Lung X-ray Image Classification And Object Detection Based On Deep Learning

Pneumonia is a disease with high morbidity and mortality for children and the elderly.With the continuous improvement of modern medical technology,remarkable achievements have been made in the treatment of pneumonia.The diagnosis of pneumonia is a very important step before treatment.However,due to the interference of other lung diseases,the explosion of medical data and the lack of professional pathologists,the accurate diagnosis of pneumonia is difficult.How to achieve "early detection and early treatment" of pneumonia is of great significance for the treatment effect of pneumonia patients.With the development of modern medicine,clinical diagnosis rely more on medical image data.With the increasing amount of data,a powerful data processing model is urgently needed to provide favorable support for the field.As a new branch in the field of machine learning,deep learning has been widely used in many fields of medical image processing,providing new methods and approaches for intelligent detection of medical images.Web technology allows real-time interaction between users and servers via Internet.It has the ability of large-scale data transmission and the advantages of simple and convenient operation.Users can send requests to the server to implement corresponding functions.The combination of deep learning with Web technology provides a solution for the problems of shortage of professional doctors,low diagnostic efficiency and accuracy resulting from medical data explosion.Therefore,this paper studies lung X-ray image classification and object detection based on deep learning,and builds the lung X-ray image data processing platform using Web technology.The main research contents and results are as follows:(1)Classification and recognition of lung X-ray images based on deep learning.Firstly,VGG19 network model is used to recognize pneumonia / non-pneumonia from more than5000 lung X-ray images,and its performance is evaluated from six aspects.The results showe that the accuracy of VGG19 model is 98.3%,the recall rate is 99.0%,the precision is98.7%,the specificity is 96.3%,the AUC value is 0.976,and the F1 score is 0.988.Then,VGG19 model and two improved models,namely VGG19+SVM model and VGG19+XGBoost model,are used to classify viral pneumonia and bacterial pneumonia.Meanwhile,the difference of the classification effect of the three models is compared.The results show that the average accuracy of the three models are all above 85.9% and the test stability of VGG19+XGBoost model is better than VGG19 model.The highest accuracy of VGG19+XGBoost model is close to 90%,the recall rate is 88.1%,the precision is 96.8%,the specificity is 92.7%,the AUC value is 0.902,and the F1 score is 0.923.This indicates that the algorithm model of deep learning combined with machine learning has practical significance.(2)Object detection of lesion target in pneumonia images based on deep learning.Firstly,two typical object detection algorithms,SSD and Faster-RCNN,are used for detection of pneumonia lesion target.Then,the FPN network model is introduced into the Faster-RCNN network model for optimization to better recognize lesion target.Finally,500 randomly selected images are tested to compare the performance of the three detection algorithms in terms of classification accuracy,regression accuracy and misdetected lesion target.The results show that for SSD algorithm,the average classification accuracy of SSD is 86.6%,the average regression accuracy is 67.4%,and the number of misdetected lesion target is 11;for Faster-RCNN algorithm,they are 90.20%,74.6%,and 3.respectively;For Faster-RCNN optimization algorithm,they are 93.7%,79.8%,and 0.Therefore,Faster-RCNN optimization algorithm has the best detection effect.(3)Construction of a Web service platform for lung X-ray image processing.A variety of lung X-ray image processing models are integrated into one system using Web technology.The user can process all the lung X-ray images via Internet.The results show that the frontend page of the Web platform has clear functions,friendly interface,good operability and stability,and meets actual demands of accuracy and efficiency.Experiments have proved that the Web service platform using the Internet as a medium,can not only eliminate the regional differences in medical levels,but also help non-professional pathologists share the pressure of related professional doctors,thereby providing better medical services for patients with pneumonia.