Liver segmentation in CT images is a complex and challenging process. This is due to wide variability of liver sizes and shapes from one image to another, in addition to the inhomogeneity of the gray-level within the liver region and the low contrast to the background levels. In this paper, a fully automatic approach for liver segmentation is introduced. The approach consists of three main stages; pre-processing, segmentation and post processing. Watershed segmentation algorithm is used in the main processing stage to detect the borders and edges accurately between the liver regions and the background. However, because of the over-segmentation caused by the watershed algorithm, region merging algorithm is applied in the post processing stage. The merging criteria were proposed to maximize the disparity between the liver regions and the background and in the same time to keep the variance of the gray-level in the liver regions under certain threshold. The algorithm achieved 91% overall accuracy when evaluated using CT images from the MICCAI dataset.
CITATION STYLE
Mohamed, A. S. E. D., Salem, M. A. M., Hegazy, D., & Shedeed, H. A. (2017). Improved watershed algorithm for CT liver segmentation using intraclass variance minimization. In Communications in Computer and Information Science (Vol. 756, pp. 164–176). Springer Verlag. https://doi.org/10.1007/978-3-319-67642-5_14
Mendeley helps you to discover research relevant for your work.