Related Experiment Video
Updated: Jun 17, 2026
04:29
Three-Dimensional Modeling of the Left Atrium and Pulmonary Veins with a Precise Intracardiac Echocardiography Approach
Published on: June 30, 2023
Accuracy of a Deep Learning Model in Intracardiac Echocardiography.
Devi Nair1, Jeffrey Winterfield2, Jonathan C Hsu3
1St. Bernards Heart & Vascular Center, Jonesboro, Arkansas, USA.
JACC. Advances
|June 15, 2026
Summary
This study introduces Auto-Contour, an AI tool for segmenting intracardiac echocardiography (ICE) images. It shows promise for standardizing procedures and improving safety in cardiac interventions.
Area of Science:
- Cardiology
- Medical Imaging
- Artificial Intelligence
Background:
- Intracardiac echocardiography (ICE) interpretation is operator-dependent, lacking standardized views.
- AI applications in ICE lag behind other echocardiography modalities.
Purpose of the Study:
- Develop and evaluate Auto-Contour, a deep-learning pipeline for ICE.
- Assess its feasibility for real-time procedural guidance and multistructure semantic segmentation.
Main Methods:
- Retrospective analysis of 5,496 ICE cine loops from 249 patients.
- Expert annotation of 65,117 segmentations across 20 procedural views.
- Training a deep-learning model for semantic segmentation.
Main Results:
- High segmentation performance for left atrium (Dice 0.94) and left ventricle (Dice 0.82).
- Acceptable performance for smaller structures like the left atrial appendage and pulmonary veins.
- Mean per-frame inference time under 0.03 seconds for real-time application.
Conclusions:
- Auto-Contour achieves robust, real-time multistructure segmentation of ICE anatomy.
- Supports prospective evaluation for AI-assisted ICE in standardizing, enhancing efficiency, and safety.
