Comparing Artificial Intelligence-Enabled Electrocardiogram Models in Identifying Left Atrium Enlargement and...

Area of Science:

• Cardiology
• Artificial Intelligence
• Medical Diagnostics

Background:

• The diagnostic capability of AI-enabled ECG models for left atrial enlargement (LAE) is not well-defined.
• The utility of AI-enabled single-lead ECG for LAE surveillance requires further investigation.

Purpose of the Study:

• To develop AI-enabled P-wave and single-lead ECG models for LAE detection.
• To compare the prognostic performance of AI-assessed severe LAE with echocardiography.

Main Methods:

• Retrospective analysis of 382,594 adult patients with paired ECG and echocardiography.
• Utilized UNet++ for P-wave segmentation and ResNet-18 for deep learning ECG models.
• External validation included 11,753 patients from an additional hospital.

Main Results:

• AI-enabled 12-lead ECG models demonstrated superior LAE classification performance.
• Single-lead AI-enabled ECG models showed excellent performance for LAE detection at a 50 mm cutoff.
• AI-identified severe LAE was a stronger predictor of cardiovascular disease than echocardiography.

Conclusions:

• P-wave analysis is critical for AI-driven LAE detection using ECG.
• AI-enabled ECG models offer superior prediction of cardiovascular diseases associated with severe LAE compared to echocardiography.