Reproducible segmentation of white matter tractograms using artificial intelligence and spatial fuzzy sets

Tractography is the only in-vivo technique that allows the extraction of white matter fibers from the brain in a non-invasive way. It is used as a support in neurosurgeries and as research tool to study structural connectivity in both healthy and pathological subjects. It generates, from diffusion Magnetic Resonance (MR) volumes, millions of 3D polylines that are estimates of the path followed by real axons. One important clinical application is the automatic segmentation of precise and anatomically well-defined white matter tracts from whole-brain tractograms, but this task is particularly difficult to accomplish since the large number of streamlines makes the computational costs very high and the anatomical tracts definitions found in literature are usually vague or imprecise. The purpose of this master thesis is to model, in a reproducible way, the tracts anatomical definitions using fuzzy logic and to extract them from whole-brain tractograms of multiple subjects with the help of clustering techniques. The whole process have been implemented on a desktop computer with normal characteristics paying particular attention on memory and time issues, so that it can be ideally used by an external operator without the need of a high performance machine and in a reasonable time. This thesis, firstly intoduces the physics of the diffusion MR and the importance of tractography as a novel support in nowadays research and clinical applications. It then explains the state of the art segmentation techniques that can be used to extract individual fiber bundles. It briefly describes the tool used to model the anatomical definitions and proceeds with the explanation of the algorithm created for tracts segmentation, whose aim is to lead the fuzzy set approach from a proof of concept state to a validated technique. Finally, the results are presented together with the ones obtained using the state-of-the-art techniques to assess the validity of the outputs and to show the improvements achieved.