Atlas Based Automatic Liver 3D CT Image Segmentation On GPU Platform

There are more than one million death cases in patients with Hepatocellular Carcinoma each year. Liver surgery has been the preferred solution to cure. Volume, tumor location and other digital parameters of liver can give doctors a good feasibility study. To achieve the purpose, the first step is to segment liver. Due to traditional manual and semi-automatic segmentation need to spend a lot of time, automatic segmentation algorithm is urgently needed. But automatic segmentation of the liver is also a grand challenge: firstly, various liver's surface; secondly, the intensity of tissues around the liver is very close to that of the liver; thirdly, the existing of tumors and lesions increasing the difficulty of the segmentation; fourthly, the time consuming is far away for doctor's patience.In this paper, 3D non-rigid registration from a refined atlas which has the prior knowledge of liver-outline is used to do the segmentation. Firstly, twenty sets of training images are used to create an atlas. Secondly, the liver initial region is searched and located automatically. Thirdly, threshold filtering is used to enhance the robustness of segmentation. Finally, this atlas is non-rigidly registered to the liver in CT images with affine and B-spline in succession. After registration, Segmentation of the liver atlas can be represent segmentation of the target liver and then the liver volume was calculated The algorithm is transplant to GPU platform to speed up. Experiment results show that this method is feasible for clinical liver CT image segmentation.