Research On Pancreas Automatic Segmentation Algorithm Based On Deep Learning

Pancreatic cancer is a common malignancy,which is difficult for early diagnosis,challenging to cure,and accessible for relapse.It is known as the "king of cancer",which seriously threatens human life.Compared with other large organs such as heart,liver and kidney,pancreas is characterized by small proportion,high specificity and blurred boundary,which makes it extremely difficult to segment pancreas.At present,computed tomography(CT)has been widely used for the early diagnosis,analysis and monitoring of pancreatic diseases.However,hand-delineating the pancreas on abdominal CT images is time-consuming,laborious and requires a high level of expertise.Therefore,automatic and accurate segmentation of pancreatic organs in abdominal CT images has become a key task in the treatment of pancreatic cancer.With the advances in science and technology,and in particular the emergence of deep learning,computer aided segmentation provides significant opportunities.In recent years,the application of deep learning in medical image tasks has become increasingly popular.Many kinds of segmentation models based on artificial neural networks have been widely used in medical image segmentation,especially in the segmentation of large organs,and have achieved excellent results,promoting the development of automatic segmentation algorithms for medical images.However,for small organs such as pancreas,automatic segmentation is faced with challenges such as class imbalance,background information interference and non-rigidity,resulting in the existing segmentation accuracy is still difficult to meet clinical requirements.Since deep learning technology still has great potential in automatic pancreatic segmentation,this paper uses deep learning algorithm to realize automatic pancreatic segmentation.In order to further improve the accuracy of automatic segmentation of pancreas,integrated learning strategy is introduced in this paper,and lots of experiments prove that the proposed algorithm is effective.Specific research work is as follows:1.A pancreas segmentation algorithm based on dual input 3D convolutional neural network is proposed.Firstly,due to the small density difference between pancreatic organs and surrounding tissues,in order to highlight the pancreatic boundary,the image with contextual residual information was used as additional input to enhance the original input.Secondly,in order to avoid the degradation of deep network,a residual module is added to the basic network structure.Finally,in order to increase effective features and suppress noise features,the squeezing and stimulating attention mechanisms are embedded to improve the accuracy of pancreatic segmentation.For proving the effectiveness of the proposed algorithm,ablation experiments are conducted in this paper,and the experiments prove that each module plays an effective role.In addition,according to experimental verification,the input of context residual information as additional information is universal in each axis.2.In order to further improve the accuracy of automatic segmentation of pancreas,according to the proposed dual-input 3D convolutional neural network algorithm of pancreas segmentation,a pancreas segmentation algorithm based on ensemble learning was proposed.By introducing ensemble learning,context residuals from three different perspectives were fused as the segmentation results of the network model with additional input,so as to further improve the segmentation accuracy.Compared with the base classifier,the experimental results show that the integrated classifier complements the information of the base classifier,and the accuracy of the automatic segmentation algorithm of pancreas is further improved.Compared with the current mainstream pancreas segmentation algorithm,the experimental results show that the proposed algorithm has better performance than the current mainstream algorithm.3.Based on the proposed algorithm,an automatic pancreas segmentation tool based on abdominal CT images was designed to realize the visualization function of pancreas segmentation on abdominal CT images.The tool mainly includes reading and displaying medical images,pancreas organ segmentation of abdominal CT images,display and preservation of segmentation results.The display of segmentation results also includes two different options: overlay of prediction results to the original image and direct display of prediction results.In addition,the two-dimensional CT original image and segmentation image of required pixel points can be displayed by specifying coordinates.The tool is simple to operate,using this tool can achieve the purpose of reducing the workload of doctors and reducing the time of patients to see a doctor.