Diagnosis And Analysis Of COVID-19 Based On Multi-View Machine Learning

Since the end of 2019,the novel coronavirus pneumonia epidemic(COVID-19)has raged around the world and brought huge challenges to the safety of human life.The sudden outbreak of medical and health public incidents has greatly tested the Chinese government’s ability to deal with health emergencies.In the early days of the outbreak in Wuhan,Hubei,RT-PCR detection results were used as the gold standard for diagnosing COVID-19.However,with the developing of epidemic,it is facing many urgent problems.First of all,in the face of a sudden outbreak,medical personnel are also gradually exploring the process of understanding the virus.The National Health Commission of China has iteratively released a total of eight versions of the ”COVID-19 Diagnosis and Treatment Plan”.Among them,in order to ensure that patients should be collected as much as possible,the fifth edition published on February 5,2020 specially revised the suspected case diagnosis standards for Hubei,and tried to reduce the spread of the epidemic,which caused a significant increase in the suspected cases in the report.This sparked widespread heated debate.In response to this problem,this article established an multi-dimensional siscriminative evaluation system based on machine learning to theoretically analyze and explain the plan made at that special period.Secondly,we are faced with insufficient RT-PCR detection capabilities,insufficient detection reagents,long detection time,and large number of patients.To cope with this problem,this article proposes a rapid diagnosis model of COVID-19 based on CT images.Specifically,the main contributions of this article can be summarized as follows:(1)Faced with two groups of patients,that is,the clinical manifestations are consistent with the suspected cases in the diagnosis and treatment plan but the RT-PCR detection results are different,this article proposes an evaluation system based on machine learning.From the two dimensions of increasing information and increasing method complexity,the correlation,clustering and discriminativeness of the two groups of patients were explored.Experiments have verified that the two groups of patients have low discrimination,and at the same time it also proved the professionalism and timeliness of the diagnosis and treatment plan at that special period.(2)Diagnosis of COVID-19 with structured latent multi-view representation learning.Combined with multi-view learning,in order to fully dig out the multiple types of features that describe CT images from different angles,the model learns a unified latent representation that can completely encode different aspects of the feature information(completeness),and has a good category separability(structured).The latent representation encoded from observations is complete and versatile,thus enhances the prediction performance,while the clustering-like classification schema in turn enhances the separability of the latent representation.Specifically,The latent representation encoded from observations is complete and versatile,thus enhances the prediction performance,while the clustering-like classification schema in turn enhances the separability of the latent representation.A large number of experimental results have verified the effectiveness of the model and shows quite stable performance when the amount of training data changes.