Parallel spatial-domain liver segmentation of CT abdominal images

Abstract

Liver segmentation in abdominal CT images is vital prior to liver disease diagnosis and computer-aided diagnosis. Since an abdominal CT image is some complex in which the pixels of the liver region and some other adjacent parts often distribute in the same value range, commonly used value-based segmentation methods have difficulty in dealing with this situation. In order to overcome this shortcoming, in this paper we present a spatial-domain segmentation method in which both the value information of the pixels and the spatial relationship between them are utilized in the segmentation process and meanwhile the spatial information, such as the area, boundary or location, of the target is extracted, thus the target region to be segmented can be distinguished from the other parts in same value range by use of the spatial information. Furthermore, a parallel algorithm is designed and implemented on GPU for improving the computation efficiency of the spatial-domain segmentation method. Experiment results obtained in the work confirm the effectiveness of the new segmentation algorithm.