Jove
Visualize
Contact Us
JoVE
x logofacebook logolinkedin logoyoutube logo
ABOUT JoVE
OverviewLeadershipBlogJoVE Help Center
AUTHORS
Publishing ProcessEditorial BoardScope & PoliciesPeer ReviewFAQSubmit
LIBRARIANS
TestimonialsSubscriptionsAccessResourcesLibrary Advisory BoardFAQ
RESEARCH
JoVE JournalMethods CollectionsJoVE Encyclopedia of ExperimentsArchive
EDUCATION
JoVE CoreJoVE BusinessJoVE Science EducationJoVE Lab ManualFaculty Resource CenterFaculty Site
Terms & Conditions of Use
Privacy Policy
Policies

Related Experiment Video

Updated: Jun 21, 2026

Real-Time Cardiac Mapping with a Noninvasive Imageless Electrocardiographic Imaging System
10:17

Real-Time Cardiac Mapping with a Noninvasive Imageless Electrocardiographic Imaging System

Published on: April 11, 2025

Differentiating Ischemic From Nonischemic T-Wave Inversion Using a Multimodal Vision-Language Model With

Yunzhang Cheng1, Zhongkai Wang2, Wen Zhang1

  • 1School of Health Science and Engineering, University of Shanghai for Science and Technology, Room 1002, 10th Floor, Zhuoyue Building, No. 334 Jungong Road, Shanghai, Yangpu District, 200093, China, +86 18566664556.

JMIR Medical Informatics
|June 19, 2026
PubMed
Summary

Related Concept Videos

You might also read

Related Articles

Articles linked to this work by shared authors, journal, and citation graph.

Sort by
Same author

Intelligent Prediction Platform for Sepsis Risk Based on Real-Time Dynamic Temporal Features: Design Study.

JMIR medical informatics·2025
Same author

Optimizing sepsis treatment strategies via a reinforcement learning model.

Biomedical engineering letters·2024
See all related articles

A novel AI framework, ECG-R1, accurately differentiates ischemic from nonischemic T-wave inversion (TWI) on ECGs using reinforcement learning. This approach improves diagnostic accuracy and provides interpretable reasoning for clinical decision support.

Area of Science:

  • Artificial Intelligence in Cardiology
  • Machine Learning for Medical Diagnostics
  • Multimodal Medical Data Analysis

Background:

  • Differentiating ischemic from nonischemic T-wave inversion (TWI) on electrocardiograms (ECGs) is a critical diagnostic challenge.
  • Visual interpretation of TWI has low positive predictive value (~50%), leading to high false-positive rates and unnecessary procedures.
  • Existing deep learning models for TWI lack multimodality and transparency, limiting clinical trust and adoption.

Purpose of the Study:

  • To develop a novel diagnostic framework for accurately differentiating ischemic from nonischemic TWI.
  • To leverage a multimodal vision-language model trained with reinforcement learning (RL) for improved accuracy and interpretable reasoning.
  • To address the limitations of current AI models in processing complex data and providing transparent clinical insights.
Keywords:
artificial intelligenceAIECGT-wave inversioncomputer-aided diagnosiselectrocardiogrammultimodal artificial intelligencemyocardial ischemiareinforcement learningvision-language models

Related Experiment Videos

Last Updated: Jun 21, 2026

Real-Time Cardiac Mapping with a Noninvasive Imageless Electrocardiographic Imaging System
10:17

Real-Time Cardiac Mapping with a Noninvasive Imageless Electrocardiographic Imaging System

Published on: April 11, 2025

Main Methods:

  • Developed ECG-R1, a multimodal framework using the Qwen2-VL-2B vision-language model, analyzing ECG images and clinical text.
  • Employed an RL paradigm with group relative policy optimization, avoiding supervised fine-tuning (SFT).
  • Trained the model to generate structured outputs with explicit reasoning traces and diagnostic answers, using a rule-based reward function for accuracy and format adherence.

Main Results:

  • ECG-R1 achieved state-of-the-art in-domain accuracy (75.21%) and AUC (84.18%) on a dataset of 12,917 TWI cases.
  • Demonstrated robust cross-hospital generalization with 72.93% out-of-domain accuracy and 81.56% AUC.
  • RL paradigm significantly improved performance (6.69% in-domain, 11.48% out-of-domain) over SFT baselines, indicating superior learning of physiological features.

Conclusions:

  • The RL-based ECG-R1 framework significantly outperformed SFT baselines in diagnostic accuracy and robustness.
  • ECG-R1 offers a transparent clinical decision support system by modeling interpretable reasoning and using probabilistic language.
  • The framework is designed for safe integration into clinical workflows via a human-in-the-loop paradigm, paving the way for trials.