Research On Pancreatic CT Image Segmentation Based On Convolutional Neural Network With Attention Mechanism

The world is suffering from the New Coronavirus Pandemic,a large number of patients need to use the CT machine to scan their diseased organs.Analyzing these mounts of CT images cost a lot of time and energy of professional physicians.Using computer to automate organ segmentation of CT images becomes urgent situation.By using a computer for CT image segmentation,it is possible to accurately and accurately locate the lesions of a large number of patients in a short time and to assist doctors in disease diagnosis.Therefore,how to effectively apply the image segmentation technology to medical imaging has become a task urgently solved by a large number of researchers.As we all know,in the process of CT image shooting,you will encounter various problems,such as large changes in target size,shape,and position in different environments.The target image may cause interference from external noise in the actual shooting.The limitations of these factors make a single traditional algorithm usually unable to deal with different complex situations.As the deep learning technology gradually matures,it provides new ideas for CT image segmentation.In this thesis,a convolutional neural network based on an attentional module is used to study the automatic precision of pancreatic CT images using the pancreatic Segmentation Methods.The objective of the study is to establish a convolutional neural network segmentation model to automatically identify and segment a complete pancreatic organ from CT images.Starting from a multi-scale convolutional neural network-based segmentation model of pancreatic CT images,the paper writes a feature map coding module and a feature The graph decoding module model enables automatic segmentation of the pancreatic organ.By comparing with common deep convolutional neural network learning algorithms,the proposed method in the thesis,both in terms of accuracy and number of model parameters are well improved.Then,to address the problem that multi-scale convolutional neural networks produce fine fragments during segmentation,we propose an attention-based mechanism for the Convolutional neural network approach,which fine-tunes the extraction of semantic information using attentional modules,and spatial and semantic information The method is guided by the fact that by increasing the number of parameters to a smaller number,the method achieves greater accuracy and not only effectively divides the intact pancreas,but also the entire pancreas.organs and alsosignificantly improved the accuracy of pancreatic organ segmentation.The main points of the paper are as follows.(1)Investigate a CT image segmentation method based on multi-scale convolutional neural network,which consists of two modules.The feature map encoding module connects the convolutional modules in parallel,and the method can simultaneously extract the spatial and semantic information of feature maps.The feature map decoding module finely reduces the feature map to its original size by using a convolutional layer with bilinear difference.The experimental comparative analysis on the pancreatic dataset shows that the model can effectively segment the pancreatic organ,but there are computational problems in the prediction.Low efficiency,poor spatial continuity and fine fragmentation of the segmentation results.(2)Investigate a method of CT image segmentation based on an attentional module of convolutional neural network,which is based on the design of a semantic Information Enhanced Attention Module and Feature Fusion Pooling Attention Module.To address problems such as fine fragmentation in CT image segmentation methods for multi-scale convolutional neural networks,focusing on semantic information refinement and The connection between the interclass and intraclass problems is well understood by using the average pooling pathway and the maximum pooling pathway for both problems The resolution of the problem is demonstrated by a comparative experimental analysis on the pancreatic dataset to show that the model is able to effectively segment the pancreatic organ and achieve the current The mainstream model level.