Prediction of premature ventricular complex origins using artificial intelligence-enabled algorithms.

Area of Science:

• Cardiology
• Medical Artificial Intelligence
• Computational Electrophysiology

Background:

• Catheter ablation is a primary treatment for frequent premature ventricular complexes (PVCs).
• Accurate prediction of PVC origin from 12-lead electrocardiograms (ECGs) is vital but challenging, requiring significant expertise.
• Existing methods for PVC origin localization can be complex and time-consuming.

Purpose of the Study:

• To develop and assess machine learning (ML) algorithms for predicting PVC origins using ECG data.
• To compare the performance of ML models against board-certified electrophysiologists and a current classification algorithm.
• To evaluate the utility of ML in improving the accuracy and efficiency of PVC origin identification.

Main Methods:

• Developed and trained Support Vector Machine (SVM) and Convolutional Neural Network (CNN) models.
• Utilized ECG data from 116 PVCs in 111 patients undergoing catheter ablation, with origins confirmed by 3D electroanatomical mapping.
• Classified PVC origins into four groups (right/left outflow tract, other) and performed binary classification (right vs. left).

Main Results:

• The SVM model achieved a weighted accuracy of 0.85 for 4-class classification, while the CNN achieved 0.80.
• ML models demonstrated superior precision, recall, and F1-scores compared to electrophysiologists.
• SVM achieved high accuracy in binary classification (0.94 for right vs. left origin).

Conclusions:

• AI-powered algorithms can accurately predict PVC origins from ECGs.
• These ML models offer performance comparable to or exceeding existing algorithms and human experts.
• The developed algorithms show significant potential to enhance the localization of PVC origins for guiding ablation therapy.