Pancreas 3D MRI Segmentation Based On Low Rank Decomposition And Multi Atlas

In recent years,the incidence of pancreatic cancer is increasing in China,and there is still a rising trend.Compared with other cancers,pancreatic cancer is highly malignant because it's difficult to be observed and almost cannot be cured.Traditional radiation therapy is a common method of treating pancreatic cancer,which irradiates with radiation to kill cancer cells,but also causes permanent damage to normal cell tissue.Precise radiotherapy is a new tumor radiotherapy technique and is less harmful to humans.It treats cancers through accurately locating tumor,precise planning,accurate dose calculating,and precise curing on a treatment machine under the guidance of magnetic resonance images.Precise radiotherapy requires a high precise pancreas segmentation in magnetic resonance images.With the development of medical imaging technology,three-dimensional magnetic resonance imaging has been widely used in clinical diagnosis and treatment,especially in image-guided radiation therapy.Fast and accurate segmentation of pancreas from three-dimensional magnetic resonance images is the key step to precise radiotherapy for pancreatic cancer.For pancreas segmentation from three-dimensional magnetic resonance images,this paper carried out the following work:1.Pancreas segmentation from three-dimensional magnetic resonance images based on lowrank decomposition and Hessian enhance was proposed in order to solve the problem of severe adhesion between the pancreas and the surrounding tissues and organs.The original three-dimensional magnetic resonance image sequence is decomposed into low rank sequence and sparse sequence by low rank decomposition method,and the adhesions phenomenon is effectively reduced in the sparse sequence.Then the low rank and sparse sequences were respectively used the Hessian matrix combined with the spatial anatomical information to enhance the pancreas.Next through the human-computer interaction to extraction the pancreas and repair the result based on the three-dimensional information of pancreas.The final extraction of the pancreas is obtained by fusing two result of the low rank and sparse sequences.This method is compared with two kinds of popular segmentation methods in medical image segmentation,which shows that the method has a good segmentation effect.2.The results of the pancreas segmentation often differ greatly in morphology.To solve this problem,a magnetic resonance pancreas segmentation method based on Multi-Atlas was proposed.This method uses the pancreas prior information contained in the Multi-Atlas,registering the images need to be segmented and template in the atlas respectively by rigid registration and non-rigid registration.The label can be moved to the images need to be segmented by registration.Then migrate label which sorted by the similarity measure and finally select some images need to be segmented with the most similar to atlas after registration to get the final results.This method is compared with the results of the first work,indicating that the method is more stable and can obtain a better shape of the pancreas,the accuracy of pancreatic segmentation is overall improved.