Recognizing Deviations from Normalcy for Brain Tumor Segmentation

Abstract

Abstract.A framework is proposed for the segmentation of brain tumors from MRI. Instead of training on pathology, the proposed method trains exclusively on healthy tissue. The algorithm attempts to recognize deviations from normalcy in order to compute a fitness map over the image associated with the presence of pathology. The resulting fitness map may then be used by conventional image segmentation techniques for honing in on boundary delineation. Such an approach is applicable to structures that are too irregular, in both shape and texture, to permit construction of comprehensive training sets. The technique is an extension of EM segmentation that considers information on five layers: voxel intensities, neighborhood coherence, intra-structure properties, inter-structure relationships, and user input. Information flows between the layers via multi-level Markov random fields and Bayesian classification. A simple instantiation of the framework has been implemented to perform preliminary experiments on synthetic and MRI data.

Support Vector Random Fields for Spatial Classification

“…We focused on the task of brain tumor segmentation in MRI, an important task in surgical planning and radiation therapy currently being laboriously done by human medical experts. There has been significant research focusing on automating this challenging task (see [10]). Markov Random Fields have been explored previously for this task (see [10]), but recently SVMs have shown impressive performance [11,12].…”

“…There has been significant research focusing on automating this challenging task (see [10]). Markov Random Fields have been explored previously for this task (see [10]), but recently SVMs have shown impressive performance [11,12]. This represents a scenario where our proposed Support Vector Random Field model could have a major impact.…”

Brain Tumors: How Can Images and Segmentation Techniques Help?

“…This is prohibitive for brain tumor segmentation applications and several authors have proposed MRF-based strategies to deal with this disadvantage by integrating information about edges or region properties. (Gering et al, 2003) proposed a framework for brain tumor segmentation, extending EM-based segmentation with region-level properties and deriving a multi-level MRF. Another iterative MRF-based segmentation method was proposed by (Chen et al, 2003) in which a Gibbs Prior model and a deformable model are combined.…”

Automatic Segmentation of Cardiac MRI

“…More sophisticated inter-layer communication is possible using multi-level Markov random fields or Bayesian belief propaga-tion networks. A more detailed discussion of CDN is available in [5] which develops the initial proposal in [4] for the application of segmenting brain tumors.…”