Level set based prior models for image segmentation and analysis

The role of medical imaging is expanding and the medical image analysis community has become preoccupied with the challenging problem of creating quantification algorithms that make full use of the information in the image data. Image segmentation still remains one of the most fundamental and difficult problems in medical image analysis. The noise and artifacts presented in medical images, combined with the complexity and variability of the anatomic shapes of interest, limit the effectiveness of simple image processing routines. Statistical prior models provide application-specific context to the problem by incorporating information derived from a training set consisting of instances of the problem along with the solution. In this thesis, we explore the benefits of statistical prior models for medical image segmentation and functional Magnetic Resonance Image (fMRI) analysis.; A novel method for the segmentation of multiple objects from 3D medical images using inter-object constraints is presented. Our method is motivated by the observation that neighboring structures have consistent locations and shapes that provide configurations and context that aid in segmentation. A Maximum A Posteriori (MAP) estimation framework is used to combine both the constraining information provided by neighboring objects as well as image gray level information. To further take advantage of the available prior information, we also include a shape-intensity joint prior knowledge into image segmentation. Inspired by the observation that the shape and the gray level intensity of objects in an image have consistent relations, we introduce a representation of the joint density function of the object and the gray level values. We formulate these segmentation models in terms of level sets as opposed to landmark points of the shape. In our segmentation methods, the level set distribution model (LSDM) is used as an approximation of the distribution of the object shape. We evaluate its performance by comparing it with the point distribution model using the Chi-square test. The experiment results show that the two models are quite similar in describing the main variations of regular object shape. The two distributions are statistically indistinguishable.; The idea of using prior models in anatomical image segmentation and analysis can also be applied to the analysis of functional Magnetic Resonance Images (fMRI). We develop an integrated framework combining both anatomical as well as functional prior information. This framework has been successfully used in functional parameter estimation (used for activation detection) with additional structural information, as opposed to the traditional functional-only estimation technique. The activation estimation resulting from our model provides a more clustered, coherent and stronger activation zone.; Our segmentation methods have been applied to a variety of imaging modalities, including magnetic resonance and computed tomography. Both our segmentation models and functional MR analysis models have been applied and validated using various medical studies, such as neuroimaging and analysis of autism spectrum subjects. We find the methods to be robust to noise, able to handle multidimensional data while avoiding the need for point correspondences during the training phase.