University of Houston

Department of Computer Science

In Partial Fulfillment of the Requirements for the Degree of
Doctor of Philosophy

Alok Bandekar

Will defend his dissertation

A Unified Knowledge-based Segmentation
Framework for Medical Images

Abstract

Cardiovascular disease is the leading cause of death in developed countries and is rapidly becoming the number one killer in the world. There is a critical need to identify patients who are asymptomatic but are at risk for coronary events. Noninvasive imaging techniques such as non-contrast cardiac computed tomography (CT) can provide information to help in assessing the individual risk of patients. Recent evidence indicates that pericardial fat and abdominal fat are significant cardiovascular risk factors. There is an increased demand for automated, robust, and accurate quantitative measures of these tissues in the heart region. An accurate segmentation allows for quantitative analysis. Thus, the development of fully automated methods which will provide unbiased and consistent results is highly desirable.

This dissertation proposes a unified knowledge-based medical image segmentation framework for CT data. Our framework consists of a training phase and a deployment phase. In the training phase, we construct a 3D probabilistic atlas of the thoracic region which associates every voxel with a specific organ. We select optimal texture-based features to discriminate between organs. Furthermore, we define relationships between organs in the atlas using prior anatomical knowledge and set image analysis methods for segmentation of specific tissues or organs in the region of interest (ROI). In the deployment phase, we initialize the ROI using the probabilistic atlas and use the predefined methods for segmentation of organs. Specifically, this framework has been applied to automatic segmentation of fat tissue in abdominal and cardiac CT data. This unified knowledge-based segmentation framework provides automatic, standardized measurement of important cardiovascular risk factors in routine cardiac CT imaging.

Faculty, students, and the general public are invited.