CVAI Catalog

·

View Catalog

DeepECG-SSL

EfficientNetV2 (self-supervised)

Montreal Heart Institute (HeartWise.AI) (Avram et al.)

12-lead ECG

Filter catalog by Modality:
ECG

General Purpose / Multi-task

Filter catalog by Disease / Trait:
General / Foundation

LV systolic dysfunction (LVSD)

Filter catalog by Disease / Trait:
Cardiac Function & Hemodynamics

Atrial fibrillation

Filter catalog by Disease / Trait:
Arrhythmia

Long QT syndrome (LQTS)

Filter catalog by Disease / Trait:
Arrhythmia

Multi-label classification

Filter catalog by Task Type:
Classification

Binary classification

Filter catalog by Task Type:
Classification

Multi-class classification

Filter catalog by Task Type:
Classification

CNN (2D)

Filter catalog by Architecture:
Convolutional (CNN)

PyTorch

Filter catalog by Framework:
PyTorch

Self-supervised EfficientNetV2-based 12-lead ECG foundation model pretrained via contrastive learning and masked-lead modeling on 1.9 million ECGs (Montreal Heart Institute plus CODE-15% and MIMIC-IV), then fine-tuned for the same 77-condition ECG interpretation task and digital-biomarker extraction as DeepECG-SL. Outperforms the supervised counterpart on label-scarce digital-biomarker tasks, with the largest gains on LQTS genotype classification (AUROC 0.931 vs. 0.850, n=127 ECGs) and 5-year atrial-fibrillation risk (AUROC 0.742 vs. 0.734, n=132,050 ECGs), and outperforms ECG-FM and ECGFounder on shared external diagnostic classes.

memory Specifications

category

Architecture

CNN (2D)

EfficientNetV2 2D-CNN backbone pretrained via self-supervised contrastive learning + masked-lead modeling on 12-lead ECG, then fine-tuned per downstream task (77-class interpretation, LVEF, AF risk, LQTS)

code

Framework

PyTorch

calendar_month

Added to catalog

2026-08-10

description Publication

database Training & evaluation data

Brazil

Publicly available 15% subset of the CODE (Clinical Outcomes in Digital Electrocardiology) dataset from the Telehealth Network of Minas Gerais; full CODE dataset (>2 million exams) is request-only.

DeepECG Multi-Site External Validation Cohort

test

Aggregate of 11 geographically diverse external validation cohorts totaling 881,403 ECGs: four public datasets (UK Biobank, Canadian Longitudinal Study on Aging [CLSA], MIMIC-IV, PTB) and seven private health-system datasets (UCSF, MGH, Cedars-Sinai, JGH, UW, NYP, CHUM).

public 161,352 subjects · USA

800,035 12-lead ECG-report pairs from 161,352 subjects at Beth Israel Deaconess Medical Center.

Montreal Heart Institute ECG Database (MHI-ds)

pretrain

public 184,210 subjects · Canada

Over 1,000,000 12-lead ECGs from 184,210 patients treated at the Montreal Heart Institute, with automated free-text extraction of 77 AHA-derived diagnostic labels from ECG reports; primary training data for DeepECG-SL and part of the DeepECG-SSL pretraining corpus.

science Capabilities & performance

Multi-label 12-lead ECG interpretation across 77 AHA-derived cardiac conditions

Multi-label classification General Purpose / Multi-task
0.99 AUROC MHI-ds internal test · internal
0.981 AUROC External public datasets (UKB, CLSA, MIMIC-IV, PTB) · external
0.983 AUROC External private health-center datasets (UW, UCSF, JGH, NYP, MGH, CSH, CHUM) · external

Reduced LVEF (<=40%) classification from 12-lead ECG

Binary classification LV systolic dysfunction (LVSD)
0.926 AUROC DeepECG external validation (LVEF<=40%) · external

5-year atrial fibrillation risk prediction from 12-lead ECG

Binary classification Atrial fibrillation
0.742 AUROC DeepECG external validation (5-year AF risk) · external

Long QT syndrome (LQTS) detection from 12-lead ECG

Binary classification Long QT syndrome (LQTS)
0.767 AUROC DeepECG external validation (LQTS detection) · external

Long QT syndrome (LQTS) genotype classification from 12-lead ECG

Multi-class classification Long QT syndrome (LQTS)
0.931 AUROC DeepECG external validation (LQTS genotype) · external