| In recent years,novel coronavirus(COVID-19)has become the most serious epidemic in the world,and humans may coexist with its virus for a long time.Because of Covid-19’s strong infectivity,the medical systems of various countries are facing a strong test.Traditional nucleic acid detection results need to wait for a certain time,while the manual diagnosis of lung images is too dependent on professional knowledge.When serious suspected patients are faced,they need to be diagnosed quickly so as to get timely treatment.The flat panel detector in DR system provides effective image features of lungs after receiving X-rays,and medical staff can judge the infection of lungs from the images.However,how to realize the efficient diagnosis of COVID-19 has become a difficult problem in primary medical units with insufficient medical resources,limited area or a large number of patients.Under this background,to realize accurate and rapid auxiliary diagnosis of patients,relevant applicable equipment and image recognition algorithms can be used.In this paper,the current DR system is systematically expounded.In order to assist medical staff in diagnosis,the image recognition of COVID-19 is mainly studied,and the image recognition algorithm of COVID-19 is designed.At the same time,in order to facilitate the flat-panel detector to receive X-rays,a multifunctional buckywall frame structure is designed to expand the application range of the flat-panel detector.The related deep learning framework and simulation software provide the foundation for this research.To realize the image recognition of COVID-19,Tensor Flow deep learning framework and open source lung image data set are adopted to design the image recognition algorithm.Based on the deep learning algorithm,Res Net-50 is used as the skeleton network for downsampling,and the decoder of U-Net is used for upsampling,so as to segment the lung region of the image,thus eliminating the influence of the image feature information of the non-lung region on the recognition effect.Compared with other networks as the skeleton network,its segmentation accuracy rate is more than 90%,which is better.In order to realize the accurate diagnosis of COVID-19’s disease on a small model,the segmented lung region was identified by Efficient Net-B0 network,which effectively identified COVID-19’s disease,normal disease and viral pneumonia.Experiments show that compared with other recognition networks,Efficient Net-B0 is more concise and efficient.Under various thresholds,the comprehensive evaluation index AUC of this model reaches 99.9%,which is higher than the recognition networks such as Xception-V3 and Xcept.The flat panel detector in the DR system receives X-rays and images them.As the clamping device of the flat panel detector,the chest X-ray stand can only move the flat panel detector up and down.Therefore,at last,this paper designs a multifunctional chest X-ray stand,which is simple and easy to use,and can adapt to various shooting situations.On this basis,the recognition model is deployed visually.The widespread spread of Covid-19 has brought great challenges to primary medical institutions in many countries.In this paper,COVID-19 image recognition algorithm is designed to improve the diagnosis efficiency of primary medical units,which can provide reference for related disease detection. |
