Search papers, labs, and topics across Lattice.
This study introduces EchoBridge, a novel framework that enhances ECG representations by aligning them with echocardiography-derived cardiac findings, addressing the challenges posed by long-tailed distributions of low-prevalence conditions. Utilizing Complementary Shared鈥揚rivate Projection (CSPP) and Adaptive Prototype Boundary Calibration (APBC), the method effectively reduces modality-specific entanglement and organizes class-specific prototypes to improve classification performance. The results demonstrate significant improvements in classifier-free AUROC, AUPRC, and F1 scores over existing baselines, particularly benefiting low-prevalence findings.
EchoBridge boosts classification performance by nearly 8 points in AUROC, especially for rare cardiac conditions that are often overlooked.
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.
