Inter-Patient Atrial Flutter Classification Using FFT-Based Features and a Low-Variance Stacking Classifier.

Area of Science:

• Cardiology
• Biomedical Engineering
• Artificial Intelligence

Background:

• Atrial Flutter (AFL) is a common arrhythmia originating from macroreentry circuits.
• Catheter ablation success relies on precise circuit identification via invasive electrophysiological (EP) study.
• Variable atrial anatomy complicates AFL circuit determination.

Purpose of the Study:

• To investigate machine learning (ML) for determining AFL macroreentry circuits from surface electrocardiograms (ECGs).
• To assess if ML can predict specific AFL circuits before invasive EP study.
• To develop a non-invasive method for AFL circuit classification.

Main Methods:

• Utilized 12-lead ECGs, reduced to eight independent leads (I, II, V1-V6).
• Developed an algorithm using ventricular complex cancellation to isolate atrial activity.
• Applied and combined Support Vector Machine (SVM), Random Forest (RF), and k-Nearest Neighbors (KNN) classifiers.

Main Results:

• Analyzed 419 retrospective AFL patient ECGs.
• Achieved 98% test accuracy and 0.97 f1 score using lead V5.
• Demonstrated high classification accuracy (95+%) on an unbalanced dataset.

Conclusions:

• Machine learning methods can automatically determine AFL macroreentry circuits from surface ECGs.
• This ML approach shows significant potential for clinical application in AFL management.
• Non-invasive ECG analysis with ML offers a promising alternative for pre-ablation AFL circuit identification.