The Research of Key Techniques on Computer-Aided Planning before Liver Transplantation
|School||Zhejiang University of Technology|
|Keywords||liver transplantation computer-aided planning medical image segmentation active appearance model statistical shape models pre-location|
In the 21st century, organ transplantation will be more and more prominent in the field of surgery, especially the liver transplantation. Before liver transplantation surgery, the computer-aided planning is an important aspect for the operation. It plays an important guiding role in the follow-up operation work. Medical liver image segmentation, liver’s three-dimensional reconstruction and liver volume measurement are the related technologies of the computer-aided planning. The liver image’s segmentation play a crucial role in the following operation, and it is the major difficulty in research that restrict the development of computer-aided diagnosis (CAD), and it’s also the main content of this article. Medical image segmentation is a classical ill-posed problem in the field of pattern recognition. It has not a better solution now.?There are thousands kind of Medical Image Segmentation methods, but it is hard to find a general method for medical image segmentation. Most theories and methods are proposed for a particular application and they have obvious limitations. Deformable models in medical image segmentation method is the most commonly used method, has been extensive researched, the statistical model method created by British scholar Cootes is the representative.This paper mainly studied the active appearance model of statistical modeling method, and firstly applied the model to the real liver CT data. We built a statistical appearance model of liver. We got a better result when we applied the model to the testing set data of liver CT; Then, our paper evaluate this AAM model using the traditional evaluation method. The result of evaluation shows that active appearance model performs robustly when we apply it to the different testing set data; Finally, we find some problem when we are building and applying the model. The first problem is that the mean shape selection is unscientific, and the second problem is that some liver shapes are irregular in the testing set. We analyzed the problem deeply and then we proposed two new solutions for the trouble. The first solution is genetic algorithm based on the feature points of liver shape, another is pre-location algorithm. At the experimental section, we used the variation genetic algorithm to generate the mean shape of our liver training set. Then, we combined the pre-location algorithm and active appearance model and applied it to some testing data which has irregular shape. The results are both ideal.