A new expectation-maximization framework for partial volume segmentation of medical images

Segmentation of medical images is an approach which is widely used in medical settings to produce a qualitative and quantitative analysis of human internal anatomy through tissue classification. There are two basic approaches in the segmentation of medical images. Hard segmentation classifies each voxel as a single tissue type. However, due to the limited spatial resolution of imaging equipment, some voxels contain multiple tissue types, especially those voxels near the tissues borders. This is a major limitation for the traditional hard image segmentation. There is another group of segmentation methods that uses an assumption that each voxel can be composed of several tissue types. These methods are called soft (partial volume) segmentation methods.; This work investigates a new partial volume (PV) image segmentation framework. The new framework utilizes an expectation-maximization (EM) algorithm to iteratively estimate tissue fractions in each image voxel and statistical model parameters of the image data. A Markov random field (MRF) model is used to utilize imaging spatial information.; Main medical imaging principles, segmentation techniques and segmentation principles used in the algorithm are discussed in chapter 1 and chapter 2 of this dissertation. A new partial volume algorithm and its multi-spectral data extension are presented in chapter 3. A validation of the new algorithm, as well as comparison study of the algorithm with other segmentation techniques, is analyzed in chapter 3 and chapter 4.