Automatic delineation of white matter fascicles by localization based upon anatomical spatial relationships

摘要

Delineation of white matter fascicles is generally achieved with tractography by specifying seeding and exclusion regions of interest (ROIs) defined by anatomical landmarks. In practice, the most popular approach has been to manually draw the ROIs for each scan which requires extensive training, is strongly subject to inter- and intra-expert variability and is highly time consuming. Fully automatic localization of the ROIs is of central interest, particularly for white matter investigations involving a large number of subjects. In this work, we propose an original approach in which the ROIs are localized using the fuzzy set theory by discovering stable anatomical spatial relationships in the brain anatomy. Our approach relies on a learning procedure, in which stable relationships are identified from a limited number of training templates supplied with manually delineated ROIs. For a new subject, the spatial relationships are applied and the ROIs localized. We show that our approach enables successful automatic delineation of the ROIs in the individual. Importantly, we show that this localization is robust across subjects age.