Subclinical atrial fibrillation prediction based on deep learning and strain analysis using echocardiography

Area of Science:

• Cardiology
• Artificial Intelligence
• Medical Imaging

Background:

• Subclinical atrial fibrillation (SCAF), or atrial high-rate episodes (AHREs), presents asymptomatic heart rate elevations linked to increased cardiovascular risks.
• Deep learning (DL) models are established for cardiac function analysis via echocardiography, but their use in AHRE prediction is novel.

Purpose of the Study:

• To develop and validate a novel deep learning framework for automated detection of atrial high-rate episodes (AHREs) using echocardiographic data.
• To assess the framework's performance in segmenting the left atrium and classifying AHREs.

Main Methods:

• A deep learning framework was developed, involving left atrium (LA) segmentation, LA strain feature extraction, and AHRE classification.
• The study analyzed echocardiograms from 117 patients, utilizing 80% for development and 20% for testing.
• A transformer-based model integrated patient characteristics for AHRE prediction.

Main Results:

• Left atrium segmentation achieved high accuracy with Dice coefficients of 0.923 (cavity) and 0.741 (wall).
• The transformer model demonstrated strong performance in AHRE classification, with a mean AUC of 0.815, accuracy of 0.809, sensitivity of 0.800, and specificity of 0.783 for a 24-hour threshold.

Conclusions:

• The developed DL framework provides a reliable method for assessing AHREs.
• This approach shows significant potential for early subclinical atrial fibrillation detection, improving clinical decisions and patient outcomes.