PaperarXiv:2611.18809
LEAP-LLM: Preserving lead-time structure for question-conditioned ECG interpretation
arXiv preprint (cs.LG; cs.AI)
TL;DR
The clinical question steers ECG feature extraction; a Lead branch keeps per-lead temporal patterns and a Temporal branch summarises across leads at each time step, so the LLM gets compact, question-relevant ECG tokens. Trained on ECGInstruct (1.16M pairs) + ECG-Grounding (30K); evaluated on four ECG-Bench sets: PTB-XL and CODE-15% (multi-label; macro AUC/F1/Hamming loss) and G12EC and CSN (multiple choice).
Abstract
The electrocardiogram (ECG) records cardiac electrical activity across multiple leads and is used for detecting cardiac abnormalities. Recent ECG-language models enable large language models (LLMs) to interpret cardiac signals and answer clinically relevant questions. However, existing ECG-language models typically encode the ECG independently of the clinical question, forcing the ECG representation to retain broad and potentially irrelevant signal information. Moreover, these models often aggregate information across leads which can obscure lead-specific details needed for a particular question. To tackle these limitations, we introduce LEAP-LLM, an ECG-LLM architecture for Lead-aware ECG Alignment with Prompt conditioning. LEAP-LLM uses the clinical question to guide ECG feature extraction while preserving explicit lead and temporal structure. Task-trained instruction representations distill information from the question and guide two complementary branches. The Lead branch captures question-relevant temporal patterns within each ECG lead, while the Temporal branch summarizes relevant information across leads at each temporal position. Together, they provide the LLM with compact, question-relevant ECG information while preserving lead and temporal structure. LEAP-LLM reduces modality-token count by 89.0% and prefill FLOPs by 7.2x on average relative to other ECG-LLMs baselines. Across four ECG benchmarks, LEAP-LLM achieves better performance, including a 3.4-percentage-point improvement in G12EC over the strongest prior ECG-specific method. We further show that the model adaptively emphasizes the clinically relevant anatomical lead groups for different cardiac abnormalities, with lead importance patterns consistent with established diagnostic knowledge, providing interpretable evidence that the predictions are supported by clinically meaningful ECG structure.
Verbatim from arXiv (opens in a new tab) · arXiv metadata (abstracts are released under CC0)
Results on ChatPPG leaderboards
All leaderboardsNo results from this paper on ChatPPG leaderboards yet. Browse the leaderboards.
Tasks & signals
Signals
Keywords
Datasets used
How it works
Cite this paper
Agarwal, A., Shen, Y., Fu, X., & Hu, R. (2026). LEAP-LLM: Preserving lead-time structure for question-conditioned ECG interpretation. arXiv preprint. https://arxiv.org/abs/2611.18809
Agarwal, Aakash, Yidan Shen, Xin Fu, and Renjie Hu. 2026. "LEAP-LLM: Preserving lead-time structure for question-conditioned ECG interpretation." arXiv preprint. https://arxiv.org/abs/2611.18809.
Agarwal, A., Shen, Y., Fu, X. and Hu, R. (2026) 'LEAP-LLM: Preserving lead-time structure for question-conditioned ECG interpretation', arXiv preprint. Available at: https://arxiv.org/abs/2611.18809
Agarwal, Aakash, et al. "LEAP-LLM: Preserving lead-time structure for question-conditioned ECG interpretation." arXiv preprint, 2026. https://arxiv.org/abs/2611.18809.
A. Agarwal, Y. Shen, X. Fu, and R. Hu, "LEAP-LLM: Preserving lead-time structure for question-conditioned ECG interpretation," arXiv preprint, 2026, [Online]. Available: https://arxiv.org/abs/2611.18809.
@misc{agarwal2026leapllm,
title = {{LEAP-LLM: Preserving lead-time structure for question-conditioned ECG interpretation}},
author = {Agarwal, Aakash and Shen, Yidan and Fu, Xin and Hu, Renjie},
year = {2026},
howpublished = {arXiv preprint},
eprint = {2611.18809},
archivePrefix = {arXiv}
}Added to the ChatPPG research hub