Combining Convolution Neural Network And Improved Level Set Model For The Technical Study Of Pancreas Segmentation

In recent years,the incidence of pancreatic diseases is increasing year by year,which is seriously endangering people's health.And the CT scan has become the primary examination method for the diagnosis of pancreatic diseases,which can detect pancreatic lesions at the early stage;thereby,it can improve the survival rate of patients.However,because of the complex anatomical structure of the pancreas,a large number of CT images,and the susceptibility to the noise,it is difficult for experts to segment the pancreas accurately manually.Therefore,it is a crucial issue to segment the pancreas quickly,accurately,and effectively.In this paper,we proposed an algorithm of pancreatic tissue segmentation based on the convolutional neural networks(CNNs)and the level set method.The algorithm mainly includes three stages:(1)In the preprocessing stage: it was proposed an enhancement method by using the fractional differential to improve the contrast between the pancreas and surrounding background.Then,we flipped,translated,and rotated the image to augment the data.(2)In the initial segmentation stage: CNNs were employed to extract the boundary of the pancreas as the initial one.(3)In the refine segmentation stage,the initial segmentation result was considered as the prior information,and the level set model was modified in terms of restriction of the prior information,length,area,and regularization.We achieved the final segmentation results by evolving the initial boundary.The proposed level-set model can not only effectively extract the weak edge,but also overcome the problem of under-segmentation,over-segmentation,and overflow of the traditional level set methods.Compared with other traditional level set methods,the experimental results show that the proposed method will obtain more accurate segmentation results and has clinical application prospects.