Improved level set method for spinal vertebrae segmentation

• Main Author: Lim, Poay Hoon
• Published: University of Nottingham 2013
• Subjects: 616.73075
• Online Access: http://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.594415

Accurate segmemation of spinal vertebrae is important in the study of spinal related disease or disorders such as vertebral fractures. Identifying the severity of fractures and understanding its cause will help physicians determine the most effective pharmacological treatments and clinical management strategies for spinal disorders. Detection and segmentation are the crucial steps towards this quantitative framework. Although these quantitative image analysis techniques have received increasing interest recently, accurate detection and segmentation methods are still lacking. The complexity of vertebrae shapes, gaps in the cortical bone, internal boundaries, as well as the noisy. incomplete or missing information from the medical images have undoubtedly increased the challenge. Level set methods are effective for image segmentation. They are well-known for their ability to handle abrupt topological changes. However, the methods suffer from limitations such as slow convergence and leaking problems. As such, over the past two decades, the original level set method has evolved in many directions, including integration of prior shape models into the segmentation framework. For highly challenging medical images, incorporating prior shape into the segmentation framework have shown great success in recent years. In this work, a new shape model and an associated energy for level set segmentation is proposed. The proposed shape model presents a new dimension to extract local shape parameter directly from the shape model, which is different from previous work that focused on an indirect manner of feature extractions and in global sense. For the 2D and 3D segmentation of spinal vertebrae, a new segmentation framework combining the use of Willmore functional and kernel density estimation is proposed. While the kernel density estimator provide a dissimilarity measure with the shape prior during the level set evolution, the Willmore energy helps to regularize and smoothen the surface in the process. The introduction of Willmore functional into the 3D segmentation framework has solved the commonly encountered irregularity problem on extracted surface. Experimental results clearly demonstrate the feasibility of the proposed framework when tested on CT images of spinal vertebrae. The ultimate goal of this work is to provide a quantitative platform for efficient and accurate diagnosis of spinal disorder related diseases.