Beyond Machine Interpretation: Learning from Expert Over-Reads Improves ECG Diagnosis

Sunwoo Kwak, Fengbei Liu, Nusrat Binta Nizam, Ilan Richter, Nir Uriel, Peter M. Okin, Mert R. Sabuncu

Keywords: ECG, Expert Over-read, Self-training, Contrastive learning, Clinical AI

TL;DR: Over-read ECG Model

Abstract: Automated machine-read ECG interpretations are widely used in clinical practice but often unreliable, leading to systematic diagnostic errors. This work investigates how training with cardiologist over-reads impacts model accuracy and clinical reliability. Using a large paired corpus of over two million ECGs containing both machine and expert interpretations, we evaluate three learning paradigms: (i) supervised learning on expert over-read labels, (ii) Self-training that extends expert supervision to public ECGs, and (iii) multi-modal contrastive learning with CLIP and NegCLIP. Across all settings, models trained with expert over-read data consistently outperform those trained on machine-read labels, especially for rare but clinically important conditions. Self-training and NegCLIP further demonstrate scalable strategies to propagate expert knowledge beyond labeled datasets. These findings highlight the essential role of expert over-reads in developing trustworthy and clinically aligned ECG AI systems. Code and model can be found at this link : https://github.com/tyoung089/Overread_ecg

Primary Subject Area: Application: Cardiology

Secondary Subject Area: Detection and Diagnosis

Registration Requirement: Yes

Reproducibility: https://github.com/tyoung089/Overread_ecg

Visa & Travel: Yes

Read CFP & Author Instructions: Yes

Originality Policy: Yes

Single-blind & Not Under Review Elsewhere: Yes

LLM Policy: Yes

Midl Latex Submission Checklist: Ensure no LaTeX errors during compilation., Replace NNN with your OpenReview submission ID., Includes \documentclass{midl}, \jmlryear{2026}, \jmlrworkshop, \jmlrvolume, \editors, and correct \bibliography command., Did not override options of the hyperref package., Did not use the times package., Use the correct spelling and format, avoid Unicode characters, and use LaTeX equivalents instead., Any math in the title and abstract must be enclosed within $...$., Did not override the bibliography style defined in midl.cls and did not use \begin{thebibliography} directly to insert references., Avoid using \scalebox; use \resizebox when needed., Included all necessary figures and removed *unused* files in the zip archive., Removed special formatting, visual annotations, and highlights used during rebuttal., All special characters in the paper and .bib file use LaTeX commands (e.g., \'e for é)., No separate supplementary PDF uploads., Acknowledgements, references, and appendix must start after the main content.

Submission Number: 7