Full Convolutional Neural Network Algorithm And Its Application In Image Segmentation Of Lung Tumors

The development of the computer application software makes computer-aided based on medical diagnosis technology to achieve rapid development.By methods of digital imaging,medical image processing technology,combined with the analysis and calculation of high performance computer,doctors found normal lesions with suspected lesions separated from the background of normal tissue dissection.Computer can get rid of the doctor's judgment,to avoid individual doctors.So as to improve the accuracy of diagnosis,and improve the work efficiency.In the process of lung cancer diagnosis,tumor segmentation is an indispensable part.Lung segmentation refers to the steps of segmentation on lung lesions.We researched on nearly five years of domestic and foreign research,and then proposed the solution of the problem for lung tumor CT image segmentation.In this thesis,We work out image segmentation problem for lung cancer research includes the following aspects of content:1.We do summarizes of the current medical imaging technology,artificial intelligence,and significance,research status at home and abroad.2.We do introduction common techniques of image segmentation,image segmentation in medical image processing technology and the convolution neural network algorithm is related to the image segmentation technology and application.3.We do research on the convolution neural network,as well as finding the advantages and disadvantages of the convolution neural network.We put forward the U-net in tumor image segmentation.4.U-net based on full convolution neural network algorithm,it needs large amount of data,so we put forward the W-net.This thesis set up the convolution in the experimental process neural network model,optimizing the network structure and put forward W-net convolution neural network structure,research under the experiment of large data samples,to reduce the algorithm complexity,improve the convergence performance and the segmentation accuracy.This thesis W-net neural network model can accurate segmentation of lung tumor image,makes the clinical doctor's diagnosis more accurate.