Application Of Deep Learning In Thorax CT Image Segmentation

With the rapid development of medical imaging and computer technology,using computer technology to analyze the clinical image data increases the probability of disease prevention and successful treatment.In the diagnosis and detection of thoracic diseases,computed tomography(CT)is most commonly used.Because CT can provide high-resolution scanning images for each organ or tissue of the chest,making full use of CT scan images very important for the detection of lung disease,such as lung cancer,pulmonary nodules and other diseases.In the computer-aided diagnosis system,the accurate segmentation of the lung CT scan image is the basis and prerequisite for the subsequent thorax function analysis and reconstruction of the three-dimensional image.Accurate segmentation not only increases the accuracy of disease diagnosis,but also reduces the amount of subsequent irrelevant calculations.The accurate segmentation of thorax CT image has two requirements,one is using thorax CT image segmentation to improve the development of computer-aided diagnosis,and the other is to achieve a complete segmentation of the thorax organs.When the thoracic organ has a lesion or deformity,there will be some abnormalities in the CT image,then the segmentation algorithm need to split the organs completely.Medical image segmentation has many common methods,such as region-based method,edge-based method,and method based on special theory.Because of the characteristics of medical images such as gray un-uniform,individual differences,artifacts and noise,the above image segmentation methods have some limitations,and it is difficult to achieve the required sensitivity and accuracy when performing organ segmentation of the various organs of thorax.This paper aims at some difficulties in the segmentation of thorax CT image.Combined with the recent advanced deep-learning techniques,the feasibility of image segmentation using convolution neural network model is discussed.The main work and innovation of this paper are as follows: A convolution neural network model for feature extraction and pixel classification using image neighborhood is proposed,which is used to segment the thorax CT images precisely;A nonlinear deep learning model that learns directly from raw image to be segmented to mark image,which is end-to-end and simplifies the process of image segmentation.We have verified these two deep learning models using clinical scanning chest CT images,the experimental results show that the deep learning models proposed in this paper achieve good performance in image segmentation.