Oral Comprehensive Exam Details
Human brain MR image segmentation
| Author: | Kai Li |
|---|---|
| Date: | January 28, 2005 |
| Time: | 15:00 |
| Location: | 220 Deschutes |
| Committee: | Allen Malony (Chair) Kent Stevens Dejing Dou |
Abstract
This paper gives a survey on methods for human brain MR image segmentation. Traditional image segmentation methods, including thresholding, edge-based methods, region-based methods, classifiers, and clustering methods, are insufficient as solution approaches to the brain MR image segmentation problem. In the past two decades, the improvement in the performance of brain MR segmentation is mainly attributed to the incorporation of prior knowledge in the traditional image segmentation methods. The prior knowledge includes spatial knowledge, geometrical knowledge, and topological property. Several approaches can be used to incorporate prior knowledge. One approach extends the traditional statistical segmentation methods with the Markov Random Field model. Another approach focuses on the deformable models, which can be seen as as extension of traditional edge-based and region-based methods with consideration of geometrical knowledge, typically in the form of curvatures, of the target object. Human brain MR image segmentation poses additional difficulties and requirement with respect to general image segmentation problems. Special prior neuroanatomical knowledge, which has not attracted enough attention from the literature, suggests a promising direction for our research.