Research And Implementation Of Pulmonary Nodule Detection In CT Image System Based On Deep Learning

Statistics show that in China,lung cancer incidence rate and mortality rate are the highest among all kinds of cancer patients.However,once lung cancer is found in the early stage and treated actively,its five-year survival rate is very high.The later it is found,the lower the fiveyear survival rate is.The clinical symptoms of early lung cancer are pulmonary nodules with a diameter of about 3-20 mm,which can be examined by CT imaging.Therefore,the examination of pulmonary nodules in chest CT images has important clinical significance,which can improve the national health level.In the actual clinical examination,doctors need to review a large number of clinical images,which depends on their experience level and subjectivity.Therefore,it is easy to cause fatigue,which will lead to missed diagnosis and misdiagnosis.With the development of computer technology,researchers put forward the computer aided diagnosis(CAD),using computer to realize different algorithms to realize the detection of pulmonary nodules,in order to reduce the burden of doctors,reduce the subjective errors of doctors,and improve the accuracy of diagnosis.In the current detection algorithm and system of pulmonary nodules,the deep learning method can achieve good detection results,but there are still some problems such as manual setting of parameters and further improvement of detection accuracy.At the same time,most of the pulmonary nodule detection system can complete the detection of pulmonary nodule,but because the systems often integrates many other functions,the system are too cumbersome to use,and the costs are high.To overcoming the problems of nodule detection methods and detection system,this paper focus on them.According to the requirements of clinical application,the specific process of pulmonary nodule detection is analyzed.Based on this,the demand analysis and function design of the nodule detection system are carried out.Then the algorithm of pulmonary nodule detection based on deep learning is studied.Finally,the implementation and system test of the nodule detection system are carried out.For the detection of pulmonary nodules,we use U-NET to train a lung parenchyma segmentation model and extract the lung parenchyma as the preprocessing stage.Based on the U-net network framework,combined with the encoding and decoding structure,the multi-scale feature information is fused to build the detection framework of pulmonary nodules,i.e.an improved N-Net is used to detect the nodules.For the implementation of the system,from the functional requirements,the overall framework and functional modules of the system are designed.QT is used to build the interactive interface of the system,including the main display interface,menu bar,operation key options and image display window.VTK tool package is used to read and display medical images in DICOM and MHD formats,and VTK functions are used to realize image interaction function,so as to complete various functions of the system.After repeated testing on Luna public database,the designed system based on deep learning has less requirement for hardware environment,and can automatically and quickly detect pulmonary nodules,with low missed diagnosis rate and misdiagnosis rate...