Research On Deep Learning Model Based On Feature Distribution And Its Application In Pancreatic Tumor Image Segmentation

Pancreatic cancer is one of the most common malignant tumors,which is harmful to human health and even life-threatening.Accurate segmentation of tumors and pancreas from medical images is important for clinical diagnoses and making treatment plans for patients with pancreatic cancer.However,manual labeling of pancreases and tumors is time-consuming,and the accuracy is easily affected by the working state of physicians.The phenotypic differences,low contrast,and small voxel ratio of pancreas and tumors make the automatic segmentation task be challenging.In recent years,convolutional neural networks have shown superior performance in medical image segmentation.Full convolutional networks can extract features from input images of any size and generate pixel-level segmentation results.In view of the feature distribution of pancreas and tumors in CT images,this paper proposes two automatic segmentation algorithms based on improved full convolutional networks to realize the accurate segmentation of pancreas and tumors and achieve effective analysis of medical big data.The main work is as follows:(1)This paper proposes a temperature-guided feature-extracted 3D fully convolutional network with three temperature guided modules,namely,balance temperature loss,rigid temperature optimizer and soft temperature indictor,to realize joint segmentation of the pancreas and tumors.Specifically,balance temperature loss is designed to dynamically adjust the learning points between tumors and the pancreas to balance the selected features,and it is aimed at improving the accuracy of tumor segmentation without losing pancreas information.Rigid temperature optimizer is proposed to accept nonimproving moves probabilistically to adaptively avoid local optima.To further refine the segmentation results,we propose the soft temperature indictor to guide the network into a fine-tuning state automatically when the model tends to stability.The experimental results are more accurate than the fourteen top-ranking methods in pancreas and tumors segmentation on the MSD pancreas dataset and six top-ranking methods in brain tumors segmentation.Ablation studies verify the effectiveness of the three temperature guided modules.(2)To further enhance the automatic recognition of targets,this paper proposes a feature combination guided fully convolutional network(FCnet)for the segmentation of pancreas and tumors in abdominal CT images.For complex anatomical structure and fuzzy boundary,dynamic combinatorial learning strategy is proposed to adaptively specify the learning direction of the network by paying more attention to morphological and boundary features of pancreas and tumors.To further focus on the relationship between the learned features and the inputs in the network,dual information gate is designed to continuously discard redundant information(such as information unrelated to pancreas and tumors region)during network training while highlight effective information.In addition,we also integrate the self-attention mechanism with the convolutional neural network model,and propose Transformer fusion branch to obtain global semantic information between high-resolution image blocks and deep voxels.Experimental results on MSD pancreas dataset and LiTS dataset show the superiority of the proposed model in pancreatic tumor segmentation and its robustness,respectively.Ablation experiments verify the effectiveness of the three designed modules.