1997-2003. Neurosurgery department, Clermont Hospital
The ability of precisely quantifying cerebral tissue volumes in various diseases (Alzheimer, hydrocephaly…) must make it possible to largely improve the diagnosis, the treatment and the general understanding of pathologies. We proposed in 2000 a unsupervised fuzzy classification scheme for brain tissue segmentation. The boundary between brain in an MR image is smooth or fuzzy rather than sharp or crisp. Moreover, noise, partial volume effect and anatomical variations within pure tissue (White Matter,Gray Matter, Cerebrospinal Fluid) activities introduce uncertainty and fuzzyness in the image. We thus used a fuzzy clustering algorithm rather than a crisp one, and particularly the possibilistic clustering algorithm (PCA). PCA allows to interpret memberships as absolute degrees of belonging, whereas they are similar to degrees of sharing or relative memberships in the case of FCM. Because of the complex nature of (GM/WM) or (GM/CF) interfaces in MR images, a significant number of voxels contains a mixture of two or three tissues.