A Deep Learning-Based Multimodal Fusion Model for Recurrence Prediction in Persistent Atrial Fibrillation Patients

Area of Science:

• Cardiology
• Medical Informatics
• Artificial Intelligence in Medicine

Background:

• Long-term success of atrial fibrillation (AF) ablation is challenging, especially for persistent AF (PeAF) patients.
• Predicting recurrence risk in PeAF is complex, with current clinical assessments limited by their inability to fully utilize electrocardiogram (ECG) data.
• Integrating clinical features with ECG signals offers a promising approach to enhance prediction accuracy and personalize patient management.

Purpose of the Study:

• To develop and evaluate a deep learning model for predicting postablation recurrence in PeAF patients.
• To investigate the efficacy of combining preprocedural AF rhythm 12-lead ECG signals with clinical data for improved risk prediction.
• To enhance personalized clinical decision-making for PeAF patients undergoing radiofrequency catheter ablation.

Main Methods:

• A retrospective analysis of 77 PeAF patients who underwent radiofrequency catheter ablation between 2016 and 2019.
• Development of a multimodal fusion deep learning framework using a residual block network.
• Integration of preprocedural AF rhythm 12-lead ECG, clinical scores, and patient baseline characteristics, with fivefold cross-validation for training and testing.

Main Results:

• The fusion model achieved an average AUC of 0.74 (maximum 0.82) in predicting recurrence.
• The model significantly outperformed traditional clinical scoring systems and single-modal ECG-based models.
• Demonstrated robustness and stability, with a low variance of 0.08, even with small sample sizes.

Conclusions:

• A novel deep learning model combining AF rhythm ECG signals and clinical characteristics effectively predicts recurrence risk in PeAF patients post-ablation.
• This approach significantly improves prediction performance, supporting personalized clinical decision-making.
• The model shows substantial potential for clinical application in managing PeAF patients.