Multimodal Patient-Specific Identification of Atrial Flutter Circuits From ECG Time Series Using Explainable Machine...

Area of Science:

• Cardiology
• Medical Imaging
• Computational Biology

Background:

• Surface electrocardiogram (ECG) criteria for atrial flutter (AFL) are often unreliable for pre-procedural mechanism identification.
• Invasive electrophysiological (EP) studies are typically required to confirm AFL circuit definitions and guide ablation strategies.

Purpose of the Study:

• To develop and evaluate an interpretable, non-invasive framework for stratifying AFL mechanisms.
• To improve the accuracy of pre-procedural AFL identification to streamline mapping and ablation planning.

Main Methods:

• Analysis of 97 patients undergoing EP study with simultaneous 12-lead ECG.
• Reconstruction of atrial vectorcardiograms (VCGs) and extraction of interpretable loop morphology and kinematic descriptors.
• Development of an explainable tree-ensemble model integrating VCG descriptors and clinical variables.

Main Results:

• VCG loops demonstrated subtype-specific archetypes for different AFL mechanisms.
• The multimodal Random-Forest model achieved high discrimination (AUROC 0.790-0.900) across AFL subtypes.
• The model showed high sensitivity for common AFL and very high specificity for perimitral clockwise AFL.

Conclusions:

• The developed interpretable framework offers a practical, non-invasive method for mechanism-oriented AFL stratification.
• This approach supports targeted mapping and more efficient ablation planning for atrial flutter.
• Future research will focus on multicenter validation and broadening applicability by removing adenosine dependency.