Research On Deep Learning Methods For Medical Image Recognition Of Coronavirus Disease 2019

Medical imaging is an important means for radiologists to diagnose diseases.The imaging doctor judges the disease by analyzing and comparing the medical images of the living tissue of the patients.However,the diagnostic accuracy depends on the level and energy of radiologists and the level of medical equipment.In recent years,artificial intelligence technology in computer vision,natural language processing and many other fields has excellent performance.The application of artificial intelligence technology in medical imaging has attracted the attention of clinical experts and artificial intelligence experts.The end-to-end deep learning method is a kind of artificial intelligence technology.Compared with traditional image processing technology,deep learning method has been proved to have strong feature representation ability,and has excellent performance in the field of image analysis.In order to improve the accuracy of disease diagnosis and relieve the pressure of medical staff,the deep learning method was applied to analyze medical imaging to assist radiologists in the diagnosis of diseases.Since the outbreak of coronavirus disease 2019(COVID-19),the epidemic situation in many affected countries or regions has not yet been effectively controlled by disease.Timely detection,isolation and treatment of patients can effectively prevent the spread of the virus,so it is urgent to develop intelligent and efficient diagnostic methods.It is a safe and effective way to use deep learning method to detect infection in medical images.Hospital hardware equipment is poor;Compared with natural images,medical images contain less features.To solve the above problems,the following two lightweight intelligent methods for automatic recognition of covid-19 medical images are proposed.(1)This paper proposes a deep learning method for recognition of cvid-19 chest X-ray(CXR)images.Because CXR images have high similarity of sample features of the same category and low variability of sample features of different categories,CXR images contain less features.A dual path multi-scale fusion(dual path multi-scale fusion)is proposed,Based on this,a lightweight convolutional neural network model DD covidnet is proposed to realize three classification.The method is verified on two datasets.The experimental results show that the sensitivity of DD covidnet model to cowid-19 recognition is 96.08%,Compared with other models,DD covidnet model has faster detection speed and more accurate results.(2)A deep learning method based on computed tomography(CT)image is proposed to detect covid-19 images.The overall characteristics of covid-19 chest CT images are multiple patchy exudative shadows in both lungs,most of them are ground glass shadows,and some of them are local consolidation.In view of the above characteristics,A parallel channel shuffle(PCS)module with strong sensing ability is proposed.PCS module is divided into two versions,which are PCS-D module using for downsampling and standard version PCS-S module.Coordinate attention mechanism is introduced into the two modules to obtain PCS-DCA module and PCS-S-CA module.Based on PCS module,a lightweight convolutional neural network model PC covidnet is proposed In PCS module,maximum fuzzy pooling is introduced to enhance the robustness of the model;channel shuffling is introduced to reduce the complexity of the model;coordinate attention mechanism is introduced to focus on the channel information and coordinate information of features,so as to enhance the classification effect of the model.The experimental results show that PC covidnet model has excellent detection performance.