Benchmark RadarECG-InterpBench
ECG-InterpBench evaluates interpretability of ECG foundation models using sparse autoencoders with matched capacity, measuring reconstruction fidelity, clinical feature accessibility, and cross-seed reproducibility across 450 cells.
- Released
- 2026-07-29
- Readiness
- Paper only
- Primary field
- Health & Life Sciences
Why it matters
Performance-focused ECG benchmarks ignore whether representations are interpretable. This benchmark provides a controlled, reproducible framework for comparing models on interpretability, aiding clinical adoption where understanding model decisions is critical.
Motivation
Existing benchmarks for electrocardiogram foundation models primarily evaluate downstream predictive performance, providing limited insight into whether their internal representations can be faithfully decomposed, clinically interpreted, or reproduced across independent analyses.
Primary resources
Benchmark Radar records only publicly supported details and links back to primary sources for verification.