Research On Assisted Diagnosis Method Of CT Images Of Pulmonary Nodules Based On Deep Neural Network

The incidence and mortality of lung cancer ranks first in China’s cancer,and the number of new lung cancer patients in China also tops the world every year.The early detection,diagnosis and treatment of lung cancer can reduce the mortality of lung cancer.CT imaging is the most effective technique for lung cancer detection.The early symptoms of lung cancer are mostly manifested as lung nodules on medical images.Therefore,in order to improve the survival rate of lung cancer,lung nodule detection and diagnosis based on lung CT images are useful for lung cancer.Early diagnosis is of great significance.With the rapid growth of CT image data and the difficult task of lung nodule screening,radiologists can easily cause misdiagnosis and misdiagnosis.Therefore,it is necessary to use computer-assisted diagnostic technology to provide reference opinions for radiologists,assist doctors in diagnosis,and improve the diagnostic efficiency and accuracy of radiologists.At the same time,with the continuous expansion of the application of deep learning,due to the powerful feature learning and feature expression capabilities of the network structure,deep learning has also achieved ideal results in medical image analysis and processing tasks.Therefore,this paper proposes to use deep neural network-based methods to detect and diagnose lung nodules.The main work of this article includes:(1)the significance of computer-assisted diagnosis of pulmonary nodules was elaborated,the research status of lung nodule detection methods and lung nodule diagnosis methods were fully investigated,and algorithms based on traditional machine learning and deep learning were summarized and analyzed.(2)Lung nodule detection method using two-dimensional and three-dimensional convolutional neural networks.Aiming at the problems of low efficiency and large number of false positives in the detection of pulmonary nodules in the existing methods,this paper proposes an end-to-end two-dimensional fully convolutional neural network(2D U-net)and three-dimensional convolutional neural network(3D CNN)combined lung nodule detection method,in which a two-dimensional full convolutional neural network is used for object localization,and a three-dimensional convolutional neural network is used for stereo target classification.Experiments on the LUNA2016 and LIDC-IDRI datasets show that the proposed method is effective and effective,and the classification accuracy rate can reach 96.9%.(3)benign and malignant diagnosis of pulmonary nodules based on multi-branch feature convolutional neural network.Aiming at the problem that the current deep learning models for benign and malignant classification of lung nodules are unexplainable and low in practicality,this paper presents an interpretable multi-branch attribute convolutional neural network model(Interpretable Multi-branch Attributes Convolutional Neural Network).,IMACNN)is used for classification of benign and malignant lung nodules.Experiments show that the proposed model can not only obtain interpretable results of lung nodule classification,but also achieve better classification performance of lung nodules with an accuracy rate of 97.8%.(4)based on the lung nodule detection and lung nodule diagnosis algorithms proposed above,an intelligent assistant diagnosis system for lung nodules based on deep learning is designed and implemented.