Personalization of epileptogenic network models based on connectivity analysis

Abstract

The use of computational brain models in epilepsy constitutes a powerful tool not only to improve our understanding of the disease, but also to optimize and personalize treatments in silico. In this direction, properly extracting functional connectivity (FC) measures from epileptogenic networks is crucial for a proper personalization of brain models. This thesis focuses on investigating the effectiveness of three connectivity metrics—cross-correlation, phase locking value (PLV), and h2 correlation—for extracting FC values from empirical and simulated stereo-electroencephalography (SEEG) data. The main goal was to identify the most suitable metric for fitting the epileptogenic computational models used at Neuroelectrics in the framework of the Galvani project. To achieve this goal, a broad analysis was performed on empirical SEEG recordings from epilepsy patients, and the simulated SEEG data obtained from their personalized models. We computed the cross-correlation, PLV, and h2 correlation between signal pairs, and their respective ability to capture typical epileptogenic FC patterns was assessed and tested using bivariate surrogates. Although the findings revealed substantial variability in the connectivity results obtained with each metric and in each patient, the h2 correlation shows a better performance, especially in the epileptogenic zone, which is the most critical area in our modelling pipeline. This superior performance of the h2 correlation fits the state-of-the-art literature and suggests its potential as a valuable tool for personalized modelling, ultimately leading to more effective, personalized and life-changing treatments for epilepsy patients. This work introduces a novel aspect by incorporating surrogates in the analysis of simulated epileptogenic SEEG data, and provides empirical evidence of the efficacy of connectivity metrics in functionally characterizing epileptogenic networks and in supporting the development of personalized and accurate brain models.