LISTA DE ABREVIATURAS

Di(cid:27)usion tensor magnetic resonance imaging characterizes the three-dimensional distribution of water di(cid:27)usivity in biological tissues. The pro(cid:28)les of the distribution are represented by tensor (cid:28)elds containing local information about the structure and organization of tissues. For e(cid:30)cient analysis, however, measures that use full tensor information that enhance local di(cid:27)erences between structures at an acceptable computational cost are desired. In this thesis, we present a novel tensor similarity measure, the polynomial dot product. It is shown by comparisons with established similarity measures such as the J divergence, Riemannian metric, and Log-euclidean distance, that the polynomial dot product keeps the desired tissue di(cid:27)erenciation properties with the bene(cid:28)t of a reduced computational cost. Assessments were made with synthetic datasets and real datasets of the brain and heart, with and without abnormalities. Results from this work will enable the development of more e(cid:30)cient segmentation, regularization and registration algorithms which will aid reseachers in studies of microstructural characterization of normal-appearing tissue and in the presence of pathologies.