A view-flexible deep learning framework for automated analysis of 2D echocardiography

Area of Science:

• Cardiology
• Artificial Intelligence
• Medical Imaging

Background:

• Traditional echocardiography demands expert operators for image acquisition and interpretation.
• Handheld cardiac ultrasound (HCU) is often used by novice practitioners, limiting clinical decision-making.
• Current methods struggle with view variability and user expertise.

Purpose of the Study:

• To develop a view-flexible deep learning framework for estimating key cardiac parameters.
• To assess the framework's performance across different echocardiography datasets and user expertise levels.
• To broaden the clinical utility of echocardiography by reducing reliance on operator skill.

Main Methods:

• A deep learning framework was designed to be view-flexible.
• The model was trained to estimate left ventricular ejection fraction (LVEF), patient age, and patient sex.
• Performance was evaluated on retrospective transthoracic echocardiography (TTE) and prospective HCU datasets.

Main Results:

• The framework demonstrated consistently strong performance across TTE datasets.
• Comparable performance was observed between prospective HCU and TTE for LVEF, age, and sex estimation.
• Model performance on HCU data was similar whether collected by experts or novice users.

Conclusions:

• A view-flexible deep learning approach can accurately estimate LVEF, age, and sex from cardiac ultrasound.
• This technology shows promise in overcoming limitations associated with user expertise in echocardiography.
• The framework has the potential to significantly expand the clinical applications of echocardiography.