Research map / Time series / Foundation models for time series
HeartLLM: Discretized ECG Tokenization for LLM-Based Diagnostic Reasoning
Jinning Yang, Wenjie Sun, Wen Shi
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 |
|---|---|---|---|
| discretize ECG into FSQ tokens (vs continuous embeddings) (ablation) | HeartLLM · Zero-shot ECG-QA QA-Average, MIMIC-IV-ECG to PTB-XL | EM accuracy (%): 35.77 -> 40.71 | Improved |
| use lead-wise ResIncept encoder instead of MultiCNN (ablation) | HeartLLM · ECG-QA QA-Query (MIMIC-IV-ECG) | EM accuracy (%): 16.71 -> 22.05 | Improved |
| use lead-wise ResIncept encoder instead of CausalCNN (ablation) | HeartLLM · ECG-QA QA-Query (PTB-XL) | EM accuracy (%): 29.45 -> 32.55 | Improved |
| use lead-wise encoder instead of independent U-Net 1D (ablation) | HeartLLM · ECG-QA QA-Choose (PTB-XL) | EM accuracy (%): 57.02 -> 59.74 | Improved |
| use lead-wise encoder instead of Unet2D (ablation) | HeartLLM · ECG-QA QA-Choose (PTB-XL) | EM accuracy (%): 59.49 -> 59.74 | No clear effect |
| use HeartLLM versus best baseline | HeartLLM · ECG-QA QA-Query (MIMIC-IV-ECG) | EM accuracy (%): 15.24 -> 22.05 | Improved |
| use HeartLLM versus strongest baseline | HeartLLM · ECG-Report BLEU-4 (MIMIC-IV-ECG) | BLEU-4 (%): 20.60 -> 29.74 | Improved |
| use HeartLLM versus ST-MEM on zero-shot QA-Choose | HeartLLM · Zero-shot PTB-XL QA-Choose | EM accuracy (%): 41.08 (ST-MEM) vs 39.56 | Hurt |
| remove LoRA fine-tuning, remove tabular features (ablations) (ablation) | HeartLLM · PTB-XL ECG-QA/Report | EM accuracy / BLEU | Hurt |
Closest papers by meaning
- ECG-Reasoning-Benchmark: A Benchmark for Evaluating Clinical Reasoning Capabilities in ECG Interpretation (NeurIPS 2026 · similarity 0.90)
- Reading Your Heart: Learning ECG Words and Sentences via Pre-training ECG Language Model (ICLR 2025 · similarity 0.89)
- ECG-R1: Protocol-Guided and Modality-Agnostic MLLM for Reliable ECG Interpretation (ICML 2026 · similarity 0.88)
- From Token to Rhythm: A Multi-Scale Approach for ECG-Language Pretraining (ICML 2025 · similarity 0.88)
- GEM: Empowering MLLM for Grounded ECG Understanding with Time Series and Images (NeurIPS 2025 · 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.
