Design And Implementation Of Pulmonary Nodule Detection And Segmentation System Based On Deep Learning

At present,the incidence and mortality of lung cancer are the highest among cancer.Early diagnosis and treatment of lung cancer are extremely important to improve the survival rate and prognosis of patients.Pulmonary nodules are the early manifestation of lung cancer,and the clinician can conduct the early screening of lung cancer by observing the characteristics of pulmonary nodules in the chest CT images.However,the huge CT image data and the characteristics of pulmonary nodules have brought great challenges to the screening work.Deep learning technology can realize automatic detection and segmentation of pulmonary nodule by learning a large number of pulmonary nodule data,providing reference for doctors’ clinical diagnosis.Therefore,based on the research of pulmonary nodule detection and segmentation algorithm,a pulmonary nodule detection and segmentation system based on deep learning was designed and implemented,aiming to improve the work efficiency of doctors and avoid the influence of subjective factors on the diagnosis results.Firstly,the domestic and foreign research status of pulmonary nodule detection and segmentation algorithm was introduced,which mainly include the traditional method based on manual feature extraction and the deep learning method based on automatic feature extraction.By comparing the advantages and disadvantages of the both methods,the deep learning method was adopted.Secondly,the pulmonary nodule detection algorithm was studied.Early screening of pulmonary nodules is an important means to effectively prevent lung cancer.Since pulmonary nodules are small in size and difficult to detect at the early stage of growth,this paper proposed a multi-scale VNet network to detect pulmonary nodules based on the three-dimensional characteristics of pulmonary nodules The performance of the network was evaluated by using three-dimensional pulmonary nodule images from LIDC-IDRI lung nodule data set and clinical data,and the results showed that multi-scale VNet can achieve better performance in pulmonary nodule detection,which basically met the requirements of clinical pulmonary nodule detection,and can be used in the system for pulmonary nodule detection.Then,the pulmonary nodule segmentation algorithm was studied.A Dense-UNet network was proposed for contour segmentation of the pulmonary nodules which have been detected.The network performance was evaluated by using two-dimensional pulmonary nodule images from LIDC-IDRI lung nodule data set and clinical data,and the results showed that Dense-UNet can achieve better performance in pulmonary nodule segmentation,which basically met the requirements of clinical pulmonary nodule segmentation,and can be used in the system for pulmonary nodule segmentation.Finally,by analyzing the needs of chest CT image detection and diagnosis,this paper designed the overall framework of pulmonary nodules detection and segmentation system,and developed the functional modules of the system,such as basic functional module,pulmonary nodule detection and segmentation algorithm module,and diagnostic information management module.The system can realize user login,picture browsing,data preprocessing,pulmonary nodule detection and segmentation,diagnosis information management and other functions.After the system development was completed,the software functional testing and performance testing were carried out.The test results showed that the functions and performance of the system basically met the design requirements and had certain application value for the clinical detection and segmentation of pulmonary nodules.