Computational and Experimental Benchmarks for High Resolution Paradigms in Brain Source Localization

Electroencephalography (EEG) is a technique for non-invasively measuring neuronal activity in the human brain using electrodes placed on the scalp. However, the low spatial resolution of standard EEG limits accurate localization of brain activity sources. High-Resolution Electrical Source Imaging (ESI) techniques integrate high-density EEG with source localization algorithms, enabling a more spatially precise reconstruction of volumetric brain currents, thus establishing ESI as a valuable tool for brain imaging and diagnostics. This thesis provides an in-depth validation of different ESI techniques, focusing on testing and comparing their accuracy and resolution. It tackles both theoretical and practical challenges through a computational and experimental assessment. The validation framework encompasses both the forward problem (FP) and the inverse problem (IP) in ESI, leveraging computational techniques and an experimental setup to assess model accuracy. The FP maps the current distribution of a set of known sources inside the brain to the electric potential at the scalp electrodes, while the IP infers the volumetric current distribution that best fits the observed electrode measurements. Nevertheless, the IP is inherently ill-posed, lacking a unique and stable solution, which complicates precise neural source localization. This thesis addresses this challenge by evaluating both FP and IP models and testing multiple source localization algorithms. The FP is tackled through a symmetric Boundary Element Method (BEM) formulation, selected for its accuracy. The Boundary Element Method (BEM) forward model is benchmarked through a comparison with the Finite Element Method (FEM) in COMSOL Multiphysics, establishing a robust reference for the symmetric BEM approach under selected mesh resolutions and conductivity contrasts. Following this, localization algorithms, including Minimum Norm Estimation (MNE), Weighted Minimum Norm Estimation (WMNE), and standardized Low-Resolution Electromagnetic Tomography (sLORETA), are validated both numerically and experimentally. Among these, sLORETA emerged as the most accurate, particularly for deep source configurations. For experimental validation, a custom cylindrical phantom filled with a saline solution and a NaCl-Agar gel is used. A coaxial cable is employed as a current source, modeling neural activity. This setup provides a high-fidelity assessment of inverse algorithms within a controlled laboratory environment. Additionally, a thorough characterization of the conductive materials, such as NaCl in H2O and NaCl-Agar gel, was performed prior to the experiment to ensure the correct conductivity of the materials, thus enhancing experimental reliability. This thesis concludes with a discussion of the implications of these findings for ESI methodologies and their potential applications in clinical diagnostics and neuroscience. By providing theoretical insights and practical tools for high-resolution brain source localization, this work establishes a validated framework for future ESI advancements, laying a solid foundation for the development of more accurate and reliable EEG-based neuroimaging techniques.