Research On X-ray Image Classification Algorithm Of Pneumonia Based On Deep Learning

The rapid spread of the Corona Virus Disease 2019(COVID-19)has led to a global health crisis,posing a huge risk to all aspects of health care,economy and security.The highly infectious,mutable and insidious nature of the novel coronavirus has made prevention and control of the outbreak much more difficult.Rapid detection and isolation and treatment of potentially infected patients is an effective way to curb the spread of COVID-19.Recent advances in deep learning and medical imaging have shown advances in the ability of computers to extract information from medical images,which in turn can play an important role in combating disease detection by building computer-aided diagnostic systems.However,the further development and optimisation of deep learning techniques for the characteristics of medical images and medical data for COVID-19 is important and challenging research.To address the above issues,this thesis focuses on COVID-19 computed tomography(CT)and chest x-ray(CXR)medical images to study and construct a COVID-19 medical image classification model with excellent performance,and the main research content and research work in this thesis are as follows.(1)The features of CT images include distinct local lesion features and global features.Therefore,this thesis proposes a convolutional neural network and Transformer two-branch model by investigating different structures of deep classification models to improve the extraction of local and global features of the model,and combine local and global features to improve the classification accuracy of the model for lesions.Firstly,the input images are data converted into two-dimensional images and one-dimensional vectors that meet the input requirements of Convolutional Neural Networks(CNN)and Transformer.Secondly,they are fed into the CNN and Transformer branches respectively,and the extracted local and global features are fused by feature stitching to achieve COVID-19 for CT medical image classification.Finally,model evaluation and experimental results analysis were carried out by in a set of CT medical images of COVID-19.The experimental results showed that the accuracy of disease detection in the triple classification task of CT medical images with COVID-19 reached 95.06%.(2)CXR medical image abnormality detection by COVID-19.Since CXR images have more dispersive local lesion features and global features compared to CT images.In this thesis,a fusion classification model based on convolutional neural network and Transformer is studied and constructed.Specifically,firstly,the model consists of a feature extraction sub-network,a feature focus sub-network and a feature classification subnetwork.The global and local features are extracted from the feature extraction subnetwork,and the global and local feature information is fused by the convolutional and Transformer feature interaction modules.Next,the extracted fused features are fed into the feature focus sub-network for deep feature extraction,and then,the final extracted fused features are classified by the feature classification sub-network.Finally,model testing and experimental results analysis were conducted to perform anomaly detection on a set of real CXR images of COVID-19.The experimental results showed that the accuracy,precision,recall and F1 scores of the fusion classification model of convolutional neural network and Transformer on CXR images of COVID-19 were 97.09%,97.16%,96.93% and 97.04%,respectively.Through the above techniques,this thesis investigates the abnormality detection technique for COVID-19 medical images.The experimental results show that the model by fusing local and global features of lesions has the accuracy and feasibility to help the rapid diagnosis and treatment of patients with COVID-19.