EUVIMEDEUVIMED entdecken
Open Access · CC BY-NC-ND 4.0

Cardiologists’ Viewpoints on Clinically Useful AI-ECG: What ECG Signals Carry, Where Current Models Fall Short, and Six Principles

Yen-Chen Lin, Gen-Min Lin

Mayo Clinic Proceedings: Digital Health · 2026-11 · 100436

DOI: 10.1016/j.mcpdig.2026.100436 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Mayo Clinic Proceedings: Digital Health, 2026-11. Verfasst von Yen-Chen Lin und 1 weiteren Autor:innen. Die Arbeit nennt 52 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY-NC-ND 4.0). Gefördert von: Ministry of National Defense Republic of China Taiwan Medical Affairs Bureau, Hualien Armed Forces General Hospital.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

A model that reads disease or risk from the electrocardiogram (ECG) by artificial intelligence is called an AI-ECG. Many AI-ECG models for diseases or outcomes not directly defined by the ECG admitted recordings without eligibility rules tailored to the clinical question, had a thin clinical comparator or none, and were judged mainly by discrimination. The ECG signal is a sample of one moment, transmitted through a body and a recording chain and obtained for a reason, and a clinician reads it against a relatively stable clinical picture and earlier recordings. We measure current models against what the signal carries and how the ECG is used, and propose six principles. A strong and relatively stable clinical risk model, built from integrated clinical data, is the foundation, whose risk the AI-ECG refines rather than replaces. As a physician compares a repeat ECG with earlier ones, the AI-ECG is better designed for serial comparison and combined interpretation of a patient's recordings. The body is both a set of risk factors and the medium the signal crosses, two roles kept apart. Training recordings are chosen by the clinical question and, because the signal carries much that a model can learn in place of the heart, admitted carefully to minimize confounded learning. The AI-ECG must agree with itself on a stable patient and hold where it is used. Its performance and incremental value are judged for patients, beyond discrimination and within subgroups. Four testable suggestions follow. We introduce concepts, and no model has been trained.

Förderung

  • Ministry of National Defense Republic of China Taiwan Medical Affairs Bureau
  • Hualien Armed Forces General Hospital

Zitieren

APA

Lin, Y., & Lin, G. (2026). Cardiologists’ Viewpoints on Clinically Useful AI-ECG: What ECG Signals Carry, Where Current Models Fall Short, and Six Principles. Mayo Clinic Proceedings: Digital Health, 100436. https://doi.org/10.1016/j.mcpdig.2026.100436

Vancouver

Lin Y, Lin G. Cardiologists’ Viewpoints on Clinically Useful AI-ECG: What ECG Signals Carry, Where Current Models Fall Short, and Six Principles. Mayo Clinic Proceedings: Digital Health. 2026:100436. doi:10.1016/j.mcpdig.2026.100436

BibTeX
@article{Lin2026,
  author = {Lin, Yen-Chen and Lin, Gen-Min},
  title = {Cardiologists’ Viewpoints on Clinically Useful AI-ECG: What ECG Signals Carry, Where Current Models Fall Short, and Six Principles},
  journal = {Mayo Clinic Proceedings: Digital Health},
  year = {2026},
  pages = {100436},
  doi = {10.1016/j.mcpdig.2026.100436}
}
Kostenlos starten →