MR brain image segmentation by growing hierarchical SOM and probability clustering
Andrés Ortíz, J. M. Górriz, Javier Ramı́rez, Diego Salas-Gonzalez · Electronics Letters · 2011
A fully automatic tool to assist the segmentation of brain magnetic resonance images (MRI) is presented. Thus, the figured out regions can be evaluated for the diagnosis of brain disorders. The main problem to be handled consists in discovering different regions on the image without using apriori information. The new approach consists in hybridising multiobjective optimisation for feature selection with a growing hierarchical self-organising map (GHSOM) classifier and a probability clustering method. The segmentation results yield average overlap metric values of 0.32, 0.75 and 0.69 for white matter, grey matter and cerebrospinal fluid, respectively, over the Internet Brain Segmentation Repository database. These results mean an improvement over the values reached by other existing techniques.