A MULTI-COMPARTMENT MODEL FOR PATHOLOGICAL CONNECTOMES

The brain is the most complex organ in our body, and understanding the mechanisms that govern it is crucial for the treatment and prevention of neurodegenerative diseases. Magnetic Resonance Imaging allows for the acquisition of detailed, non-invasive in vivo images of brain tissue. Additionally, diffusion-weighted magnetic resonance imaging, which is particularly sensitive to the movement of water molecules in tissues, enables the reconstruction of the main axonal pathways within white matter using tractography algorithms. These algorithms, combined with the connectome framework, allow us to model the brain as a network, where nodes represent gray matter regions and edges correspond to neuronal pathways estimated through tractography, represented as polylines called streamlines. The weight of these edges is expected to reflect properties of the underlying microstructure, such as axonal density. Former methods for quantifying the connection strength assigned weights to the edges based on the number of streamlines reconstructed between regions. However, this approach has been shown to lack quantitative accuracy. Tractography reconstructs pathways according to the direction of water diffusion within axons, but it does not inherently consider the underlying microstructure. To address this limitation, a class of techniques known as microstructure-informed tractography was developed. These methods integrate information from diffusion-weighted images with tractography, enhancing its quantitative accuracy. This class of methods relies on the assumption that a streamline represents a group of axons following a shared trajectory. Therefore, along this trajectory, the microstructural values in the acquired image are assumed to remain constant. However, this assumption does not hold in the presence of focal pathologies that locally disrupt white matter integrity. This PhD project addresses the need to quantitatively estimate the reconstruction of axonal pathways in the white matter (structural connectivity) without introducing bias due to pathology. To achieve this, we first studied the effects of applying microstructure-informed tractography methods in the presence of focal lesions. We identified what we call the “butterfly effect”: errors in connectivity estimation caused by lesions also affect healthy fiber bundles, not just those directly impacted, rendering the model nearly insensitive to pathology. We proposed an extension of the forward model used by this class of methods consisting of adding a compartment to account for signal attenuation within lesions in addition to the compartment already employed to fit the intra-axonal signal. Finally, we introduced the refinement of the weights of streamlines passing through lesions by assigning them the minimum contribution within the lesion, following the principle that “a chain is only as strong as its weakest link.” We tested our proposals on various scenarios. First, we introduced artificial lesions into a population of healthy controls to establish ground truth for comparison. Then, we applied the model to different cohorts of multiple sclerosis patients, yielding promising results. Our results demonstrated that the developed method provides accurate structural connectivity estimates in patients with focal pathologies, such as multiple sclerosis, while significantly improving sensitivity to pathology compared to the original model. This work represents a major step forward in understanding the impact of focal pathologies on structural connectivity, offering valuable insights into the mechanisms underlying these conditions. The thesis is structured as follows: Chapter 2 provides a comprehensive background, introducing the essential concepts necessary for understanding the research conducted. Chapter 3 presents an application for studying aging in healthy individuals using an innovative method that weighs structural connectivity based on myelin content. Chapter 4 reviews the literature on the challenges of quantitatively estimating pathological connectomes, which is a central issue in this project. This chapter concludes by addressing the main problem tackled in this thesis and proposing a solution. Chapter 5 elaborates on this proposed approach and its initial application on a dataset derived from real data with simulated pathology. Following that, we demonstrate the use of our method in real-world scenarios, detailed in Chapter 6. This includes its application in patients with multiple sclerosis and the evaluation of a drug's efficacy in two Phase II clinical trials. Finally, we present an additional application of the proposed model, where, instead of computing structural connectivity, we leverage the model’s estimated lesion damage to assess its correlation with motor disability, both at baseline and longitudinally. The thesis concludes with a summary of findings and final remarks (Chapter 7). The works presented in this thesis were conceptualized and developed by me in collaboration with my supervisors. I also handled the data processing, the statistical analysis, and the interpretation of the results under their guidance. However, the dedicated staff of the Basel research group carried out data acquisition and the initial pre-processing.