . However, relying on a single modality often leads to incomplete diagnostic information, thereby increasing the risk of misdiagnosis [ 15 , 9 , 1 ] . As a consequence, multi-modal approaches that integrate ECG and PCG have been developed, effectively combining their complementary features to enhance diagnostic accuracy and reliability [ 18 , 13 , 11 ] . Figure 1: Illustration of the relationship between ECG, PCG, and the coupling signal. ECG, PCG, and the coupling signal serve as three distinct yet complementary modalities for CVD detection. ECG captures the heart’s electrical activity, while PCG reflects its mechanical response. The coupling signal characterizes the transformation from electrical excitation to mechanical contraction, and PCG can be interpreted as the convolution of ECG and the coupling signal. By leveraging these interrelated modalities, the multi-modal approach provides a more comprehensive representation of cardiac function. The relationship between ECG and PCG is governed by the electromechanical coupling of the heart [ 23 ] . The electrical signals (ECG) initiate mechanical contractions that produce the heart sounds (PCG). This relationship can be modeled as PCG being the convolution of ECG with a latent coupling signal that characterizes the electrical-to-mechanical transformation [ 7 ] . Incorporating this coupling signal as a third modality can enhance diagnostic performance by explicitly modeling cross-modal dynamics [ 22 ] . However, conventional coupling signal estimation methods rely on deconvolution [ 22 , 21 ] , an ill-posed process that amplifies noise in clinical environments. Environmental noise during PCG acquisition renders these approaches unreliable in real-world settings, limiting their practical utility. To address this challenge, we propose Noise-Robust Multi-Modal Coupling Signal Estimation (NMCSE), which reformulates the problem as distribution matching via optimal transport theory. This approach: (1) operates at the distribution level to incorporate global signal characteristics; (2) circumvents ill-posed inverse operations; and (3) jointly optimizes amplitude and temporal alignment. We integrate NMCSE with our Temporal-Spatial Feature Extraction network for robust multi-modal CVD detection. Experiments on the PhysioNet 2016 dataset with simulated hospital noise demonstrate that NMCSE reduces estimation errors by approximately 30% compared to conventional methods and maintains higher correlation across all tested signal-to-noise ratios. Our approach achieves 97.38% accuracy and 0.98 AUC in CVD detection, outperforming state-of-the-art methods in noisy conditions. 2 Related Work CVD detection has progressed from single-modality to multi-modal approaches that integrate ECG and PCG signals, evolving through three main stages: Feature-level fusion methods extract features independently before combination. Li et al. [ 18 ] used specialized networks with genetic algorithm optimization, while Hettiarachchi et al. [ 12 ] employed transfer learning on scalograms. Decision-level fusion methods, as proposed by Li et al. [ 17 ] , use evidence theory to weight contributions from each modality based on confidence levels. End-to-end deep learning approaches perform joint representation learning, with Li et al. [ 16 ] utilizing a multi-input CNN framework and Zhang et al. [ 24 ] developing CPDNet with modality-specific encoders. These approaches largely treat ECG and PCG as independent information sources. Sun et al. [ 22 ] made a significant advance by incorporating coupling signals that explicitly model the electromechanical relationship between ECG and PCG, demonstrating improved diagnostic accuracy. However, their deconvolution-based estimation method is limited by noise amplification, compromising reliability in realistic clinical settings. Our NMCSE method addresses this critical limitation by providing noise-robust coupling signal estimation, enabling reliable multi-modal CVD detection even in noisy clinical environments. 3 Noise-Robust Multi-Modal Coupling Signal Estimation: Theory and Algorithm This section establishes the mathematical foundation for applying optimal transport theory to ECG-PCG coupling signal estimation. We develop a comprehensive theoretical framework that addresses the limitations of conventional approaches and provides rigorous error analysis with convergence guarantees. 3.1 Mathematical Model of Cardiac Electromechanical Coupling We begin by formalizing the relationship between cardiac electrical and mechanical activities within a mathematically rigorous framework. Theorem 3.1 (ECG-PCG Transformation Model) There exists a linear time-invariant system characterized by impulse response h(t) such that, under ideal conditions, the PCG signal X_{\mathrm{PCG}}(t) can be expressed as the convolution of the ECG signal X_{\mathrm{ECG}}(t) with h(t) : X_{\mathrm{PCG}}(t)=(X_{\mathrm{ECG}}*h)(t) (1) In realistic clinical environments with noise, the observed PCG signal becomes: \tilde{X}_{\mathrm{PCG}}(t)=(X_{\mathrm{ECG}}*h)(t)+\epsilon(t) (2) where \epsilon(t) represents additive noise from various sources. 3.2 Limitations of Deconvolution in Noisy Environments Theorem 3.2 (Error Amplification in Deconvolution) Let H_{t}(\omega) be the true frequency response of the coupling system and H_{a}(\omega) be the estimated response using standard deconvolution. In the presence of noise with spectrum E(\omega) , the estimation error satisfies: \|H_{a}-H_{t}\|_{2}^{2}=\int_{-\infty}^{+\infty}\left|\frac{E(\omega)}{X_{\mathrm{ECG}}(\omega)}\right|^{2}d\omega (3) For any \delta>0 , if there exists a frequency set \Omega_{\delta}=\{\omega:|X_{\mathrm{ECG}}(\omega)|<\delta\} with non-zero measure, then the error is lower-bounded by: \|H_{a}-H_{t}\|_{2}^{2}>\frac{1}{\delta^{2}}\int_{\Omega_{\delta}}|E(\omega)|^{2}d\omega (4) Proof (Proof Sketch) From the frequency domain representation of the noisy PCG signal \tilde{X}_{\mathrm{PCG}}(\omega)=X_{\mathrm{ECG}}(\omega)H_{t}(\omega)+E(\omega) , the deconvolution estimate becomes H_{a}(\omega)=H_{t}(\omega)+\frac{E(\omega)}{X_{\mathrm{ECG}}(\omega)} . The error term \frac{E(\omega)}{X_{\mathrm{ECG}}(\omega)} is amplified when X_{\mathrm{ECG}}(\omega) approaches zero, resulting in the error bound above. This theorem mathematically proves why deconvolution fails in realistic clinical scenarios: even small amounts of noise can cause unbounded errors when the ECG signal has low power in certain frequency bands, which is unavoidable in physiological signals. 3.3 Distribution-Based Formulation via Optimal Transport To address the limitations of deconvolution, we reformulate coupling signal estimation as a distribution matching problem solved through optimal transport theory. Instead of operating in the traditional frequency domain with point-wise divisions, we optimize a parameterized coupling filter by minimizing the discrepancy between distributions: h_{\theta}^{*}=\arg\min_{h_{\theta}}\mathcal{D}(P_{\mathrm{ECG}*h_{\theta}},P_{\mathrm{PCG}}) (5) where h_{\theta} is a parameterized coupling filter, P_{\mathrm{ECG}*h_{\theta}} represents the distribution of the ECG signal transformed by the filter, P_{\mathrm{PCG}} is the distribution of the PCG signal, and \mathcal{D} is a distribution divergence measure. For this application, the Wasserstein distance from optimal transport theory is particularly well-suited due to three key advantages: W_{p}(P_{1},P_{2})=\left(\inf_{\gamma\in\Gamma(P_{1},P_{2})}\int_{\mathcal{X}\times\mathcal{X}}d(x,y)^{p}d\gamma(x,y)\right)^{1/p} (6) First, it naturally accounts for the underlying metric structure of the signal space, incorporating

Read the complete text