Research On Key Techniques In Living Donor Liver Transplantation Surgical Planning System

The morbidity and mortality of liver cancer are very high in China, andcurrently the most effective treatment for this disease is liver resection andliving donor liver transplantation. Due to the complex intrahepaticanatomical structures such as the distribution of vessels, surgeons need tohave a clear understanding of intrahepatic structures before the surgery, inorder to achieve the optimal surgery plan and improve the success rate ofthe surgery. Living donor liver transplantation surgical planning system isof great significance for the achievement and implementation of surgeryplans. It can help surgeons reduce their dependence on experience andaccurately learn liver structures and the distribution of intrahepatic vesselsand tumors. It can also provide accurate quantitative parameters such asvolume of the liver and tumors, and the size of vessels. As a result, it canassist surgeons make the best surgery strategy and ensure the success ofthe surgery. Most of current surgical planning systems rely much on userinteractions, and they can hardly obtain robust results, which has limitedits clinical applications.To help surgeons achieve the optimal surgery plan, this paperinvestigated a new liver surgical planning system, which was aimed toobtain stable results for complex cases in clinical environments and toenhance its robustness and clinical applicability. The proposed systemconsists of the segmentation of liver and intrahepatic tissues, liver segmentapproximation and visualization. The most important work was thedevelopment of a liver shape prior and minimally supervised classificationbased segmentation framework, which was designed to address the challenges caused by the low contrast, blur edges and the influence oftumors and other metastasis.In the modeling of shape prior, this paper employed sparse shapecomposition (SSC) to deal with complex variations of liver shapes, whichcould hardly be addressed by traditional methods. The SSC represents aninput shape as a sparse linear combination of shapes in the shape repository,and it successfully overcome three challenges in shape modeling. Firstly, itdoes not need any assumptions on the statistical distributions of livershapes, and can effectively deal with complex variations of liver shapes.Secondly, SSC model the gross errors in the input shape explicitly, and isable to capture non-Gaussian errors, thus overcomes theunder-segmentation and over-segmentation in the input shape. Thirdly, theSSC has an excellent ability to reserve local details in the input shape, sothat it can realize accurate modeling of shape priors. The paper alsoinvestigated fast solutions for the optimization problem of SSC. TheHomotopy method was employed to continuously transform the L1optimization problem into a series of simple problems which were fast tocompute, and it remarkably improved the efficiency of the SSC.This paper realized the intrahepatic vessels based liver segmentapproximation method. This method automatically obtained the skeleton ofvessels from the segmentation results and then converted it to a tree’sstructure. Surgeons only need to identify root nodes of vessels for liversegments and then liver segment approximation can be finished atomically.Only few user interactions are required and the results are patient-specific.On the basis of liver segmentation and segment approximation, theproposed system visualizes different tissues and liver segments, whichprovides surgeons with anatomical structures in a detailed and intuitiveway. The system also provides quantitative analysis of liver segments andtumors, such as the measurement of volume, so that it can help surgeonslearn the liver’s structures and functions more comprehensively. Finallythe best surgery strategy will be achieved, which is the most beneficial to both the donor and recipient.The proposed liver surgical planning system was validated by a largeamount of experiments and analysis. Results of experiment showed thatthe SSC was very effective to model shape priors for livers. Comparedwith Principal Component Analysis (PCA), the SSC has a betterperformance in dealing with complex shape variations, non-Gaussianerrors and local details. The experiment on18patients showed that theproposed segmentation framework could achieve very accuratesegmentation results. The average symmetric surface distance evaluationfor hepatic parenchyma, portal veins, hepatic veins and tumors was1.07±0.76mm，1.09±0.28mm，0.92±0.35mm and1.13±0.37mm, respectively.It was shown that the proposed liver surgical planning system remarkablyimproved the robustness when dealing with complex clinical liver shapes.It can not only obtain accurate segmentation results for healthy personsand common patients, but also has a high robustness when dealing withspecific patients with large variations of liver shapes in complex clinicalenvironments, thus it has an excellent clinical applicability.