目的 研究 CT 视觉定量评估和人工智能（AI）对进展期新型冠状病毒肺炎（COVID⁃19）严重 程度的评估和预测价值。方法 回顾性研究 86 例 COVID⁃19 患者进展期临床及影像资料，其中发展为重 型⁃危重型 31 例，普通型 55 例。分析并比较两组患者特征 CT 参数差异。利用受试者工作特征曲线（ROC 曲线）分析肺段累及数量和 CT 评分两种视觉定量评估方法评估和预测普通型与重型⁃危重型 COVID⁃19 的 能力。同时应用人工智能软件判断COVID⁃19患者严重程度，计算AI预判重型⁃危重型COVID⁃19的特异性 及敏感性。结果 重型⁃危重型组 CT 评分中位数和肺段累及数量中位数明显高于普通型组（P < 0.001）， 两组患者磨玻璃影（GGO）、实变、胸膜下线、“铺路石”征发生率差异无统计学意义。ROC 曲线分析结果表 明，CT 评分用于预判重型⁃危重型 COVID⁃19 的曲线下面积（AUC）为 0.879，截断值为 8.5 时，敏感性为 71.0%，特异性为 90.9%。肺段累及数量用于预判重症⁃危重型 COVID⁃19 的 AUC 为 0.883，截断值为 10.5 时，敏感性为 87.1%，特异性为 81.8%。AI 预判重型⁃危重型 COVID ⁃19 的敏感性为 77.4%，特异性为 92.7%。结论 CT 视觉定量评估和人工智能可用于评估和预测进展期 COVID⁃19 严重程度，为临床普通型

Objective To study the evaluation and prediction value of CT visual quantitative assessment combined with artificial intelligence（AI）for the severity of Coronavirus Disease 2019（COVID ⁃ 19）. Method The clinical and imaging data of 86 patients with COVID⁃19，including severe/critical cases（n = 31）and ordinary cases（n = 55），were studied retrospectively. The characteristic CT parameter variation of thetwo groups was analyzed and compared. The number of involved pulmonary segments was analyzed using the receiver operating characteristic curve（ROCcurve）. CTscores and two visual quantitative assessment methods were used to evaluate and predictthe ability ofordinary and severe⁃critical COVID⁃19. Moreover，the artificial intelligence software was utilized to predict the severity of COVID ⁃19 and calculatethe specificity and sensitivity of severe ⁃critical COVID ⁃ 19. Results The median CT score and the median number of involved pulmonary segments in thesevere ⁃critical groupwere signifi⁃ cantly higher than those in the ordinary group（P < 0.001）. The difference in ground⁃glass opacity（GGO），consoli⁃ dation，subpleural line，and“crazy⁃paving”sign between the two groupswas not statistically significant. The results of ROC curve analysis showed that the area under the curve（AUC）is 0.879 when the CT score was used for prediction of severe⁃critical COVID⁃19，and the sensitivity was 71.0%，and the specificity is 90.9% at the cutoff value of 8.5. The AUC was 0.883 when the number of involved pulmonary segments was used to predict the severe⁃ critical COVID ⁃ 19. When the cut ⁃ off value was 10.5，the sensitivity was 87.1% and the specificity was 81.8%. When AI was used for prediction of severe⁃critical COVID⁃19，the sensitivity was 77.4% and specificity was 92.7%. Conclusions CT visual quantitative assessment combined with artificial intelligence could be used to assess and predict theseverity of COVID⁃19，