2-D MRI Pancreas Segmentation Based On Transfer Learning

With the continuous development of the economic and the rising of people’s living standard, people begin to pay more and more attention to their own health problems. Among them, pancreatic cancer is the fourth fatal tumor among the adult population, and patients with pancreatic adenocarcinoma have a poor prognosis with cumulative 5 year survival <5%. With the vast increase in the number of medical images, how to locate and analyze the pancreas effectively and efficiently become a urgent subject for computer aided medical diagnosis. However, pancreas is a soft tissue located in internal abdomen, with the lack of a fixed shape, and having close and important relationship with other organs near by. Resulting that the edge of pancreas is difficult to be determined, so the analysis of abdomen image based on the technology of magnetic resonance imaging(MRI) has become an important method for diagnosis pancreatic cancer. In order to solve this problem, this paper pay attention to how to segment pancreas from the MRI sequence images effectively, and the main works of this paper are summarized as follows:(1) Aiming at the purpose that segmenting the pancreas from the abdominal MRI sequence images effectively while the object images have a complex background, and the object area, with the problem that edge is difficult to be found, and similar to the stomach in the gray level. By combining edge detection with region growing method, and the priori knowledge that the stomach has a stable shape as a constraint, this method we proposed can segment multiple target regions at the same time, namely the pancreas, stomach and liver are segmented along with, greatly improved the speed and accuracy for image processing.(2) Here we proposed a method to segment abdominal MRI sequence images based on dictionary transfer learning. By training the target region and background region separately, we get the target dictionary and background dictionary. Then use them to approximate the following images, and get the target area or the background area of that MRI image. At the same time, updating the dictionary to obtain a robust method while segmenting pancreas from medical sequence images.(3) A segmentation method based on statistical learning is proposed for extracting the target regions from the abdominal sequence images. Firstly, the parameters of the single Gaussian Model will be carried out while analyzing the target samples, and then establish the normal distribution model. As long as we own the model, we mark the subsequent data under test using the Bayesian classifier. By calculating the weighted density of the target pixels in all of the areas obtained when using the mean shift method to pre-segment the MRI image, and then we extract the target areas with a higher degree of precision segmentation.