Microstructure informed tractography with anatomical priors

It is known from brain anatomy that axons in the white matter are structured and organized in groups, called fascicles or bundles. Historically, these white matter structures have been studied in detail using invasive techniques, such as dissection, with the aim to elucidate their organization and possible effects of neurological diseases on brain connectivity. With the advent of magnetic resonance imaging, which is a noninvasive acquisition modality, today it is possible to perform "virtual dissections" using tractography. In fact, this technique allows estimating in vivo the macroscopic trajectory of these fascicles using 3D polylines called streamlines, and this unique ability opened a number of exciting possibilities to study the anatomy of the brain noninvasively and to characterize alteration in its structure due to pathology. However, tractography is not perfect and some limitations have recently been pointed out in many studies and received a lot of attention in the scientific community. One of the major problems is the high number of invalid white matter structures reconstructed, i.e., false-positive connections between cortical and subcortical regions, which were shown to potentially bias the characterization of brain connectivity. The main objective of this thesis is to investigate and propose new strategies to improve the anatomical accuracy of tractography. In particular, we extended a microstructure informed tractography framework by introducing the important anatomical prior that axons are organized in groups and using this information as a constraint to reduce ambiguities in the reconstructions. We tested different solutions and showed that it is possible to improve the specificity of connectivity estimates without affecting their sensitivity, both in numerical simulations and real brain data. We implemented a pipeline using a supervised tractography segmentation tool to extract bundles and evaluate the performance of our proposals using in vivo data. In the first proposal, we introduced the possibility of using the definition of groups of streamlines as a constraint in the optimization problem, which produces a big improvement in the connectivity generated with the reconstructions. Later in the second strategy, we extended the framework to use as a constraint a multilevel hierarchical organization of streamlines, which is able to reproduce the connectivity results removing implausible streamlines that are inside the groups that create the connections. Finally, we proposed a robust augmentation to the cost function of the framework that adjusts for the reliability of the original measurements. With the research performed in this thesis, we proved that adding information on the organization of the white matter benefits reconstructions made with tractography. Our proposals represent an additional step forward to improve the anatomical accuracy of tractography and our understanding of how different brain regions are interconnected.