Study On The Method Of Liver Tumor Segmentation Of CT Images

The liver has a complicated anatomical structure with intrahepatic vessels system asthe largest solid organ in abdomen of the body. There are many kinds of diseases withhigh morbidity. Hepatocellular carcinoma constructs a serious threat to human life andhealth because of higher grade malignancy as the second cancer killer, the diagnosis andtherapy of which is a significant and difficult problem in medicine.With the development of modern imaging equipment and technology such as US,CT, MRI and PET-CT, the progress of hepatobiliary surgery improved the diagnosis andprognosis of patients. Making a perfect operation plan is always the pursuit aboutprecise surgery for the industry, based on the anatomical structure, evaluation of liverfunctional reserve and the extent of the lesion. This is involved in the field of medicalimage segmentation.On the brief review of the research background, achievements of medical imagesegmentation in recent years, this paper will formulate several mainstream methodtheories, such as medical image segmentation based on graph theory, medical imagesegmentation based on active contour, medical image segmentation based on regiongrowing, and medical image segmentation based on mathematical morphology.On the basis of the CT image characteristic of liver and the clinical demand of thehepatobiliary surgery, this paper is to propose a new image segmentation method for thesets of pipeline system of liver, after combining the image segmentation based on kernelgraph cut and the level set method. This new method can achieved fast segmentation ofliver and intrahepatic vessels, establish good groundwork for3D reconstruction to guidethe hepatobiliary surgery in feedback way.The main research context is as follows:（1）This paper reviewed the image principle of CT and MRI, dynamiccontrast-enhanced spiral CT performance of liver, and image reading and pre-processingbased on DICOM protocol, and summarized acquiring the clinical image of liver.（2）This paper reviewed several mainstream method theories, such as medicalimage segmentation based on graph theory, medical image segmentation based on activecontour, medical image segmentation based on region growing, and medical imagesegmentation based on mathematical morphology. Comparisons and analysis were madein terms of experiment results. （3）On the basis of analysis on image segmentation based on multi-regions,nuclear induction, namely mapping liver CT data to high dimensional space usingkernel function and extracting the tumor area using graph cut algorithm, effectivelyavoids building a complex model for different image areas. It also improves theversatility and popularization of graph cut algorithm.（4）On the basis of accurate initial contour generated by coarse segmentation ofmedical image in kernel graph cut, fine segmentation was conducted in the improved LiChunming model, and effectively improved the accuracy of segmentation result.（5）specific experiments were carried out in the above combining segmentationmethod by MATLAB and VS2005tools. Our results showed that the combination ofkernel graph cut and LCVmodel get a good effect, which optimized liver and tumorsegmentation of CT images to assist clinical diagnosis and treatment.