Analysis Of Pulmonary AI Imaging Of Novel Coronavirus Infection And Correlation With Clinical Prognosis

ObjectiveThis study quantifies the volume of lung lesions at the time of the most severe clinical symptoms in COVID-19 patients by using the Artificial Intelligence(AI)diagnostic imaging system-Coronavirus Disease 2019(COVID-19),and analyzes the volume of lung lesions at the time of the most severe clinical symptoms in COVID-19 patients with different clinical prognosis.The basic clinical data of COVID-19 patients with different clinical prognosis,chest CT performance at the time of the most severe clinical symptoms,to explore the lung AI images of COVID-19 patients and their association with clinical prognosis,to provide a basis for clinical diagnosis and treatment,and to improve radiologists’ understanding of chest CT imaging performance of COVID-19 patients.MethodsA case-control study method was utilized.A total of 122 patients with COVID-19 were included,including 75 patients in the good prognosis group and 47 patients in the poor prognosis group.The volume of pulmonary lesions,including ground-glass opacity(GGO)volume and solid shadow volume,the main distribution location of lesions,bronchial inflation sign,mediastinal lymph node enlargement and pleural effusion were recorded using the medical AI diagnostic imaging system-COVID-19 auxiliary analysis software.The chi-square test and Fisher’s exact probability method were selected to analyze the basic clinical data,chest CT performance at the worst clinical symptoms of COVID-19 patients with different clinical prognosis;logistic regression and Cox regression were used to analyze the factors influencing the poor prognosis of COVID-19patients;KM survival curve was drawn using R language,and Nomogram prediction model was constructed with ROC,DCA to evaluate this Nomogram prediction model.Results1.Age,underlying diseases(including hypertension,diabetes,coronary heart disease,chronic kidney disease,cerebral infarction,hepatitis B,tumor,and others),and clinical staging were statistically significant(P<0.05)between the good prognosis group and the poor prognosis group.There was no statistically significant difference between the good prognosis group and the poor prognosis group for gender,exposure history or history of travel to Hubei area,and clinical presentation(P>0.05).2.The differences in GGO volume,solid shadow volume,location of the main distribution of lesions,and bronchial inflation sign in COVID-19 patients with the most severe clinical symptoms were statistically significant between the good prognosis group and the poor prognosis group(P<0.05).mediastinal lymph node enlargement and pleural effusion were less frequent in COVID-19 patients,and the differences between the good prognosis group and the poor prognosis group were not statistically significant(P> 0.05).3.The last follow-up chest CT lesions had the most peripheral distribution,and the GGO volume,solid shadow volume,bronchial inflation sign,mediastinal lymph node enlargement,and pleural effusion were significantly reduced compared with the chest CT at the time of the most severe clinical symptoms,and the differences between the two groups were statistically significant(P<0.05).4.The differences in clinical prognosis were statistically significant(P<0.05)between the mild and severe disease group,the underlying disease group,the main distribution location of the lesions,and the bronchial inflation group.5.Single-factor logistic regression analysis showed that age,underlying disease,clinical staging,GGO volume,solid shadow volume,distribution of lesions in the central+ periphery,and bronchial inflation sign were independent risk factors for poor prognosis in patients with COVID-19.Multifactorial Cox regression analysis showed that age,underlying disease,and GGO volume were independent risk factors for poor prognosis in patients with COVID-19.Based on the results of logistic regression,Cox regression and clinical experience,age,underlying disease,GGO volume,solid shadow volume,main distribution location of lesions,and bronchial inflation sign were selected to construct a Nomogram prediction model,and the ROC curve analysis showed that the model AUC value was 0.864,sensitivity was 91.5%,specificity was 65.3%,and the area under the curve of each risk factor The DCA showed that using this prediction model to assess the prognosis of COVID-19 patients could bring higher clinical benefit under certain risk threshold.Conclusion1.Patients with COVID-19 have fever and cough as the most predominant clinical symptoms,followed by malaise and shortness of breath;hypertension,diabetes and coronary heart disease are the three most common underlying diseases in patients with COVID-19.2.In the good prognosis group and the poor prognosis group,patients with COVID-19 both showed GGO and solid shadow on CT at the time of the most severe clinical symptoms,and the lesions were most distributed in the peripheral + central area,and the GGO volume,solid shadow volume and lesion distribution were larger in the poor prognosis group than in the good prognosis group.Patients with COVID-19 were less likely to have enlarged mediastinal lymph nodes and pleural effusions.3.The mean follow-up time in the poor prognosis group was approximately twice as long as in the good prognosis group.During the follow-up period,patients in the severe group,the group with underlying disease,the group with lesions distributed peripherally+ centrally,and the group with bronchial inflation signs had a worse prognosis than the corresponding control group.4.A prognostic model based on imaging and clinical variables that incorporates age,underlying disease,GGO volume,solid shadow volume,location of major distribution of lesions,and bronchial inflation signs was developed for noninvasive,individualized prediction of COVID-19 prognosis.With reference to this Nomogram prediction model,patients can directly derive the predicted probability of poor prognosis based on individual information for risk ranking.