Probablistic-based framework for medical CT images segmentation

Abstract

Liver segmentation is a difficult process due to wide variability of livers shapes and sizes between patients and the intensity similarity between the liver and other organs. Liver segmentation from abdominal Computed Tomography (CT) images is very useful in many diagnostic and surgical processes. It is the essential step in many clinical applications. Medical decisions are rarely taken without the use of imaging technology such as CT, Magnetic Resonance Imaging (MRI), or Ultrasound Imaging (US). In this paper, an automated probabilistic-based framework for liver segmentation from abdominal CT images is presented. The framework consists of four stages; thresholding stage, superpixels construction stage, Bayesian network construction stage and region merging stage. We train and validate our model using 20 clinical volumes. We use the MICCAI dataset (Medical Image Computing and Computer Assisted Intervention for Liver Segmentation). MICCAI dataset is used in more than 90 researches.