3D Segmentation of MR Brain Images into White Matter, Gray Matter and Cerebro-spinal Fluid by Means of Evidence Theory
Résumé
We propose an original scheme for the 3D segmentation of multi-echo MR brain images into white matter, gray matter and cerebrospinal fluid. To take into account complementary, redundancy and eventual conflicts provided by the different echoes, a fusion process based on Evidence theory is used. Such theory, well suited to imprecise and uncertain data, provides great fusion tools. The originality of our method is to include a regularization process by the mean of Dempster's combination. Adding neighborhood information increases the knowledge. The segmentation is more confident, accurate and efficient. The method is applied to simulated multi-echo data and compared with method based on Markov Random Field theory. The results are very encouraging and show that Evidence theory is well suited to such problematic.