Research Of Pulmonary Nodule Detection Based On Deep Learning

In recent years,more and more people have found pulmonary nodules in physical examinations.The scientific research shows that the probability of carcinogenesis of lung nodules is about 40%.In China,about 75% of lung cancer patients have advanced lung cancer when they first go to hospital.Therefore,early detection and diagnosis of lung cancer is an extremely urgent matter for lung cancer patients.The results of lung nodules have a significant impact on subsequent diagnosis.To help experts easily diagnose lung nodules,artificial intelligence can generate accurate predictive models through massive medical data,medical imaging,study and research of medical record,which can provide early warning for diseases.After analyzing the advantages and disadvantages of traditional detection methods and deep learning methods,this thesis designs a novel end-to-end lung nodule detection scheme.This scheme is different from most previous methods and gives the detection results directly by skipping the candidate nodule selection phase,which greatly simplifies the detection process.The main realization point is to introduce the shortcut technology into the U-Net's down-sampling to overcome the problem of insufficient depth,and named Residual U-Net(RUN).This method is validated in the LUng Nodule Analysis 2016(LUNA16)data set,and achieves a sensitivity of 90.90% at 2 false positives per scan which outperforms many state-of-the-art approaches.Then,based on the RUN algorithm proposed in this thesis,the prototype system of lung nodule detection is designed and implemented.The combination of computer-aided diagnosis and image processing is fully utilized,and the high-performance computing power of the computer is used to help clinicians diagnose lung cancer in time.The running results of the test system also show that the system works well,and can quickly realize image preprocessing,intelligent prediction of pulmonary nodules,diagnosis information input,and patient case reviewing.The implementation of the system makes it easier for doctors to perform a full range of treatment and screening of patient conditions,thus,more precise diagnosis results can be obtained.