Research On CT Image Classification Of COVID-19 Pneumonia Based On Convolutional Neural Network

With the strengthening of domestic medical technology and the improvement of medical level,more and more artificial intelligence technology is applied to the diagnosis of diseases.In this context,the application of cutting-edge science and technology to the diagnosis and prediction of the new type of coronavirus pneumonia that broke out at the end of 2019 Among them is the general trend.Previously,many researchers have developed deep learning-based models and algorithms to process pneumonia images to assist hospitals in diagnosing whether patients are infected with new coronary pneumonia.However,these methods currently have certain limitations,and the robustness of the model and the accuracy of detection cannot meet the requirements in practical applications.Therefore,based on the method of deep learning convolutional neural network,this paper studies image enhancement and image classification algorithms for COVID-19 CT images.The main research contents are as follows:Firstly,the research background and significance of this topic are outlined,the current domestic and foreign research status of image recognition and prediction of new coronary pneumonia is analyzed,and the main problems existing in the current research on imaging diagnosis of new coronary pneumonia are analyzed,and some classical algorithms related to image classification are also introduced,formulating the main research goals and directions of this paper.Secondly,in view of the difficulty of obtaining CT image data of pneumonia in the early stage of the outbreak,a data enhancement algorithm is proposed to solve the problem of data sparseness.Specifically,the pix2 pix conditional generative adversarial network is used to expand the data samples to generate sample subsets;the accuracy of image classification is low.A diagnostic framework based on an improved capsule network is constructed to capture the spatial relationship between image instances and enhance the feature extraction capability of the network.The effectiveness of image enhancement and the superior performance of model classification are verified by experiments.Thirdly,due to the different conditions of the imaging equipment,there is heterogeneity among the acquired lung CT images,and the developed model algorithm cannot adapt to this heterogeneity among the data,so the problem of poor performance is proposed in different sites.On the joint data set,the focal loss function is redesigned and added to the dual-branch neural network,and Res Net is used as the basic structure of the upper-branch neural network;on the joint site data,the improved dual-branch neural network is used for training and testing,through comparative experiments The classification effect and robustness of the proposed method are shown.Finally,from the research perspective that CT images are three-dimensional image data,a weakly supervised high-precision classification network is proposed.First,the Unet segmentation network with attention mechanism is used to segment CT images to obtain lung masks.Membrane and raw CT volume images as input to a 3D deep convolutional neural network to predict COVID-19 infection probability.The experimental results confirm that the model has excellent prediction and classification performance on 3D CT images.