Thoracic CT Image Segmentation Based On Convolutional Neural Network

In recent years,the incidence of lung malignant tumors has remained high,and the mortality rate is also the highest in the world.With the development of medical imaging technology and image processing technology,medical image segmentation technology is more and more widely used in the diagnosis and treatment of lung tumors.Computer-based automatic image processing has become one of the main directions in the field of modern medical imaging.Most medical images have the characteristics of soft tissue structure,complex micro-structure,blurred boundary of imaging target and large amount of data,which greatly affects the effect of diagnosis and treatment of doctors.Therefore,the development of excellent computer-aided diagnosis algorithm is very important.Accurate image segmentation can not only improve the efficiency of doctors,but also improve the accuracy of diagnosis.Aiming at the characteristics of chest CT images,combined with advanced deep learning technology,a new network model is proposed based on the classical U-net network structure.The main work and innovations of this paper are as follows:(1)A lung cancer segmentation method combining threshold method and convolution neural network is proposed.OTSU threshold method is used to segment lung parenchyma,and then the segmented lung parenchyma image is input into convolution neural network for lung cancer segmentation.(2)A network model for segmentation of lung parenchymal areas in chest CT images is proposed.It is an end-to-end full-convolution neural network model based on local residual network.It can be directly trained by using standard segmentation maps.(3)An automatic cascade neural network model for lung cancer segmentation is proposed.Two network models are trained.The first one is to segment the lung parenchyma.The second one is to cut the lung parenchyma image and get the segmentation result.According to the segmentation method mentioned in this paper,chest CT images are used to verify the experimental results.The results show that the proposed model has excellent segmentation performance.