VISION-ICE: Video-based Interpretation and Spatial Identification of Arrhythmia Origins via Neural Networks in Intracardiac Echocardiography

TL;DR

The feasibility and clinical promise of using ICE videos combined with deep learning for automated arrhythmia localization are demonstrated and an AI-enabled framework that leverages intracardiac echocardiography (ICE) is proposed to guide clinicians toward areas of arrhythmogenesis and potentially reduce procedural time.

Abstract

Contemporary high-density mapping techniques and preoperative CT/MRI remain time and resource intensive in localizing arrhythmias. AI has been validated as a clinical decision aid in providing accurate, rapid real-time analysis of echocardiographic images. Building on this, we propose an AI-enabled framework that leverages intracardiac echocardiography (ICE), a routine part of electrophysiology procedures, to guide clinicians toward areas of arrhythmogenesis and potentially reduce procedural time. Arrhythmia source localization is formulated as a three-class classification task, distinguishing normal sinus rhythm, left-sided, and right-sided arrhythmias, based on ICE video data. We developed a 3D Convolutional Neural Network trained to discriminate among the three aforementioned classes. In ten-fold cross-validation, the model achieved a mean accuracy of 66.2% when evaluated on four previously unseen patients (substantially outperforming the 33.3% random baseline). These results demonstrate the feasibility and clinical promise of using ICE videos combined with deep learning for automated arrhythmia localization. Leveraging ICE imaging could enable faster, more targeted electrophysiological interventions and reduce the procedural burden of cardiac ablation. Future work will focus on expanding the dataset to improve model robustness and generalizability across diverse patient populations.

Figures (3)

• Figure 1: Overview of the preprocessing pipeline. It includes masking of irrelevant regions, temporal segmentation, spatial cropping and resizing, and data augmentation.
• Figure 2: Overview of the hierarchical evaluation framework for ICE-based arrhythmia localization. For each of the ten folds, patients are split into independent training, validation, and test subsets to prevent data leakage. Within each fold, four models are trained independently, one per anatomical view, using early stopping based on sample-level metrics. Clip-level predictions are then obtained by majority voting across heartbeats within each clip. At the final fusion stage, predictions from all available views are aggregated by majority voting to produce cross-view patient-level decisions. Final performance metrics are computed and averaged across folds to yield the overall evaluation summary.
• Figure 3: Grad-CAM visualizations derived from the deepest convolutional layer (layer $4$) of the modified ResNet18-3D. The top row displays the original intracardiac echocardiography frames, and the bottom row shows the corresponding activation heatmaps. The two columns on the left correspond to correctly classified samples, and the two columns on the right illustrate misclassified cases. Warmer colors (red/yellow) indicate regions with higher contribution to the model’s prediction, while cooler colors (blue) represent areas of lower importance.