Font Size: a A A

Automatic Pancreas CT Scans Segmentation Based On Convolutional Neural Network

Posted on:2022-02-08Degree:MasterType:Thesis
Country:ChinaCandidate:Y J YanFull Text:PDF
GTID:2544306326476764Subject:Computer Science and Technology
Abstract/Summary:PDF Full Text Request
The pancreas plays a significant role in glucose digestion and metabolism respectively.However,the pancreas is vulnerable to diabetes,pancreatitis and pancreatic cancer.In particular,pancreatic cancer,as a common malignant tumor,its incidence rate increases year by year in the world.The early diagnosis rate of pancreatic cancer is very low,which is the main reason for the high mortality of pancreatic cancer.As a new technology of diagnosis and prognosis of pancreatic cancer,automatic pancreas segmentation is of great significance to assist doctors in diagnosis,treatment and operation.Automatic organ segmentation from abdominal images is an important research topic in medical image processing.It includes locating the region of interest that contains the target organ from computed tomography(CT)or magnetic resonance imaging(MRI)to help accurately segment the target organ and improve the diagnosis and treatment effect.However,the size,shape and location of the pancreas are greatly influenced by the gender,age and obesity of patients.Therefore,compared with the liver,kidney and other organs,accurate segmentation of the pancreas is an extremely challenging task.In this paper,we propose two 2.5D pancreas segmentation models.One is an improved 2.5D U-net network with a hybrid attention module.Attention mechanism can improve the segmentation accuracy without extra supervision,and can suppress the interference of irrelevant information from background region.The other is called 2.5D U-net+which is composed of several U-net networks of multiple depths.2.5D U-net+can autonomously learn to find the best network depth for a given dataset.2.5D networks can capture more spatial information along the third dimension than 2D networks,and require less computational resources and GPU(Graphics Processing Unit)memory than 3D networks.In order to improve the segmentation accuracy,we apply a coarse-to-fine framework for segmentation,which uses a smaller image obtained from the coarse segmentation as the input of the fine segmentation.In this paper,the proposed approaches are evaluated on the NIH Pancreas dataset.Experimental results demonstrate that the proposed approaches outperform state-of-theart methods in recent years.
Keywords/Search Tags:Pancreas Segmentation, 2.5D U-net, Attention Mechanism, Coarse-to-Fine Segmentation Framework
PDF Full Text Request
Related items