Skip to content
DraconiticResearch
Search Sources API
226,742 studies31 sourcesUpdated 36 min ago
Peer-reviewedCC BY 4.0Nature Cardiovascular Research

A multiview video–text model detects obstructive coronary stenosis and predicts 1-year cardiovascular events from angiography

10 affiliations
  1. 1Division of Cardiology, Department of Medicine, Montreal Heart Institute, Université de Montréal, Montreal, Quebec Canada
  2. 2HeartWise.ai, Montreal, Quebec Canada
  3. 3Faculty of Engineering, McGill University, Montreal, Quebec Canada
  4. 4Department of Medicine, Université de Montréal, Montreal, Quebec Canada
  5. 5Division of Cardiology, Department of Medicine, University of California, San Francisco, CA USA
  6. 6Cardiovascular Research Institute, University of California, San Francisco, CA USA
  7. 7Bakar Computational Health Sciences Institute, University of California, San Francisco, CA USA
  8. 8Department of Cardiology, Smidt Heart Institute, Cedars-Sinai Medical Center, Los Angeles, CA USA
  9. 9Department of Bioengineering, University of California Los Angeles, Los Angeles, CA USA
  10. 10Department of Medicine, Kaiser Permanente, San Francisco, CA USA

Abstract

Invasive coronary angiography is the reference standard for diagnosing coronary artery disease, yet assessment is subjective, and existing artificial-intelligence tools typically assess stenosis in individual frames or videos and do not integrate the full multiview angiographic study. Here we show that DeepCORO-CLIP, trained on 203,808 angiographic videos and reports from 28,117 patients, interprets whole multiple-video angiograms at once. It detects obstructive stenoses (≥70%), with an area under the receiver operating characteristic curve (AUROC) of 0.89 internally and externally, and identifies chronic total occlusion (AUROC = 0.90 to 0.95), intra-coronary thrombus and calcification (AUROC = 0.90 to 0.91). It also predicts 1-year cardiovascular events (AUROC = 0.79, 95% CI of 0.74 to 0.84, versus 0.65 for a model derived using clinical variables) and left ventricular ejection fraction (mean absolute error 7.3% versus 8.8% for a previous model), both achieved using coronary angiography alone. In a blinded review of 365 evaluations across 288 angiograms, its reports were clinically usable in 76% of evaluations versus 88% for cardiologists, with comparable false positives (17% versus 14%, P = 0.38); residual missed stenoses reflected an adjustable threshold rather than an absent signal.

Related

  1. Nonlinear and Vessel-Specific Association Between Coronary Artery Calcium Burden and Coronary Computed Tomography Angiography Underestimation of Coronary Stenosis
    Xuzhe Wang, Chengzhuo Wang, Yu Xie +2
    Reviews in cardiovascular medicinePeer-reviewedCC BYFull textCardiovascular medicineEurope PMC
  2. Incremental diagnostic value of coronary CT angiography-derived pericoronary adipose tissue radiomics for identifying hemodynamically significant coronary stenosis
    Wei Fan, Hong-Xuan Wang, Bai-Hui Zhang +7
    Frontiers in cardiovascular medicinePeer-reviewedCC BYFull textCardiovascular medicineEurope PMC
  3. Combined non-contrast enhancement 3D bright-blood and 2D black-blood coronary magnetic resonance angiography for the assessment of coronary artery stenosis
    Qiu-Ju Hu, Di Tian, Dong-Sheng Jin +7
    Quantitative imaging in medicine and surgeryPeer-reviewedCC BY-NC-NDFull textMedical imaging & radiologyEurope PMC
  4. Prognostic Value of Plaque Burden Beyond Stenosis Severity in Patients with Stable Chest Pain Undergoing Coronary CT Angiography
    Ana Ezponda, José Marlon Rodríguez, Miguel Sogbe +6
    Medical sciences (Basel, Switzerland)Peer-reviewedCC BYFull textGeneral medicineEurope PMC
  5. Single-Stage Computed Tomography Angiography, Percutaneous Coronary Intervention, and Transcatheter Aortic Valve Replacement in an Octogenarian With Femoral Neck Fracture, Non-ST-Elevation Myocardial Infarction, and Severe Aortic Stenosis: A Case Report and Literature Review of Multidisciplinary Approaches
    Ming Zhang, Xietian Pan, Lina Yang +9
    Reviews in cardiovascular medicinePeer-reviewedCC BYFull textCardiovascular medicineEurope PMC
BibTeX
@article{harrabi2026multiview,
  title = {{A multiview video–text model detects obstructive coronary stenosis and predicts 1-year cardiovascular events from angiography}},
  author = {Sarra Harrabi and Yichen Wu and Omar Majwal and Gabriel Asslo and George Azar and Geoffrey H. Tison and Minhaj Ansari and Milos Vukadinovic and David Ouyang and Joshua P. Barrios and Jacques Delfrate and Robert Avram},
  journal = {Nature Cardiovascular Research},
  year = {2026},
  month = oct,
  doi = {10.1038/s44161-026-00890-8},
  url = {https://doi.org/10.1038/s44161-026-00890-8},
  note = {CC BY 4.0},
}
RIS
TY  - JOUR
TI  - A multiview video–text model detects obstructive coronary stenosis and predicts 1-year cardiovascular events from angiography
AU  - Sarra Harrabi
AU  - Yichen Wu
AU  - Omar Majwal
AU  - Gabriel Asslo
AU  - George Azar
AU  - Geoffrey H. Tison
AU  - Minhaj Ansari
AU  - Milos Vukadinovic
AU  - David Ouyang
AU  - Joshua P. Barrios
AU  - Jacques Delfrate
AU  - Robert Avram
PY  - 2026
DA  - 2026/10/13
JO  - Nature Cardiovascular Research
DO  - 10.1038/s44161-026-00890-8
UR  - https://doi.org/10.1038/s44161-026-00890-8
AB  - Invasive coronary angiography is the reference standard for diagnosing coronary artery disease, yet assessment is subjective, and existing artificial-intelligence tools typically assess stenosis in individual frames or videos and do not integrate the full multiview angiographic study. Here we show that DeepCORO-CLIP, trained on 203,808 angiographic videos and reports from 28,117 patients, interprets whole multiple-video angiograms at once. It detects obstructive stenoses (≥70%), with an area under the receiver operating characteristic curve (AUROC) of 0.89 internally and externally, and identifies chronic total occlusion (AUROC = 0.90 to 0.95), intra-coronary thrombus and calcification (AUROC = 0.90 to 0.91). It also predicts 1-year cardiovascular events (AUROC = 0.79, 95% CI of 0.74 to 0.84, versus 0.65 for a model derived using clinical variables) and left ventricular ejection fraction (mean absolute error 7.3% versus 8.8% for a previous model), both achieved using coronary angiography alone. In a blinded review of 365 evaluations across 288 angiograms, its reports were clinically usable in 76% of evaluations versus 88% for cardiologists, with comparable false positives (17% versus 14%, P = 0.38); residual missed stenoses reflected an adjustable threshold rather than an absent signal.
SN  - 2731-0590
ER  -