Segmentation Of Celiac Trunk Based On Multi Channel Convolution Neural Network

Gastric cancer affecting all human beings,grievously impacting people’s health and daily life.The direction of celiac trunk plays an important part in the planning of gastric cancer surgery.Computed tomography（CT）is universally applied in the diagnosis of gastric cancer,so how to get accurate and precise segmentation results of celiac trunk is still a hot and challenging task.At present,the celiac trunk segmentation mainly relies on manual labeling,which not only labor-consuming but also is time-consuming.In computer vision algorithm,convolutional neural networks（CNNs）with its powerful performance are more and more used in various tasks in various fields.It can intelligently extract the information on the basis of vast amount of data,so as to achieve a certain prediction.Aiming at the problem of judging the direction of celiac artery in the process of gastric cancer diagnosis,based on the enhanced CT data of upper abdomen,this paper studies the semantic segmentation of celiac artery using multi-channel convolution neural network.The contributions are as below:（1）There is a certain information loss in the data type conversion of 3D CT data to 2D data,and the segmentation of 2D CT data with single axial plane has the problems of insufficient information and poor segmentation effect.Thus a multi-channel convolutional network algorithm is proposed for the process of gastric cancer diagnosis,which can fuse CT multi-axis data through the two-dimensional slice data of CT data.The relevant data of other axis are used to supplement the global information in order to make the calculation under the control,and the X-module part is used iteratively in the network,which can generate more global information to makes the network use more abundant feature information in the segmentation process,makes the network produces more accurate semantic segmentation results possible.Through the comparison of 60 data sets with some efficient semantic segmentation networks,it can find that the algorithm can be significantly improved in different ways,which further shows the effectiveness of the algorithm.（2）In the research of semantic segmentation of 3D CT data,the performance of sparse object is usually unsatisfactory in the edge part and small object part.In order to further utilize the original CT data with the information in a larger value range and further refine the segmentation of the edge part of the target,this paper proposed a novel method in a convolutional network based on multi supervision in the segmentation of celiac trunk within CT three-dimensional data.The network structure of this algorithm can be combined with the contour information of celiac trunk to assist the semantic segmentation of celiac trunk.To a certain extent,the semantic segmentation of celiac trunk is constrained and restricted.Through the comparative experiments of 60 data sets,it can be found that this algorithm can not only improve the overall level of network segmentation,but also make the final result of the network in the edge part more accurate,so as to realize the auxiliary diagnosis of the direction of gastric celiac trunk in the diagnosis of gastric cancer.