Robust segmentation of the thalamus using Kohonen algorithm from diffusion tensor images

J. Dauguet, V. Frouin, Y. Cointepas, D. Hervé, N. Ayache, P. Hantraye CEA, Service Hospitalier Frédéric Joliot, Orsay, France, INRIA, Epidaure Project, Sophia Antipolis, France Synopsis : Most of the neuroscience applications of diffusion tensor imaging (DTI) concern the tracking of fibers [1], but it may be used for tissue segmentation in structures like the thalami where classical imaging yields no contrast between main nuclear groups [2]. In this work, we describe the robust processes to derive parameters from raw DTI images and identify the different kinds of information that can be used for the segmentation process. This process is performed using Kohonen auto-organized maps combined with a hierarchical ascending classification.