Research On Pulmonary Nodule Segmentation Based On Convolution Neural Network

In recent years,the mortality rate of lung cancer has increased year by year,and it is particularly important for the early prevention of lung diseases.Pulmonary nodules are one of the main signs of early lung cancer.The detection of pulmonary nodules becomes the best way to prevent lung cancer.As an important medium for computer-aided treatment,medical images carry abundant information,and their role in medical research and diagnostics is increasingly prominent.Non-invasive method is used for CT to obtain internal tissue images,which is non-visible light reconstruction imaging with better perspective ability,and because of convenient check-up,clear structure and other characteristics,CT has become an important tool for doctors to diagnose the disease and evaluate the treatment effects.In order to effectively improve the diagnosis and treatment of lung cancer,lung nodules can be accurately separated from CT images of the lungs to greatly assist lung cancer assessment and diagnosis,it is of great research value that how to efficiently extract image features,accurately identify lesions,and reduce misdiagnosis rate.At present,segmentation of pulmonary nodules for CT images has become a research hotspot.However,because of the intensity inhomogeneity and blurred boundary of CT images,the traditional image segmentation algorithm can segment the nodule boundary,but the over segmentation is severe and the segmentation accuracy is not ideal.The convolutional neural network can extract image features automatically.The shallower convolutional layers extract the features of local regions,and the deeper convolutional layers extract some abstract features,which are insensitive to the size,location,and direction of the object,this helps to improve the performance of the segmentation algorithm.However,the acquisition of medical images with labels is difficult,which makes it difficult to use the convolutional neural network to directly train medical images.In order to solve the above problems,a lung nodule segmentation algorithm based on fully convolutional networks is proposed in this paper.We adapt VGGNet into fully convolutional networks and transfer their learned representations by fine-tuning to the segmentation task.It solves the problem of less data effectively.Specifically,we first pre-train the network model on the natural image dataset and obtain the weight information.Then,we make labels for medical images and perform data augmentation.Finally,the obtained weights are used as the initial value of the fine-tuning network for training and testing.The experimental results show that the features extracted by automatic learning of the fully convolutional networks based on transfer learning are more conducive to the segmentation of pulmonary nodules,and the segmentation results are superior to the traditional level set and Graph Cut segmentation methods and have high accuracy.