A Hybrid Approach for Liver Segmentation

In this paper, we propose a hybrid approach for fully automatic liver segmentation in Computed Tomography (CT) data. The approach consists of four stages: first, an intensity-based partition is applied to obtain the soft-tissue regions. Second, a region-based texture classification is used to classify all the soft-tissue regions based on texture features and generate probability images. Third, an initial region of the liver for each patient is determined from the probability images. In the final fourth step, a 95% confidence interval determined from the intensities of the initial regions is used to detect the liver region in each one of the consecutive slices in which the liver appears.