ETD

• ablation
• CARTO 3
• dice
• electroanatomical mapping
• electrocardiogram
• entropy
• jaccard
• pot-ischaemic ventricular tachycardia
• signal processing
• slope entropy
• ventricular tachycardia

The aim of this thesis is to explore functional connectivity analysis techniques, widely investigated in the context of electroencephalography and neuroscience, and apply them to electroanatomical maps (EAM) of post-ischemic ventricular tachycardia (VT) cases, with the goal of characterizing the arrhythmogenic substrate.
A graph will be constructed based on the spatial distribution of the acquisition points (and their corresponding electrograms) in the EAM maps. This network will be used to analyze connectivity patterns and identify potentially distinctive behaviors of pathological tissue, particularly in correspondence with annotated abnormal potentials. The methodology involves the creation of directed and weighted graphs, incorporating conduction velocity (CV) and Local Activation Time (LAT) to establish directionality and physiological values filtering. Connections between myocardial sites, constructing the network, are defined using statistical and phase-related metrics, such as Mutual Information (MI) and Phase Lag Index (PLI). The approach evaluates different entropy formulations—including Shannon, Renyi, and more importantly Slope Entropy—and incorporates signal consideration logics (AND/OR/NONE) to isolate pathological signals and to create different configurations (Prior and No Prior Knowledge approaches, applying entropy thresholds).
Once the connectivity matrix is defined, various graph-theoretical indices (called Network Metrics) will be computed and analyzed to highlight distinctive patterns involving the local or global structures of the substrate, using the Brain Connectivity toolbox. The distribution of these metrics, such as the Clustering Coefficient and Betweenness Centrality, will be numerically analyzed for the characterization of the arrhythmogenic substrate. As a further analysis, the work explores the robustness of the proposed approach when a Factor2 K-means subsampling is performed, leading to a partial EAM map used for the functional connectivity assessment.
The performance of the algorithm will be evaluated through spatial measures like Jaccard and Dice indexes, classification metrics (such as Sensitivity, Recall and F1-Score), and inter-rater agreement through Cohen’s Kappa (K). The Ground Truth used for the performance evaluation, in a retrospective evaluation, is represented by the EAM location of Abnormal Ventricular Potentials and Physiological Intracardiac Electrograms annotated by a consensus of expert cardiologists.
The results saw a Jaccard and Dice up to 0.79 and 0.91, respectively, in the Prior Knowledge configuration based on Slope Entropy AND logic and PLI connectivity matrix within the Clustering Coefficient Network Metric, with high Precision (up to 0.9) but an intermediate agreement (K=0.6). The results proved that the CV filtering and the Entropy thresholding, leading to the Prior Knowledge, outperformed the No Prior approach, where only the directionality of the LAT/CV was applied and No entropy threshold was considered, leading to a 0.68 Jaccard. The network metrics’ behavior changes passing from the No Prior Knowledge to the Prior, as the first identifies long distance EGM connections within the Closeness Centrality metric as the best case and the latter exposes the best results within short distance EGM connections within the Clustering Coefficient metric (both in the MI &PLI). The Subsampling by a Factor of 2 halved the performance, leading to the conclusion that, globally, the algorithm is influenced by the acquisition points’ density.