Latent Representations of Intracardiac Electrograms for Atrial Fibrillation Driver Detection

2025-07-24

AI Summary

Identifying non-pulmonary vein (non-PV) drivers in persistent atrial fibrillation (AF) remains challenging, limiting ablation efficacy. To address this, we propose an unsupervised deep learning framework based on convolutional autoencoders that automatically extracts physiologically meaningful, low-dimensional latent representations from unipolar and bipolar electrograms (EGMs) without manual annotation. The model supports real-time inference and is compatible with clinical electrophysiological mapping systems. Evaluated on 291 patient datasets, it achieves low reconstruction error and demonstrates robust performance in detecting EGM fractionation (AUC = 0.93) and AF drivers—including rotational activity and focal activations (AUC = 0.73–0.76). Key contributions are: (i) the first application of unsupervised feature learning to multimodal EGM analysis; (ii) automated, interpretable characterization of AF mechanisms; and (iii) a novel tool enabling mechanism-guided, individualized ablation for complex AF.

Technology Category

• Machine Learning: Deep Generative Models & Autoencoders
• Knowledge Representation and Reasoning: Diagnosis and Abductive Reasoning
• Computer Vision: Segmentation

Application Category

• Graph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs
• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Search and Retrieval-Augmented AI: Agentic search

Abstract

Atrial Fibrillation (AF) is the most prevalent sustained arrhythmia, yet current ablation therapies, including pulmonary vein isolation, are frequently ineffective in persistent AF due to the involvement of non-pulmonary vein drivers. This study proposes a deep learning framework using convolutional autoencoders for unsupervised feature extraction from unipolar and bipolar intracavitary electrograms (EGMs) recorded during AF in ablation studies. These latent representations of atrial electrical activity enable the characterization and automation of EGM analysis, facilitating the detection of AF drivers. The database consisted of 11,404 acquisitions recorded from 291 patients, containing 228,080 unipolar EGMs and 171,060 bipolar EGMs. The autoencoders successfully learned latent representations with low reconstruction loss, preserving the morphological features. The extracted embeddings allowed downstream classifiers to detect rotational and focal activity with moderate performance (AUC 0.73-0.76) and achieved high discriminative performance in identifying atrial EGM entanglement (AUC 0.93). The proposed method can operate in real-time and enables integration into clinical electroanatomical mapping systems to assist in identifying arrhythmogenic regions during ablation procedures. This work highlights the potential of unsupervised learning to uncover physiologically meaningful features from intracardiac signals.

Problem

• Detect atrial fibrillation drivers using deep learning
• Automate EGM analysis for arrhythmia characterization
• Enable real-time integration into clinical mapping systems

Innovation

• Deep learning with convolutional autoencoders
• Unsupervised feature extraction from EGMs
• Real-time integration into clinical systems