Unsupervised Classification of Atrial Electrograms for Electroanatomic Mapping of Human Persistent Atrial...

Area of Science:

• Cardiology
• Biomedical Engineering
• Computational Biology

Background:

• Persistent atrial fibrillation (persAF) ablation is challenging due to undefined atrial substrate and multiple arrhythmia mechanisms.
• Current methods lack a 'ground truth' for characterizing the atrial substrate, complicating ablation target identification.

Purpose of the Study:

• To implement an unsupervised classification of atrial electrograms (AEGs) to identify distinct patterns in persAF.
• To validate these AEG clusters using electrophysiological markers for improved substrate characterization.

Main Methods:

• Collected 956 bipolar AEGs from 11 persAF patients.
• Utilized CARTO variables to create a 3D space for k-means unsupervised classification.
• Analyzed identified AEG groups using nine derived markers: SampEn, dominant frequency, OI, determinism, laminarity, RR, PP amplitude, CL, and WS.

Main Results:

• Identified five distinct AEG classes (F = 582, P<0.0001) with varying degrees of organization and fractionation.
• Class 1 (25%) showed organized AEGs with high wave similarity (WS) and low sample entropy (SampEn).
• Class 5 (20%) exhibited fractionated AEGs with low WS, high SampEn, and low organization index (OI).

Conclusions:

• The unsupervised classification expands criteria for automated AEG analysis in persAF.
• Nine electrophysiological markers effectively differentiated the five identified AEG classes.
• This approach offers a more comprehensive characterization of the persAF substrate for future ablation target identification.