Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Detecting Structural Heart Disease from Electrocardiograms via a Generalized Additive Model of Interpretable Foundation-Model Predictors

Ya Zhou, Zhaohong Sun, Tianxiang Hao, Xiangjie Li

TL;DR

This work illustrates a complementary paradigm between classical statistical modeling and modern AI, offering a pathway to interpretable, high-performing, and clinically actionable ECG-based SHD screening.

Abstract

Structural heart disease (SHD) is a prevalent condition with many undiagnosed cases, and early detection is often limited by the high cost and accessibility constraints of echocardiography (ECHO). Recent studies show that artificial intelligence (AI)-based analysis of electrocardiograms (ECGs) can detect SHD, offering a scalable alternative. However, existing methods are fully black-box models, limiting interpretability and clinical adoption. To address these challenges, we propose an interpretable and effective framework that integrates clinically meaningful ECG foundation-model predictors within a generalized additive model, enabling transparent risk attribution while maintaining strong predictive performance. Using the EchoNext benchmark of over 80,000 ECG-ECHO pairs, the method demonstrates relative improvements of +0.98% in AUROC, +1.01% in AUPRC, and +1.41% in F1 score over the latest state-of-the-art deep-learning baseline, while achieving slightly better performance even with only 30% of the training data. Subgroup analyses confirm robust performance across heterogeneous populations, and the estimated entry-wise functions provide interpretable insights into the relationships between risks of traditional ECG diagnoses and SHD. This work illustrates a complementary paradigm between classical statistical modeling and modern AI, offering a pathway to interpretable, high-performing, and clinically actionable ECG-based SHD screening.

Detecting Structural Heart Disease from Electrocardiograms via a Generalized Additive Model of Interpretable Foundation-Model Predictors

TL;DR

This work illustrates a complementary paradigm between classical statistical modeling and modern AI, offering a pathway to interpretable, high-performing, and clinically actionable ECG-based SHD screening.

Abstract

Structural heart disease (SHD) is a prevalent condition with many undiagnosed cases, and early detection is often limited by the high cost and accessibility constraints of echocardiography (ECHO). Recent studies show that artificial intelligence (AI)-based analysis of electrocardiograms (ECGs) can detect SHD, offering a scalable alternative. However, existing methods are fully black-box models, limiting interpretability and clinical adoption. To address these challenges, we propose an interpretable and effective framework that integrates clinically meaningful ECG foundation-model predictors within a generalized additive model, enabling transparent risk attribution while maintaining strong predictive performance. Using the EchoNext benchmark of over 80,000 ECG-ECHO pairs, the method demonstrates relative improvements of +0.98% in AUROC, +1.01% in AUPRC, and +1.41% in F1 score over the latest state-of-the-art deep-learning baseline, while achieving slightly better performance even with only 30% of the training data. Subgroup analyses confirm robust performance across heterogeneous populations, and the estimated entry-wise functions provide interpretable insights into the relationships between risks of traditional ECG diagnoses and SHD. This work illustrates a complementary paradigm between classical statistical modeling and modern AI, offering a pathway to interpretable, high-performing, and clinically actionable ECG-based SHD screening.
Paper Structure (22 sections, 12 equations, 2 figures, 4 tables)

This paper contains 22 sections, 12 equations, 2 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Model
  3. Classical statistical modeling: interpretable but limited expressive capacity
  4. Pure deep learning: accurate but difficult to interpret
  5. A hybrid modeling framework with interpretable latent predictors
  6. Methodology
  7. ECG foundation model
  8. Interpretable predictors and post-training
  9. Nonparametric estimation
  10. Experiments
  11. Datasets
  12. Traditional ECG dataset
  13. ECG-ECHO Dataset
  14. Experimental settings
  15. Post-training for the ECG foundation model
  16. ...and 7 more sections

Figures (2)

  • Figure 1: ROC and PR curves comparison of four different models. The left panel shows ROC curves with AUROC values, while the right panel displays PR curves with AUPRC values. Models marked with "(our)" indicate models that use predictors derived from the ECG foundation model employed in this study.
  • Figure 2: Examples of estimated entry-wise functions in the proposed additive model. "Partial Effect" refers to $\hat{f}_j(x)$ defined in \ref{['def:f_hat']}, and "Predictive Value" refers to the corresponding $\sigma\{\hat{h}_j(\mathbf X)\}$, where $\hat{h}_j(\mathbf X)$ is defined in \ref{['eqn:def_hat_h']}. The titles IMI, NDT, AFIB and LVH denote inferior myocardial infarction, non-diagnostic T abnormalities, atrial fibrillation, and left ventricular hypertrophy, respectively.