Image-Decomposition-Enhanced Deep Learning for Detection of Rotor Cores in Cardiac Fibrillation

Area of Science:

• Computational cardiology
• Artificial intelligence in medicine
• Cardiac electrophysiology

Background:

• Rotors, spiral wave reentry in cardiac tissue, are key drivers of atrial fibrillation (AF).
• Traditional rotor detection methods require specialized expertise in cardiac physiology and electrogram interpretation.
• Advances in AI and smart sensing offer new avenues for diagnosing cardiac arrhythmias like AF.

Purpose of the Study:

• To develop an AI-driven framework for automatic identification of rotor cores in cardiac tissue.
• To enhance rotor detection using image decomposition techniques combined with deep learning.
• To validate the framework on both simulated and experimental optical mapping data.

Main Methods:

• Ensemble Empirical Mode Decomposition (EEMD) was used to decompose cardiac imaging data.
• The most representative decomposed component was processed by a You-Only-Look-Once (YOLO) object-detection model.
• The EEMD-YOLO framework was trained and validated using bi-domain simulation data and optical mapping data from rabbit hearts.

Main Results:

• The integrated EEMD-YOLO model achieved high accuracy in rotor core detection.
• Precision reached 97.2% for simulation data and 96.8% for optical mapping data.
• Recall was 93.8% for simulation data and 92.2% for optical mapping data, with F1 scores of 95.5% and 94.4%, respectively.

Conclusions:

• The EEMD-YOLO framework provides accurate and automated rotor core detection.
• The model's performance is comparable to established gold-standard methods in the literature.
• This AI approach has the potential to significantly aid in the diagnosis and treatment of atrial fibrillation.