Snapshot artificial intelligence-determination of ejection fraction from a single frame still image: a...

Area of Science:

• Cardiology
• Artificial Intelligence
• Medical Imaging

Background:

• Artificial intelligence (AI) shows promise for point-of-care cardiac assessment.
• Previous AI models for left ventricular ejection fraction (LVEF) estimation often required computationally intensive video inputs.
• This study aimed to develop an AI model using static frames for LVEF estimation.

Purpose of the Study:

• To develop and evaluate an AI model for estimating LVEF from static echocardiogram frames.
• To assess the model's performance across diverse datasets, including handheld cardiac ultrasound (HCU).

Main Methods:

• A two-dimensional convolutional neural network was trained on retrospective transthoracic echocardiography (TTE) data.
• Model performance was validated on multiple TTE datasets, a public dataset, and prospective HCU data.
• Evaluation included comparisons between expert and novice HCU data collection.

Main Results:

• The AI model demonstrated strong performance in LVEF classification (AUC > 0.90) when single-frame estimates were averaged.
• Performance remained robust even with a single frame per video, with AUC > 0.85 for novice-collected HCU clips.
• LVEF estimates showed slight variations based on the cardiac cycle phase of image acquisition.

Conclusions:

• Static frames from multiple videos may suffice for rapid LVEF classification deployment.
• The model's sensitivity to the cardiac cycle provides insights into its explainability.
• AI models using static frames offer a computationally efficient approach to cardiac function assessment.