Tissue segmentation by high resolution MRI: improved accuracy and stability

Tissue segmentation based on 2D and 3D feature maps derived from high resolution MR images was performed in experimental brain edema in cats, normal humans and those with brain tumors and MS. Statistical and anisotropic diffusion filters were applied to the data. The k-nearest neighborhood segmentation algorithm was utilized. Segmentation based on a 3D feature map was found to be much better than that based on a 2D feature map (p<0.01). Inter-observer variability ranged between 6.6% and 9.3% and intra-observer variability ranged between 8.3% and 10.4%. The use of high resolution MRI resulted in detection of new lesions which well correlated with neuropsychological exams. In conclusion, the presented technique is an accurate, stable and promising method of tissue characterization