基于CT图像的肺结节自动检测模型研究与实现

AbstractLung cancer is one of the fastest growing malignant tumors with incidence rate and mortality rateTimely detection of lung cancer and treatment are of great importance to human health protection.Mostof the traditional pulmonary nodule detection methods are based on the combination of professionalmedical knowledge and the characteristics of CT images to extract features,and then select usefulfeatures to train in the classifier,which is not only cumbersome,but also poor generalization ability ofthe trained model,and can not play a very good auxiliary role.In recent years,deep leaming technologyis becoming more and more mature,which plays an important role in the detection of pulmonarynodules.In this paper,a Faster RCNN algorithm based on deep learning is proposed to detectpulmonary nodules,which solves the problem that traditional methods can not automatically detectpulmonary nodules.The main contents of this paper are as follows:(1)using LabelImg tool to label lung cancerdatasets in the format of LIDC-IDRI and DICOM in Affiliated Hospital of Ningxia Medical Universityrespectively;(2)using Faster RCNN algorithm to detect lung nodules,aiming at the problem that FasterRCNN cannot detect automatically,this paper proposes an improved Faster RCNN algorithm,combining Faster RCNN with 3D spatial features,realizes automatic detection and improves theaccuracy of pulmonary nodule detection;(3)based on Faster RCNN algorithm,a pulmonary noduleauxiliary detection system is designed by python.The improved Faster RCNN algorithm is based on Faster RCNN by adding dual path and u-netnetwork structure,so that the improved algorithm proposed in this paper can directly process medicaldata in DICOM format without conversion to standard JPG image,and can capture rich 3D imagefeatures in DICOM format.Experiments show that the recognition accuracy of the improved fast RCNNalgorithm is up to 96.6%,which is more accurate than the original Faster RCNN 1.70%.The improvedFaster RCNN not only simplifies the detection of complex pulmonary nodules,but also improves thedetection rate.This study provides a good application value for the accurate detection of lung cancerand reducing the burden of doctors.Key words:pulmonary nodules,CT images,Faster-RCNN,Deep learning川