AUTOMATIC TUMOR SEGMENTATION USING KNOWLEDGE-BASED TECHNIQUES

A system that automatically segments and labels glioblastoma-multiform e tumors in magnetic resonance images (MRI's) of the human brain is pr esented. The MRI's consist of T1-weighted, proton density, and T2-weig hted feature images and are processed by a system which integrates kno wledge-based (KB) techniques with multispectral analysis. Initial segm entation is performed by an unsupervised clustering algorithm, The seg mented image, along with cluster centers for each class are provided t o a rule-based expert system which extracts the intracranial region. M ultispectral histogram analysis separates suspected tumor from the res t of the intracranial region, with region analysis used in performing the final tumor labeling. This system has been trained on three volume data sets and tested on thirteen unseen volume data sets acquired fro m a single MRI system. The KB tumor segmentation was compared with sup ervised, radiologist-labeled ''ground truth'' tumor volumes and superv ised k-nearest neighbors tumor segmentations. The results of this syst em generally correspond well to ground truth, both on a per slice basi s and more importantly in tracking total tumor volume during treatment over time.