Visualization Study Of Pulmonary Nodules Based On 3D U-Net Bottom Layer Features And Edge Detection

Lung cancer is one of the cancers with the highest incidence rate in the world.In the early screening of lung cancer,visualization of lung nodules is helpful for accurate diagnosis of lung cancer.In recent years,deep learning technology has been widely used in various tasks such as pulmonary nodule detection,segmentation,and diagnosis,and the diagnosis of pulmonary nodules is inseparable from the accurate extraction and visualization of pulmonary nodule edge features and morphological features.In order to solve the problem of insufficient extraction of edge and morphological features in the visualization of pulmonary nodules,a study on visualization of pulmonary nodules based on deep learning and edge detection was carried out in this thesis,and it is verified on the public LIDC-IDRI dataset.Lung CT images are an important basis for screening lung-related diseases.The complete structure of pulmonary nodules can be extracted from lung CT images by using image segmentation technology based on deep learning technology.By analyzing the features extracted by each convolution layer in the lung nodule segmentation model based on deep learning technology,it is found that the first two layers of convolution operations of the deep learning model usually extract some low-level features of the image,such as the edge,shape and texture features of the image etc.Because these low-level features are helpful to extract the shape information of pulmonary nodules,and the classic edge detection algorithm can effectively extract the edge information of pulmonary nodules.In order to give full play to the advantages of deep learning and classic edge detection operators in the shape and edge feature extraction of pulmonary nodules,a visualization method for pulmonary nodules that combines the underlying features of deep learning models and traditional edge detection algorithms was proposed in this thesis.Since the 3D U-Net model has a skip connection information transmission method,which can ensure accurate segmentation of object edges,it has been widely used in 3D pulmonary nodule segmentation tasks.In this thesis,the underlying feature information of 3D U-Net and traditional edge detection operators are used to carry out visual analysis of pulmonary nodules.First,the 3D U-Net deep learning network is used to segment the pulmonary nodules on the LIDC-IDRI dataset,and the calculation results of the first two layers of convolution operations of the 3D U-Net are obtained.Secondly,the result is binarized,and the non-pulmonary nodule area in the binarized image is eliminated using the erosion-dilation algorithm.Then,multiply the result of pulmonary nodule binarization area processing with the corresponding elements of the original image of pulmonary nodules,and set the pixels of non-pulmonary nodules in the CT image to 0,so as to obtain an image O containing only the main body of pulmonary nodules.Thirdly,use the edge detection operator to process the original image of the pulmonary nodule to obtain the result image E containing the edge information of the pulmonary nodule.Finally,image O and image E are fused using a fusion algorithm to finally obtain a visualized image of pulmonary nodules.Since the visualization analysis of pulmonary nodules has not formed a quantitative unified evaluation index,in order to evaluate the effectiveness of the lung nodule visualization method proposed in this thesis,a scoring questionnaire(scoring interval is[0,10])was designed.And 26 graduate students and doctors in the field of lung CT image processing were invited to fill out the questionnaire.The results show that the visualization method of pulmonary nodules that combines the underlying features of the deep learning model and the traditional edge detection algorithm can obtain an average score of 8.74,which proves the effectiveness of the visualization method of pulmonary nodules proposed in this study.In addition,in order to display the results of pulmonary nodule visualization more intuitively,a simplified version of pulmonary nodule visualization software was designed using PyQt5 development technology,which will help improve the efficiency of pulmonary nodule auxiliary diagnosis.