EchoBridge: Long-Tail-Aware ECG-Echocardiography Text Alignment for Echocardiography-Derived Cardiac Findings

2026-07-27

AI Summary

This work addresses the challenges of modality entanglement and supervision sparsity for low-frequency cardiac pathologies arising from long-tailed distributions in ECG–echocardiography text alignment. To this end, the authors propose a shared-private complementary projection mechanism coupled with an adaptive prototype boundary calibration strategy. By constructing orthogonally constrained bimodal projections on a normalized hypersphere, enforcing bidirectional alignment, incorporating frequency-adaptive angular margins, and applying spherical Riesz repulsion, the method effectively disentangles modality-specific factors and enhances fine-grained semantic alignment. Evaluated under multiple protocols, the approach significantly outperforms baseline models, achieving absolute improvements of 7.88, 5.61, and 4.54 points in AUROC, AUPRC, and F1 score, respectively, while demonstrating consistent advantages in cross-center transferability and recognition of rare valvular lesions.

Abstract

Standardized echocardiography conclusions provide meaningful supervision for learning ECG representations of echocardiography-derived cardiac findings. Global ECG--text alignment may entangle modality-specific factors, while long-tailed finding distributions provide sparse positive supervision for low-prevalence conditions. We propose EchoBridge with Complementary Shared--Private Projection (CSPP) and Adaptive Prototype Boundary Calibration (APBC). CSPP maps each modality into shared and auxiliary private projections, reduces directional redundancy via within-modality orthogonality, and bidirectionally aligns normalized shared projections. APBC organizes the shared hypersphere with class-specific prototypes, training-frequency-adaptive angular margins, and spherical Riesz repulsion. We evaluate EchoBridge on EchoNext-Mini and independent PKUPH and SHTMU cohorts under four protocols: prompt-based inference without downstream classifier training, in-domain frozen linear probing, target-domain cross-center frozen linear probing, and source-only cross-center transfer, supplemented by finding-specific analyses. EchoBridge improves classifier-free AUROC, AUPRC, and F1 over the strongest baselines by 7.88, 5.61, and 4.54 points, respectively, and achieves the highest point estimates across all in-domain and target-domain probing budgets and both source-only transfer cohorts. Finding-specific analyses show gains for most conditions, including several low-prevalence valvular findings.

Problem

• ECG-echocardiography alignment
• long-tail distribution
• cardiac findings
• modality-specific factors
• sparse supervision

Innovation

• ECG-echocardiography alignment
• long-tail learning
• shared-private projection
• prototype calibration
• cross-modal representation