The Key Technology Research Of Automatic Detection For Pulmonary Nodules

Lung cancer is one of the leading causes of cancer death in the world today.Pulmonary nodule detection based on low-dose CT has great significance for early treating lung cancer and increasing patient survival.This paper takes the key technology of automatic detection for pulmonary nodules based on deep convolutional neural network as the research object,and aims to develop an accurate and efficient computeraided detection system for pulmonary nodules.The main research contents and achievements are as follows:Firstly,a 3D-RPN candidate pulmonary nodules detection network was proposed,which can perform end-to-end training for the generation task of suspicious nodules.The network uses the improved U-Net as the skeleton.RPN utilizes the convolutional feature maps to simultaneously predicts object bounds and object scores.The 3D SERes Net module allows the deep network easier to be optimized and dynamically recalibrates channel-wise features of residual learning,which is beneficial to boost nodule feature discriminability.Secondly,a false positive reduction network of pulmonary nodules based on 3D SE-Res Net was proposed,which reduces false positive nodules during candidate nodule detection stage.3D SE-Res Net module improves the classification ability of the network.The 3D SE-Res Net module integrates the advantages of Res Net and SENet.The residual learning for feature reuse solves the problem of deep network optimization,and the squeeze-and-excitation operation for adaptive feature recalibration enhances the learning ability of original basic modules.Finally,based on the CT image and 3D deep convolutional neural network framework,a new method for pulmonary nodules detection is developed and an intelligent pulmonary nodules detection system is developed.The system mainly includes the function modules of patient data information management,image preprocessing,candidate pulmonary nodules detection and false positive reduction.The system can realize the rapid detection and localization of pulmonary nodules,which can assist doctors to improve the detection efficiency.Experimental results on LUNA16 demonstrate superior effectiveness of the proposed approach in pulmonary nodule detection task.