PVCsNet : A Specialized Artificial Intelligence-Based Model to Classify Premature Ventricular Contractions From ECG...

Area of Science:

• Cardiology
• Artificial Intelligence
• Medical Imaging

Background:

• Premature ventricular complexes (PVCs) disrupt normal heart rhythm, necessitating accurate origin identification for effective treatment.
• Current methods for PVC detection and origin localization have limitations in accuracy and data processing.
• Pre-operative identification of PVC origins can optimize surgical duration, reduce radiation exposure, and improve ablation success.

Purpose of the Study:

• To develop and evaluate PVCsNet, a deep-learning network for classifying PVCs in ECG images based on their origin.
• To assess the performance of PVCsNet in accurately categorizing PVCs from various cardiac regions.

Main Methods:

• A novel deep-learning network, PVCsNet, was designed using convolutional layers, residual connections, and attention mechanisms.
• The network was trained and validated on ECG images, classifying PVCs into six origin categories: RVOT, LVOT, PM, VA, summit, and HPS.
• Specific configurations, including the SE Block with MaxPool and a ratio of 4, were tested to optimize performance.

Main Results:

• PVCsNet achieved an overall accuracy of 94.49% in classifying PVC origins from ECG images.
• The network demonstrated high precision for critical PVC categories, including RVOT, PM, and HPS.
• PVCsNet exhibited a moderate parameter size, indicating efficiency in its design.

Conclusions:

• PVCsNet shows significant potential for clinical diagnostics in accurately classifying PVC origins from ECG data.
• The study highlights the efficacy of deep learning, particularly with attention mechanisms, in analyzing complex cardiac signals.
• Accurate PVC origin classification using PVCsNet can contribute to improved patient management and future research in cardiac electrophysiology.