Design Of Software For Detection And False-Positive Reduction Of Pulmonary Nodules Based On Deep Learning

Computer-aided diagnosis plays an increasingly important role in smart imaging medicine,and the detection of pulmonary nodules is one of the most typical applications.At present,the incidence of pulmonary cancer in the world and CT image data is rapidly increasing.The work pressure of screening the nodules in the CT images of the pulmonary with the naked eyes and experience of the doctor is growing,and the detection of the pulmonary nodules may be caused by the fatigue or lack of experience of the doctor.With the help of computer-aided detection of pulmonary nodules and processing of lesion images,it is of great value to improve the efficiency of pulmonary cancer diagnosis and reduce the cost and time of pulmonary CT image annotation.The current commercial software systems for pulmonary nodule detection have practical problems such as single function and high price.There are many inconveniences in the actual detection of pulmonary nodules and the annotation of scientific research data.Therefore,the development of a software platform with high pulmonary nodule detection performance and image data annotation function has become the research focus of this thesis.Based on the design platform of the PyQt5 software interface,starting from the actual needs,on the basis of the realization of the pulmonary nodule detection network and the false-positive reduction network,a software system for the detection and false-positive reduction of pulmonary nodules is designed based on deep learning in the thesis.The pulmonary nodule detection network used in this thesis is based on the DeepLung system detection network,and the false-positive reduction network is a three-dimensional convolutional network implemented on the basis of the residual network.In the design of the software system,first select the CT image data to be detected,and then obtain the candidate results of the detection through the pulmonary nodule detection network,then input the results into the false-positive network for false-positive detection processing.The detection results are marked on the CT image.The visualization functions in the software system include the display of pulmonary CT images,detection results and false-positive reduction results and their annotation in CT images,and users can crop and annotate images in the visualization window Through practical application,it is found that the pulmonary nodule detection network and the false-positive reduction network need to exist at the same time.The main reason is that in the pulmonary nodule detection stage,in order not to miss the pulmonary nodules,some false-positive nodules are allowed.Therefore,it is necessary to design a false-positive reduction network as a cascade network to reduce the false positive rate and further improve the accuracy of pulmonary nodule detection.In this thesis,the LUNA2016 pulmonary image data set and the high-resolution CT data set built by the team are used to train and test the false-positive reduction network.The experimental results show that the sensitivity of the output of the false-positive reduction network is 97.83%.The pulmonary nodule detection network in the software system only uses the original training detection network parameters of the DeepLung system,and the new data set is not used for retraining.The software system for pulmonary nodule detection and false-positive reduction designed in this thesis has the functions of pulmonary nodule detection,false-positive reduction,visualization,and result labeling.It has certain application value to the computer-aided diagnosis and auxiliary processing of pulmonary CT images.