Research On Assisted Diagnosis Technology Of New Coronary Pneumonia X-ray Chest Film Based On Deep Learning

Pneumonia is one of the common diseases that threaten the health of the lungs.As a kind of pneumonia with a wide spread,strong contagion,rapid onset,and severe impact.Once infected,New coronary pneumonia will severely affect the function of the respiratory system and severe cases will lead to respiratory failure and death in a short time.The key to controlling the epidemic is early detection,early isolation,and early treatment.How to assist doctors in quickly identifying patients with new coronary pneumonia is very important.With the continuous development of artificial intelligence,deep learning technology has become increasingly mature and widely used in the medical field.This dissertation applies deep learning technology to the identification and diagnosis of new coronary pneumonia,aiming to improve the recognition accuracy of new coronary pneumonia,quickly identify new coronary pneumonia,and reduce the risk of infection of social workers and medical staff.Based on the research on the auxiliary diagnosis of new coronary pneumonia,the paper proposes a pneumonia classification and detection method based on deep learning technology and oriented to pneumonia X-ray images.Firstly,the FC＿COVID-19 four-category data set was constructed and the data was preprocessed to reduce the impact of data-level problems on the model;secondly,the basic Dense Net-201 network was reused and the classification and detection research and detection were performed on the two new coronary pneumonia data sets and evaluate the results,select the commonly used hyperparameters and training optimization methods.Then the basic Dense Net-201 network is improved for the performance of the basic model in the evaluation results,and the maximum pooling layer,feature fusion layer,etc.are added,and then change the Re LU activation function in the network to the Leaky Re LU activation function.Finally,Dropout regularization is applied in the feature classification layer of the network to improve the performance by improving the feature extraction and nonlinear representation capabilities of the model.Finally,according to the proof of selected confusion,accuracy,loss value,precision rate,recall rate,F1 value and other evaluation criteria,the improved model are comprehensively evaluated and compared from two aspects of training process and experimental results by means of graphs and tables.The experiment focuses on the two data sets of COVID-19 and FC＿COVID-19 to verify the method proposed in the dissertation.Experimental results show that the accuracy rate under the COVID-19 data set has increased from 96.21% to 98.79%,which is a 1.09% improvement over the model of Chowdhury et al.,and a 0.69% performance improvement over the model of Feng Yibo et al.The accuracy rate of the new coronary pneumonia in the FC＿COVID-19 data set is improved from 79.36% to 84.52%.The accuracy,precision,recall and F1-score value of the improved model of new coronary pneumonia are as high as 100% on the COVID-19 data set,and in the FC＿COVID-19 data set both are close to 100%.At the same time,other categories on the two data sets have significantly improved on the four indicators.Through the detection and research of various types of pneumonia images,it is possible to classify various types of pneumonia accurately,improve the rapid detection and recognition ability of new coronary pneumonia,and reduce misdiagnosis and missed diagnosis.It provides an important reference basis for clinically assisting doctors in the diagnosis and treatment of new coronary pneumonia.