Automatic Segmentation and Classification of Multiparametric Image Data in Medicine

Abstract

In this paper cluster analysis and classification methods in combination with image processing algorithms are presented for automatic segmentation and classification of tissues in medical images. Based on multiparametric image data, which are typically generated in magnetic resonance tomography (MRT), different tissue structures are segmented by using pyramidal histogram analysis methods in combination with a merging algorithm. Though the amount of data is large in the medical images considered, several images can be segmented simultaneously in an efficient manner. After segmentation, the 3-dimensional distribution of tissues in space is visualized. Furthermore, the tissue-specific relaxation parameter values of each segmented tissue are computed to establish a tissue data base. Based on the extracted relaxation parameter values an automatic classification of unknown tissue structures is performed using statistical classifiers. The classified tissues are marked with tissue-specific colours and visualized in tissue class images. For diagnostic support in clinical applications, the algorithms for tissue segmentation, classification and visualization are integrated in the software system Samson (‚System for Automatic Segmentation and Classification of Tissue in Magnetic Resonance Tomography’) [Handels 1992].KeywordsTissue StructureAutomatic SegmentationRelaxation ParameterCluster IndexTissue SegmentationThese keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.