Research On Image-assisted Diagnosis And Real-time Detection Of New Coronary Pneumoni

As the COVID-19 pandemic continues to spread globally,effective disease detection is essential for controlling its spread.However,reverse transcription-polymerase chain reaction(RT-PCR)testing,which is widely used,may have limited sensitivity,leading to false-negative results and increasing transmission risks.As a result,chest imaging has become a crucial complementary diagnostic tool to overcome the limitations of RT-PCR testing.Traditional chest imaging diagnoses require experienced doctors and may be time-consuming,which may be problematic during a pandemic outbreak.Computer-aided chest imaging diagnosis for COVID-19 has gained considerable attention due to the advancement of computing power and the development of deep learning techniques.However,the current COVID-19 chest imaging assisted diagnosis techniques are limited by the size of the dataset,overfitting,and low accuracy.Moreover,in practical applications,device capability may limit the deployment of highperformance models,and data transmission may result in the leakage of patient privacy information.To address these issues,this thesis proposes two research directions:(1)To address the problem of over-fitting and low accuracy of the COVID-19 lung imaging-assisted diagnosis technology,this paper proposes a dual-stream network based on Efficient Net for COVID-19 chest imaging assisted diagnosis.The dual-stream network considers the significant information contained in the frequency domain of CT images that network models have frequently overlooked.To avoid overfitting the dataset,the Adversarial Propagation(Adv Prop)technique is used to generate adversarial samples as additional samples.In addition,a feature pyramid network(FPN)is used to fuse dual-stream features.Finally,the experimental results on the public dataset COVIDx CT-2A demonstrate that the proposed method achieves an accuracy of 98.70%,which outperforms existing methods.(2)To address the problem of lung image diagnosis for COVID-19 under the constrained computing environment,this paper proposes a real-time diagnosis method based on edge computing.First,the model proposed in(1)is pruned,quantized,and deployed on the edge to construct a real-time detection framework,which compresses the model size to fit the conditions for running on edge devices.Then,an improved Strength Pareto Evolutionary Algorithm(SPEA2)is used to obtain a set of balanced service migration strategies.Finally,multi-objective attribute decision-making technology is used to evaluate all service migration strategies in the set to obtain the final service migration strategy,thereby achieving joint optimization of network load balancing,service migration time,and privacy entropy.Experimental results show that the proposed method achieves network load balancing and protects patient privacy with a short service migration time.