Integrated Multiscale MRI-based Frameworks for Neurovascular Dynamics and Brain Connectivity

This thesis addresses the integration of advanced neuroimaging techniques within optimized frame-works to provide novel insight into human brain architecture and to address the complex mechanisms underlying its functions. Magnetic resonance imaging (MRI) is a powerful tool for studying brain organization and neurovascular dynamics at multiple scales, particularly when integrating imaging techniques and advanced computational approaches based on image-derived features. Nevertheless, the development of robust multimodal frameworks requires considerable methodological efforts to handle some inherent limitations of clinical imaging modalities, including the systematic variability of MRI-derived metrics resulting from the chosen setting parameters and computational strategies. This thesis addresses these challenges by implementing and assessing standardized workflows to en-hance the reliability and translational relevance of neuroimaging findings. The thesis is divided into two main sections, each focusing on a key area of multimodal neuroimaging research: neurovascular coupling (NVC) and brain connectivity. In the first section, the thesis addresses a gap in the current literature regarding the reproducibility (and exchangeability) of NVC measurements from combined functional-perfusion MRI. A prelimi-nary analysis assessed the agreement of perfusion measures across multiple brain parcellation reso-lutions. Building on these findings, a benchmarking framework was implemented to compare NVC estimates across the heterogeneous computational approaches currently adopted in clinical studies. This framework allowed identifying suitable combinations of methodological choices that may en-hance reproducibility, interpretability and cross-study comparability. The second section focuses on brain structural and functional connectivity within MRI-based frame-works relevant to the study of neurodegenerative diseases. In a first study, advanced diffusion-weighted MRI was combined with positron emission tomography (PET) in humans to investigate in-vivo the existence of direct structural pathways connecting the substantia nigra pars compacta, a sub-cortical brain region hosting dopaminergic neurons, and the thalamus, another subcortical region hosting a number of neural relay stations, which are essential for the feedback control of major cog-nitive and motor functions. This study revealed significant and reproducible connectivity patterns that, if confirmed by future ex vivo investigations, could explain a number of symptoms often ob-served in Parkinson’s disease (PD). In a second study, a graph theoretical analysis of functional con-nectomes was integrated with machine learning (ML) in a comparative framework to assess the dis-criminative power of multi-scale topological features in distinguishing newly diagnosed, drug-naïve PD patients from healthy subjects. Results indicated that both global and local features were important for the early classification of functional connectomes, highlighting their potential for defining accu-rate and non-invasive neuroimaging biomarkers for PD. Finally, this thesis provides a forward-looking perspective on the combined use of structural and functional connectomics with radiomics. Building on a preliminary ML study focused on brain aging, it lays the groundwork for future integrations of these approaches to extract complementary image-based features for the investigation of brain and neurological disorders.