An adaptive hybrid technique for pancreas segmentation using CT image sequences

This paper briefly introduces a novel semiautomatic method to segment the pancreas volume from CT image sequences. Existing hybrid Level Set Methods (LSM) is applicable for extracting the full size pancreas from single abdominal CT image. To extract shape and size varying pancreas from continuous CT image sequences, an adaptive hybrid level set method is proposed. Proposed method uses Fast Marching Method (FMM) for rough segmentation followed by Distance Regularized Level Set Method (DRLSM) for final pancreas segmentation from single CT image. To make it adaptive for a set of CT images, the optimal values of time threshold and iterations number for FMM and DRLSM respectively are computed automatically. The proposed method is evaluated on a dataset of 9 abdominal CT image sequences, which includes 140 CT slices. The performance of proposed method is quantitatively evaluated by comparing the segmentation results with ground truth CT image slices, in which pancreas region is manually marked by experienced radiologist.