DeepCORO-CLIP

Montreal Heart Institute / UCSF / Cedars-Sinai (Harrabi, Avram, Tison, Ouyang et al.)

Coronary angiography

Filter catalog by Modality:
Invasive Coronary & Intracoronary Imaging

Coronary artery disease / stenosis

Filter catalog by Disease / Trait:
Coronary & Ischemic Disease

LVEF estimation

Filter catalog by Disease / Trait:
Cardiac Function & Hemodynamics

Major adverse cardiovascular events (MACE)

Filter catalog by Disease / Trait:
Prognosis & Aging

Binary classification

Filter catalog by Task Type:
Classification

Regression

Filter catalog by Task Type:
Regression

Hybrid

Filter catalog by Architecture:
Hybrid / Multi-branch

PyTorch

Filter catalog by Framework:
PyTorch

file_copy

View model files code View code

Multi-view foundation model for coronary angiography trained with video-text contrastive learning on 203,808 angiography videos from 28,117 patients across 32,473 studies at the Montreal Heart Institute, externally validated on 4,249 studies from UCSF. Integrates multiple angiographic projections with attention-based pooling for study-level assessment spanning diagnostic, prognostic, and disease-progression tasks: significant-stenosis detection (AUROC 0.888 internal / 0.89 external), stenosis-percentage estimation (MAE 13.6% vs. 19.0% for clinical reports), chronic total occlusion, intracoronary thrombus, and coronary calcification detection. Transfer learning further enables one-year MACE prediction (AUROC 0.79) and LVEF estimation (MAE 7.3%) from the same angiography embeddings, with a mean in-hospital inference time of 4.2 seconds.

memory Specifications

description Publication

DeepCORO-CLIP: A Multi-View Foundation Model for Comprehensive Coronary Angiography Video-Text Analysis and External Validation open_in_new

arXiv preprint · 2026 · original paper

database Training & evaluation data

Montreal Heart Institute Coronary Angiography-Report Archive (DeepCORO-CLIP)

pretrain

public 28,117 subjects · Canada

203,808 coronary angiography videos across 32,473 studies from 28,117 patients at the Montreal Heart Institute, paired with angiography report text for video-text contrastive pretraining. A curated public benchmark subset ("DeepCORO-mini", ~1,000 cases / ~7,000 videos) is available via controlled-access request on PhysioNet.

UCSF External Coronary Angiography Validation Cohort (DeepCORO-CLIP)

test

USA

4,249 external validation studies from the University of California, San Francisco, used to externally validate DeepCORO-CLIP.

science Capabilities & performance

Significant coronary stenosis detection from multi-view angiography video

Binary classification Coronary artery disease / stenosis

0.888 AUROC Montreal Heart Institute internal validation · internal

0.89 AUROC UCSF external validation · external

Coronary stenosis percentage estimation from multi-view angiography video (vs. core-lab QCA)

Regression Coronary artery disease / stenosis

13.6 MAE Montreal Heart Institute internal validation · internal

Left ventricular ejection fraction estimation via transfer learning from angiography embeddings

Regression LVEF estimation

7.3 MAE Montreal Heart Institute internal validation · internal

One-year major adverse cardiovascular event (MACE) prediction via transfer learning from angiography embeddings

Binary classification Major adverse cardiovascular events (MACE)

0.79 AUROC Montreal Heart Institute internal validation · internal

CVAI Catalog

View Catalog