Deformable model-based image segmentation using region, statistical and shape information

Medical image segmentation is an important step in medical image analysis. Deformable model-based segmentation method is a popular method for this due to its natural ability to deal with topology changes and straightforward extension to higher dimensional images. This dissertation focuses on investigating medical image segmentation with deformable model-based methods.; The first part of this work presents a system for segmenting the human aortic aneurysm in CT angiograms (CTA). The system estimates a rough "initial surface," and then refines it using a level set segmentation scheme augmented with two external analyzers: the global region analyzer, which incorporates a priori knowledge of the intensity, volume and shape of the aorta and other structures, and the local feature analyzer, which uses voxel location, intensity, and texture features to train and drive a support vector machine classifier. We tested our system using a database of twenty CTA scans of patients with aortic aneurysms. Compared to the manual segmentation results, the mean values of volume overlap and volume error were 95.3% +/- 1.4% and 3.5% +/- 2.5% (s.d.), respectively. This preliminary study shows that the system is promising for assisting radiologists in abdominal aortic aneurysm segmentation, and may be of benefit to patients with aortic aneurysms.; One of the challenges for segmentation is prevention of leakage from one structure into an adjacent one. The second part of our work proposes a new directional distance aided (DDA) image segmentation method that can prevent leakage. At each iteration, the zero level set is extracted and using a new anti-shrinkage Gaussian smoothing operation. For each point on the zero level set, the directional distance term, defined as the vector starting from this point and pointing to its counterpart on the smoothed version of the zero level set, is calculated to measure its "degree of protrusion." The points that are considered to be protruding outward will experience slower growth in the next iteration compared to other points. We evaluated the new method by performing two 2D and two 3D experiments. Experimental results show that the new DDA method achieved promising performance in preventing leakage while preserving shape details.