Shape-Based Liver Segmentation Without Prior Statistical Models
In this work, we introduce a shape-based liver segmentation approach. However, unlike the other shape-based approaches, this approach is model-free, and it does not require prior shape or intensity model construction. In contrary, we exploit the relation
- PDF / 2,368,459 Bytes
- 37 Pages / 439.37 x 666.142 pts Page_size
- 87 Downloads / 181 Views
Abstract In this work, we introduce a shape-based liver segmentation approach. However, unlike the other shape-based approaches, this approach is model-free, and it does not require prior shape or intensity model construction. In contrary, we exploit the relation between consequent slices in multi-slice CT images to estimate and propagate shape and intensity constrains. Then, these constrains are integrated with a shape-based graph cut algorithm to extract the liver object in each slice. This approach needs a simple user interaction and it eliminates the burdens associated with model building like data collection, manual segmentation, registration, and landmark correspondence. Moreover, it is talented to deal with complex shape and intensity variations. This model-free approach was evaluated on 50 CT images from three different datasets with several liver abnormalities, including tumors and cysts, and it achieved high average gauged scores of 80.4, 79.2, and 81.7 for these datasets.
Introduction Liver tumors are one of the most common causes of death over the world [1], and the accurate diagnosis of these tumors helps to reduce their burden. It has shown that the utilization of computer-aided diagnosis (CAD) systems can greatly improve tumor diagnosis [2, 3], and it is useful for treatment planning, especially for liver transplantation and tumor ablation. In liver CAD systems, the liver segmentation is
A. Afifi (*) Faculty of Computers and Information, Menofia University, Shibin Al-Kawm, Menofia, Egypt e-mail: [email protected]; [email protected] T. Nakaguchi Graduate School of Engineering, Chiba University, Japan, 1-33, Yayoi-cho, Inage-ku, Chiba 263-8522, Japan e-mail: [email protected] A.S. El-Baz et al. (eds.), Abdomen and Thoracic Imaging: An Engineering & Clinical Perspective, DOI 10.1007/978-1-4614-8498-1_11, © Springer Science+Business Media New York 2014
279
280
A. Afifi and T. Nakaguchi
the first and essential process, and its accuracy is of special significance. However, this process is difficult because of low contrast between the liver and surrounding tissues, great differences in liver shape and intensity, and the existence of liver abnormalities. Hence, the conventional segmentation methods cannot produce adequate results. In literature, there are many attempts to solve the liver segmentation problem and various approaches have been proposed, including intensity- or texture-based approaches, deformable and statistical model-based approaches, and probabilistic atlases-based approaches. In the intensity-based approaches, one or multiple intensity thresholds, region growing, or watershed methods are applied to extract an initial binary volume which consequently refined using morphological filters or knowledge-based approaches. In [4], a predefined threshold was utilized on a simplified image to determine the initial liver area, and then it was refined using morphological filters and deformable contours with gradient information. Although this method showed accurate volume measure
Data Loading...
