Related Experiment Video
Updated: Jun 22, 2025
06:34
Evaluation of Left Ventricular Structure and Function using 3D Echocardiography
Published on: October 28, 2020
3.9K
Deep Learning to Estimate Left Ventricular Ejection Fraction From Routine Coronary Angiographic Images
Behrouz Rostami1, Kenneth Fetterly1, Zachi Attia1
1Department of Cardiovascular Medicine, Mayo Clinic, Rochester, Minnesota, USA.
JACC. Advances
|June 28, 2024
Summary
Deep learning models can accurately assess left ventricular ejection fraction (LVEF) from coronary angiography images. This novel algorithm surpasses human expert performance, aiding clinical decisions.
Area of Science:
- Cardiology
- Artificial Intelligence
- Medical Imaging
Background:
- Coronary angiography cine images offer more than stenosis assessment.
- Deep learning (DL) may identify left ventricular dysfunction from these images.
Purpose of the Study:
- To evaluate machine learning models for estimating left ventricular ejection fraction (LVEF) using routine coronary angiograms.
Main Methods:
- A 3D-convolutional neural network (CNN) and transformer model processed dual coronary angiogram projections (LAO-caudal, RAO-cranial).
- The model classified LVEF as reduced (≤40%) or normal/mildly reduced (>40%), validated against echocardiography.
Main Results:
- The DL model achieved an AUC of 0.87, sensitivity 0.77, and specificity 0.80 on the test set.
- Performance exceeded human expert assessment (AUC 0.76-0.77).
- Combined LAO and RAO views improved accuracy over individual views; the CNN-transformer model outperformed 3D-CNN or transformer-only models.
Conclusions:
- A novel DL algorithm accurately and rapidly assesses LVEF from coronary angiography.
- This tool can support clinical decisions and enable future data extraction from angiograms.
