Using Fuzzy Information in Knowledge Guided Segmentation of Brain Tumors

Matthew C. Clark Lawrence O. Hall Dmitry B. Goldgof
Department of Computer Science and Engineering
University of South Florida
Tampa, Fl. 33620
hall@csee.usf.edu

ABSTRACT

This paper presents a system that integrates a knowledge-based system with unsupervised fuzzy clustering to automatically segment and label glioblastoma multiforme tumors in magnetic resonance slices of the human brain. Each slice is initially segmented by an unsupervised fuzzy c-means algorithm. The segmented image, along with a set of tissue cluster centers and some knowledge gathered during ``pre-processing,'' is then given to the rule-based system which uses model-based recognition techniques and further fuzzy clustering to iteratively locate tissues of interest. These ``focus-of-attention'' tissues are analyzed by matching them with expected characteristics. Further fuzzy reclustering is aided by the use of initialization and training data created by the knowledge system.

This system has been tested on thirteen slices acquired from a single MR coil. Final tumor segmentation for each slice compares favorably with supervised, hand-labeled ``ground truth'' tumor images. Partial labeling of non-tumorous tissues was also achieved.

• Introduction
• Domain Knowledge
  • Slices of Interest for the Study
  • Cluster Center Distribution in Feature Space
  • Anatomical Knowledge
• Knowledge Applications
  • Focus-of-attention
  • Merging Clusters
  • Tissue Modeling and Matching
  • Prototypical Patterns
• Classification Stages
  • Pre-Processing and Recluster Mask Building
  • Initial Reclustering
  • CSF and Gray Matter Identification
  • Pathology Mask Creation and Reclustering
  • Tumor Segmentation and Labeling
• Results and Discussion
  • Results
  • Discussion
• References
• About this document ...