Deep Learning Model for High-Accuracy Classification of Premature Ventricular Contractions With Precordial Transition...

Area of Science:

• Cardiology
• Medical Imaging
• Artificial Intelligence

Background:

• Predicting premature ventricular contractions (PVCs) origin is difficult with transition zones in leads V3-V4.
• Electrocardiogram (ECG) analysis is crucial for diagnosing cardiac arrhythmias.

Purpose of the Study:

• Develop a deep learning model for predicting PVC origin.
• Identify key ECG features influencing the model's predictions.

Main Methods:

• A convolutional neural network (CNN) was trained on ECG data from 314 patients.
• The model used paired PVC and intrinsic QRS (iQRS) data for training.
• Gradient-weighted class activation mapping identified influential ECG regions.

Main Results:

• The CNN model achieved 92.1% accuracy and an F1 score of 0.91.
• Diagnostic performance surpassed conventional ECG indices.
• Model attention focused on V3-V4 in iQRS and inferior limb leads/V2-V3 in PVC.

Conclusions:

• The CNN model shows significant clinical utility for predicting PVC origin.
• Deep learning enhances the interpretation of complex ECG patterns.