Synergistic hybrid image segmentation: Combining model and image-based strategies

The focus of this thesis is practical model-based image segmentation. The target application in mind is segmentation and separation of the individual bony components at a joint for studying joint motion. This thesis examines this problem in three successive stages: (1) how best to combine model and image based strategies for 2D segmentation; (2) extending these to 3D segmentation; (3) utilizing these to segment (track) the same object in images corresponding to different positions of the joint.; Stage (1): Active Shape Model (ASM) is a widely used statistical model-based method for detecting and delineating structures in medical images. It efficiently searches images with a flexible and compact model by using prior knowledge derived from a training set. However, it is limited by accuracy because the segmentation results are parametric descriptions of the identified shape and they do not often match the perceptually identified boundary in images optimally. These inaccuracies will pose problems, especially for the subsequent analysis of medical images. On the other hand; purely image-based optimal boundary detection segmentation methods, such as live wire, perform well in delineation of boundaries of objects in images, although they require boundary recognition help from the user. Therefore, in this thesis, two such strategies are proposed, live wire active shape models (LWASM) and oriented active shape models (OASM) to actively exploit the synergy between model-based and image-based methods during segmentation seems to be an effective strategy.; Stage (2): ASM has proven to be an effective segmentation approach for medical images, most successfully in 2D objects with fairly consistent shape. Several difficulties arise when extending 2D ASM to true 3D. In this thesis, we apply 2D OASM for 3D segmentation in medical images. In this pseudo-3D OASM method, a small number of 2D statistical models are utilized to capture the shape variation within slices and different individuals. Each model can be matched rapidly to new images using the OASM algorithm. We demonstrate its application in the 3D segmentation of different anatomical structures in MR and CT images.; Stage (3): There are several medical application areas that require the segmentation and separation of the component bones of joints in a sequence of images of the joint acquired under various loading conditions, our own target area being joint motion analysis. This is a challenging problem due to the proximity of bones at the joint, partial volume effects, and other imaging modality-specific factors that confound boundary contrast. In this thesis, a two-step model-based segmentation strategy is proposed that utilizes the unique context of the current application wherein the shape of each individual bone is preserved in all scans of a particular joint while the spatial arrangement of the bones alters significantly among bones and scans. In the first step, a rigid deterministic model of the bone is generated from a segmentation of the bone in the image corresponding to one position of the joint by using any of the above methods. Subsequently, in other images of the same joint, this model is used to search for the same bone by minimizing an energy function that utilizes both boundary- and region-based information. We demonstrate in both MRI and CT of the tarsal complex and the cervical spine, how this strategy can be used to segment and separate bony components of a complex joint.