Automatic 3D segmentation and characterization of brain tissues in multiparametric MR image sequences.

In this paper a new segmentation method for automatic differentiation of normal and pathological brain tissues, based on multidimensional relaxation parameter histograms, is introduced. The developed histogram pyramid algorithm, which is an extension of histogram-based cluster analysis methods, is used for automatic analysis of multiparametric image data from several body slices simultaneously. The achieved segmentation results are improved by a following merging algorithm, which merges split tissue parts. Furthermore, tissue specific relaxation parameter values are computed within each 3D tissue segment and can be stored in a data base. Based on the tissue database, automatic tissue classifications are performed using statistical pattern recognition methods. Classified tissues are marked with tissue-specific colors and visualized in tissue class images. For clinical applications, the algorithms for 3D segmentation, visualization, and classification of tissue structures are integrated in the software system SAMSON (System for AutoMatic Segmentation and ClassificatiON of Tissue in Magnetic Resonance Tomography). With the software system SAMSON, a tissue database for intracranial tissues has been established based on the examination of 15 volunteers and 100 patients with neoplastic and other lesions--all verified histologically.