Research map / Label-efficient and robust learning / Recognition and long-tail
Benchmarking ECG Foundational Models: A Reality Check Across Clinical Tasks
M A Al-Masud, Juan Lopez Alcaraz, Nils Strodthoff
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use S4 backbone instead of Net1D CNN in supervised training | supervised baseline · PTB-XL (all) | macro AUROC: 0.929 -> 0.941 | Improved |
| use ECG-CPC SSM foundation model instead of supervised S4 | finetuning · PTB-XL (all) | macro AUROC: 0.941 -> 0.949 | Improved |
| use ECG-CPC instead of supervised S4 | finetuning · EchoNext (Echo) | macro AUROC: 0.819 -> 0.831 | Improved |
| use ECG-FM instead of supervised S4 | finetuning · MIMIC (Age) | standardized MAE: 0.455 -> 0.412 | Improved |
| use ST-MEM foundation model instead of supervised S4 | finetuning · MIMIC (Age) | standardized MAE: 0.455 -> 0.504 | Hurt |
| pretrain HuBERT-ECG with masked language modeling at 9.1M samples instead of S4 from scratch | finetuning · MIMIC (Age) | standardized MAE: 0.455 -> 0.579 | Hurt |
| pretrain with ECG-JEPA instead of training S4 from scratch (label efficiency) | EchoNext scaling · EchoNext | label efficiency ratio: 0.11-0.42 | Improved |
| pretrain with ECG-CPC instead of S4 from scratch (label efficiency) | EchoNext scaling · EchoNext | label efficiency ratio: 0.21-0.40 | Improved |
| pretrain with ECGFounder instead of S4 from scratch (label efficiency) | EchoNext scaling · EchoNext | label efficiency ratio: 0.30-0.62 | Improved |
| add layer-dependent learning rates in finetuning (ablation) | finetuned FMs · PTB-XL, EchoNext, CPSC-Extra | performance | Improved |
| use 2.5-second crops with test-time averaging instead of full 10-second inputs (ablation) | finetuning · ECG benchmark | performance | Improved |
| use linear head instead of learnable query-attention head in frozen evaluation for ECG-CPC | ECG-CPC · ECG benchmark | ranking | Hurt |
Closest papers by meaning
- ECG-Reasoning-Benchmark: A Benchmark for Evaluating Clinical Reasoning Capabilities in ECG Interpretation (NeurIPS 2026 · similarity 0.89)
- From Token to Rhythm: A Multi-Scale Approach for ECG-Language Pretraining (ICML 2025 · similarity 0.89)
- GEM: Empowering MLLM for Grounded ECG Understanding with Time Series and Images (NeurIPS 2025 · similarity 0.88)
- Extending Pretrained 10-Second ECG Foundation Models to Longer Horizons (NeurIPS 2026 · similarity 0.88)
- ECG-R1: Protocol-Guided and Modality-Agnostic MLLM for Reliable ECG Interpretation (ICML 2026 · similarity 0.87)
- anyECG-chat: A Generalist ECG-MLLM for Flexible ECG Input and Multi-Task Understanding (AAAI 2026 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
