Research And Application Of Tumor Image Segmentation Algorithm Based On Deep Learning

In recent years,the technology of deep learning has developed rapidly and has become brilliant in many fields,especially in the field of computer image processing.Deep learning,especially the related technology of neural network,has been gradually applied in medical image analysis.Deep learning correlation methods have great application value in medical image processing.In the field of medical image analysis,medical image segmentation can be used in image-guided interventional diagnosis and treatment,directional radiotherapy and other processes,and become an important means of computer-aided diagnosis.Accurate segmentation of tumor is the core step of computer-aided diagnosis.Accurate results can provide scientific and reliable reference for doctors and improve the efficiency of diagnosis and treatment.At present,manual segmentation is widely used in clinical diagnosis and treatment.However,it is obvious that this method has low efficiency and poor reliability.Therefore,an accurate and fast segmentation tumor auxiliary segmentation system is urgently needed to assist diagnosis.In this thesis,convolutional neural network is used to segment rectal CT images for tumor,mainly studying how to better play the role of neural network,and achieve the purpose of efficient and accurate segmentation of rectal cancer tumors.The mainstream deep learning method,using convolutional neural network to segment rectal CT image,has two problems: first,the image data is small and difficult to train; Second,the target area to be segmented is too small and the background interference is too large,so it is difficult to achieve accurate segmentation.The segmentation effect of the mainstream network model Unet is not ideal.In this thesis,based on the Unet network model,a segmentation network suitable for rectal CT image is designed,and based on this,an auxiliary diagnosis system for rectal cancer is developed.This thesis mainly improves the segmentation efficiency through two aspects.On the first hand,in order to make full use of the detailed feature information of images,Transformer architecture is applied to improve the feature extraction ability of the network because of the lack of feature extraction ability of small target images in the current mainstream network model.In the second aspect,attention mechanism is adopted to further improve the existing model by fusing the global semantic information and local semantic information of the graph by fusing the multi-scale feature image information.Through experimental testing on the data set,the accuracy is improved on the basis of Unet and FCN modelsIn this thesis,an auxiliary diagnosis system for rectal cancer with anterior and posterior separation is designed and implemented,and the optimized network model is deployed in the system.Through the front-end upload patient CT image,display segmentation results,and can calculate the tumor circumference,area and other information and save historical data,convenient comparison of tumor changes,realize the application of the algorithm.