Model-Based Segmentation of Medical Images

Motivation
	Segmentation of medical images is an important first step in the analysis of medical images. Even though a lot of research has been conducted on the segmentation of medical images, existing tools still require substantial amounts of user guidance to produce reasonable results. Segmentation problems are particularly difficult and challenging when the desired regions are not homogeneous in intensity and their region boundaries are not distinctive enough. The model-based approach uses a model to guide the segmentation algorithm. WIth the use of models that capture human knowledge, the segmentation algorithms become more intelligent, reducing the need for user guidance.
Research Goal
	Our research goal is to apply the model-based approach to the segmentation of complex anatomical structures.
Atlas-Based Segmentation
	An atlas is used to describe the shapes of the anatomical parts in the abdomen and their spatial relationship. A registration algorithm deforms the atlas to align with the boundaries of the anatomical parts. In this way, the segmentation of the anatomical parts can constrain each other, resulting in more accurate and robust segmentation.
	Sample Results The atlas contains three parts: liver (red), stomach (green), and spleen (yellow). >>> video demo Atlas (red) overlaid onto input image Segmented left femur (black)
Researchers
>>	Mr. Ding Feng, Ph.D. student, Dept. of Computer Science, Natioanl University of Singapore
>>	A/Prof. Leow Wee Kheng, Dept. of Computer Science, Natioanl University of Singapore
>>	A/Prof. Shih-Chang Wang, Dept. of Diagnostic Radiology, Natioanl University of Singapore
>>	Dr. Howe Tet Sen, Dept. of Orthopaedics, Singapore General Hospital
Publications
>>	F. Ding, W. K. Leow, and T. S. Howe. Automatic Segmentation of Femur Bones in Anterior-Posterior Pelvis X-ray Images. In Proc. Int. Conf. on Computer Analysis of Images and Patterns, 2007, pp. 205-212.
>>	F. Ding, W. K. Leow, S.-C. Wang. Segmentation of 3D CT Volume Images Using a Single 2D Atlas. In Proc. First International Workshop on Computer Vision for Biomedical Image Applications (CVBIA 2005), LNCS 3765, (in conjunction with Int. Conf. on Computer Vision, 2005), pages 459-468, Springer, 2005.

3 July 2016