影像科学与光化学2025，Vol.43Issue(5)：60-67,8.DOI:10.7517/issn.1674-0475.2025.05.08
基于CT影像预测COVID-19严重程度的AI模型建立与验证研究
Establishment and Validation of AI Model for Predicting COVID-19-associ-ated Pneumonia Severity Based on CT Images
摘要
Abstract
Objective:This study aims to use artificial intelligence(AI)technology and based on CT image data to establish and verify a risk model that can predict disease severity(normal type and severe/critical type)of COVID-19-associated pneumonia patients,in order to provide basis for clinical decision-making and resource allocation.Methods:643 patients with COVID-19-associated pneumonia admitted from January 2022 to May 2023 were included in this study.Clinical data and CT image data of patients were collected,and quantitative parameters of CT images were measured by AI algorithm.According to the severity of the disease,the patients were divided into the common group(284 cases)and the severe or critical group(359 cases).Part of the clinical data and CT image data with significant differences between the two groups were included in LASSO regression analysis.The LASSO regression screening indicators were used as independent variables,and the factors affecting the occurrence of severe or critical COVID-19-associated pneumonia were included in multivariate logistic regression analysis,and a nomogram model was built according to the influencing factors.Receiver operating characteristic(ROC)curve was used to analyze the influencing factors and the predictive efficacy of the nomogram model for severe or critical COVID-19-associated pneumonia,and then the nomogram model was verified.Results:There were significant differences in age,ground-glass opacity(GGO)volume,solid volume,total lesion volume,GGO percentage,solid percentage and total lesion percentage between common and severe groups(P<0.05).Seven predictors of severe or critical COVID-19-associated pneumonia were identified by LASSO analysis,namely,age,GGO volume,solid volume,total lesion volume,GGO percentage,solid percentage,and total lesion percentage.Multivariate Logistic regression model analysis showed that GGO volume,consolidation volume,total lesion volume,GGO percentage,consolidation percentage,and total lesion percentage significantly affected the risk factors for severe or critical COVID-19-associated pneumonia(P<0.05).ROC curve analysis showed that GGO volume,solid volume,total lesion volume,GGO percentage,solid percentage,total lesion percentage,and area under the curve(AUC)of the nomogram model were 0.936,0.891,0.914,0.754,0.876,0.859,and 0.956,respectively.When cut-off value was taken,the sensitivity of GGO volume,solid volume,total lesion volume,GGO percentage,solid percentage,total lesion percentage and nomogram model were 0.858,0.797,0.767,0.657,0.722,0.833 and 0.941,respectively.The specificity was 0.964,0.843,0.915,0.743,0.895,0.738,0.962,respectively.The internal verification of the nomogram model using Bootstrap method showed that the prediction curve of the model basically coincides with the ideal line,which indicates that the prediction ability was good.The decision curve resulted show that when the high-risk threshold was between 0.0 and 1.0,the net return was greater than 0,and the predicted results had clinical significance.Moreover,the lower the high-risk threshold,the greater the net return.Conclusion:This study successfully established a COVID-19-associated pneumonia risk prediction model based on AI and CT images,which can effectively distinguish between ordinary and severe/critical patients,and has high clinical application value.关键词
COVID-19肺炎/AI/CT影像/影响因素/风险预测模型Key words
COVID-19-associated pneumonia/AI/CT image/influencing factors/risk prediction model分类
医药卫生引用本文复制引用
张俊丽,罗恒..基于CT影像预测COVID-19严重程度的AI模型建立与验证研究[J].影像科学与光化学,2025,43(5):60-67,8.基金项目
2019年四川省第二批卫生科研项目(19PJ284). （19PJ284）
