In recent years, the incidence of pancreatic cancer, one of the cancers which have the worst prognosis, is increasing year by year. Usually, patients with pancreatic cancer are found in the advanced stage. So, the radiotherapy and chemotherapy are mostly used in combination as a palliative in clinical treatment. In addition, MRI guided radiation therapy with integrated MRI on the treatment machine has become technically feasible very recently. Consequently, the pancreas segmentation technology has become much more critical. Unfortunately,the research on this subject is almost empty. Therefore, based on the problem of the automatic pancreas segmentation in MRIs, we have mainly completed the following works:1. A manifold clustering constrained dictionary learning method(MCDL) was developed for the automatic pancreas segmentation in MRIs. Due to the considerable individual pancreas difference and the complex anatomy around pancreas, object and background dictionaries are trained using the K-SVD algorithm to obtain the preliminary segmentation results. But it can’t deal with the fuzziness of MRIs. Therefore, a manifold clustering method is used to constrain the rough preliminary segmentation. The final results were compared with three state of the art algorithms and proved to be superior in accuracy.2. A clustering method based on three dimensional gray-gradient features was developed for the three dimensional segmentation of pancreas in MRIs. The two dimensional gray-gradient features are introduced into the three dimensional space to obtain features of voxels. The three dimensional gray-gradient features of voxels are then clustered to get a preliminary segmentation with some redundant tissues. So some three dimensional mathematical morphology operations are used to remove the redundant tissues. And the final segmentation results prove to be accurate. |