Application Research On CT Image Segmentation Via Convolutional Neural Networks

Nowadays,new tools for disease diagnosis are emerging in the medical field,and more and more research is based on medical digital images.Digital imaging equipment innovation and medical image processing methods in the context of continuous innovation,medical digital images as a reflection of human organs and tissues of the internal information carrier plays an irreplaceable role.In the face of today's high-resolution digital images,how to use these image data to diagnose the disease and depth of the relevant information on the disease is now medical image processing and analysis has been studied within the scope of the difficulties and priorities.Medical digital image segmentation technology research is an important part of the current image processing field,but also for other aspects of image processing has laid a solid foundation.Many image segmentation methods have been matured for a long time,used in the field of medical digital images and achieved very good segmentation effects,but also encountered a lot of restrictions.The emergence of the depth learning network has injected new vitality into the development of image segmentation technology.The segmentation algorithm based on depth learning has become a new research direction for experts and scholars.In this thesis,the application of the convolution neural network model in deep learning to segment CT images and brain CT image data is studied.At the beginning of the article,the importance of medical CT image diagnosis and the traditional image segmentation method are analyzed,and the necessity and feasibility of CT image segmentation based on convolution neural network are emphasized.Secondly,the LeNet5 convolution neural network model is improved based on the membership probability method and the CT images of the lungs are segmented on this basis,and a good segmentation effect is achieved.Finally,on the basis of the CT image data of the brain,the neural network structure is designed and segmented,and the effect of improving the network on the CT image segmentation of the brain is confirmed.The specific research contents are as follows:1.In the CT segmentation of lungs,the pulmonary parenchyma is extracted based on the histogram threshold binarization method and the boundary tracking method.Then,on the basis of the image patch and the fuzzy membership method,the traditional convolution nerve the output layer of the network structure is improved and the parameters are trained.Finally,the lung texture segmentation is completed by the trained convolution neural network model.2.In the brain CT segmentation,the soft tissue part of the brain CT image is extracted by using the threshold and morphological algorithm.Then,by setting the number of layers of the convolution neural network,the size and number of the convolution layer kernel function,The new convolution neural network structure is used to segment and train the parameters of the brain CT image.Finally,the soft tissue segmentation of the brain is completed by the trained convolution neural network model.3.In the experimental part,we study and use the depth learning framework Caffe to complete the experimental part of this algorithm,and realize the visualization of the middle process of the convolution neural network.