Detection of Left Atrial Enlargement Using a Convolutional Neural Network-Enabled Electrocardiogram

Area of Science:

• Cardiology
• Artificial Intelligence
• Medical Diagnostics

Background:

• Left atrial enlargement (LAE) is an independent predictor of cardiovascular diseases.
• Early detection of LAE is crucial for cardiovascular risk management.

Purpose of the Study:

• To develop and validate an artificial intelligence (AI) approach for detecting LAE using 12-lead electrocardiography (ECG).

Main Methods:

• Trained a convolutional neural network (CNN) on ECG data from 3,391 elderly individuals (≥65 years) with echocardiography confirmation of LAE.
• Evaluated model performance using Area Under the Curve (AUC), accuracy, sensitivity, specificity, and F1 score.

Main Results:

• The AI-enabled ECG achieved an AUC of 0.949 for LAE detection, significantly outperforming physician diagnoses (AUC 0.610).
• The AI model demonstrated high sensitivity (84.0%), specificity (92.0%), and accuracy (88.0%) in identifying LAE.
• The AI model also showed excellent performance in classifying different degrees of LAE, with AUCs up to 0.998.

Conclusions:

• AI-enabled ECG analysis can reliably identify individuals with a high likelihood of LAE.
• This AI tool shows promise for non-invasive LAE screening, potentially improving cardiovascular disease risk stratification.
• Further refinement and external validation are recommended for clinical implementation.