Liver Segmentation Based on Expectation Maximization and Morphological Filters in CT Images

Segmentation of liver tissues in CT image is particularly challenging due to the anatomic complexity. An integrated model based statistical learning and morphology operations are presented in the paper to simplify the liver segmentation procedure and achieve qualified results simultaneously. The proposed scheme consists of two subroutines: initial segmentation and refinement. The former estimates statistical parameter vector of mixture Gaussian distribution by the EM algorithm and take initial classification based on maximum probabilities. The latter makes further refinement using morphological filters to remove foreign components and apply hole-filling routine. Experimental results show the ability of the proposed algorithm to accurately segment the liver structure in presence of liver tumor and other anatomic organs, and suggest its suitability to other medical image segmentation tasks.