Evaluation of intracardiac echocardiography and deep learning for the definition of anatomical models supporting arrhythmia ablation therapy

Atrial fibrillation (AF) is the most common sustained cardiac arrhythmia and represents a major cause of morbidity and thromboembolic complications worldwide. Catheter ablation, particularly pulmonary vein isolation, has become a cornerstone of rhythm control strategies. In this context, accurate reconstruction of left atrial anatomy is essential for procedural navigation and optimization of clinical outcomes. Intracardiac echocardiography (ICE), integrated with electroanatomical mapping systems, enables real-time intraprocedural visualization of cardiac structures without using ionizing radiation. Recent advances in artificial intelligence have enabled the development of automated ICE-based reconstruction algorithms capable of generating three-dimensional anatomical models during the procedure. The aim of this thesis was to evaluate the anatomical agreement between pre-procedural computed tomography (CT)-derived models and intraprocedural ICE-derived models reconstructed using a deep learning–based algorithm. Seven patients undergoing transcatheter AF ablation within the framework of the FATA Project were included. Morphometric comparison, landmark-based spatial analysis, and volumetric overlap evaluation were performed after rigid model registration. Morphometric analysis showed no significant differences between CT- and ICE-derived measurements for most anatomical structures, with strong correlation for the inter-carinal distance. Landmark-based centroid displacement revealed moderate spatial variability, particularly at the level of the left atrial appendage. Volumetric overlap analysis showed heterogeneous Dice similarity coefficients across patients, reflecting reconstruction differences and anatomical variability. Overall, automated ICE-based reconstruction demonstrated good morphometric agreement with CT and acceptable spatial correspondence, supporting its feasibility as a reliable intraprocedural anatomical imaging tool for atrial fibrillation ablation.