Fully Automated Echocardiogram Interpretation in Clinical Practice.

Area of Science:

• Cardiology
• Artificial Intelligence
• Medical Imaging

Background:

• Automated cardiac image interpretation can enhance clinical practice, particularly for serial assessments in primary care.
• Advances in computer vision offer potential for a fully automated echocardiogram analysis pipeline.

Purpose of the Study:

• To develop and evaluate a scalable, automated pipeline for echocardiogram interpretation using convolutional neural networks.
• The pipeline aims for view identification, image segmentation, structure/function quantification, and disease detection.

Main Methods:

• Trained convolutional neural network models on 14,035 echocardiograms for view identification and cardiac chamber segmentation.
• Quantified cardiac structure and function (ejection fraction, longitudinal strain) using segmentation output.
• Developed models to detect hypertrophic cardiomyopathy, cardiac amyloid, and pulmonary arterial hypertension.

Main Results:

• Accurate automated view identification (96% for parasternal long axis) and chamber segmentation.
• Cardiac structure measurements showed agreement with clinical values (e.g., 15-17% median absolute deviation for volumes).
• Automated function measurements (ejection fraction, strain) agreed with commercial software and demonstrated serial monitoring utility.
• Disease detection models achieved high performance (C-statistics: 0.93 for hypertrophic cardiomyopathy, 0.87 for cardiac amyloid, 0.85 for pulmonary arterial hypertension).

Conclusions:

• The automated echocardiogram interpretation pipeline provides a foundation for scalable analysis of archived cardiac imaging data.
• This technology supports serial patient tracking and broad clinical application of echocardiogram interpretation.