目的 联合人工智能预测概率与可解释的CT平扫影像学特征，构建并验证诊断主动脉夹层（aortic dissection，AD）的联合模型，以提升平扫CT在临床工作中筛查AD的价值。 方法 回顾性收集广州医科大学附属第二医院2020年7月至2024年2月行胸腹部CT平扫的患者，其中疑似AD患者221例，按7∶3的比例完全随机分为训练集（n = 153，其中AD患者76例）与验证集（n = 68，其中AD患者33例）。收集患者的临床资料及CT影像学特征，通过多因素logistic回归分析筛选AD的独立预测因子。最终筛选出的独立预测因子除年龄外均为影像学特征。分别构建基于影像学特征的传统影像模型、AI预测模型以及联合影像特征与AI预测概率的联合模型。采用受试者操作特征（ROC）曲线评估各模型的诊断效能，并通过DeLong检验比较曲线下面积（AUC）差异，模型的校准度通过Hosmer-Lemeshow检验和校准曲线（calibration curve）进行评估。 结果 多因素logistic回归显示年龄小、升主动脉管径增宽、管壁弧形高密度及主动脉腔内膜片形成是AD的独立危险因素（P < 0.05）。训练集中联合模型的AUC值显著优于传统影像模型及AI预测模型（0.921 vs. 0.897 vs. 0.779，P < 0.05）；验证集中联合模型保持稳健诊断效能（AUC = 0.841，敏感度84.8%，特异度77.1%）。模型具有较好的拟合效果（均P > 0.05），校准曲线显示联合模型诊断的AD发生风险与主动脉CTA得到的结果有较好的一致性。 结论 联合AI与CT平扫影像特征的诊断模型显著提升平扫CT对AD的识别能力，为无创筛查提供可靠工具。

Objective To develop and validate a combined diagnostic model for aortic dissection （AD） by integrating artificial intelligence （AI）?derived prediction probabilities with interpretable imaging features from non-contrast computed tomography （CT）， thereby enhancing its clinical utility for AD evaluation. Methods We conducted a retrospective study of 221 patients with suspected AD who underwent non-contrast chest and abdominal CT at our institution between July 2020 and February 2024. Patients were randomly allocated at a 7∶3 ratio into a training cohort （n = 153， 76 AD cases） and a validation cohort （n = 68， 33 AD cases）. Clinical variables and CT imaging features were extracted. Independent predictors of AD were identified using multivariate logistic regression. The final model incorporated younger age as the sole clinical predictor， alongside key imaging features. Three diagnostic models were constructed： a traditional feature-based model， a standalone AI model， and a combined model integrating both imaging features and AI-derived prediction probabilities. Diagnostic performance was evaluated using receiver operating characteristic （ROC） curve analysis， with area under the curve （AUC） values compared via the DeLong test. Model calibration was assessed using the Hosmer-Lemeshow test and calibration plots. Results Multivariate analysis identified younger age， ascending aortic dilation， crescentic high-attenuation intramural density， and an intimal flap as independent predictors of AD （all P < 0.05）. In the training cohort， the combined model yielded a significantly higher AUC than both the traditional model and the standalone AI model （0.921 vs. 0.897 vs. 0.779， respectively， all P < 0.05）. This advantage was confirmed in the validation cohort， where the combined model achieved an AUC of 0.841， with a sensitivity of 84.8% and specificity of 77.1%. All models demonstrated adequate calibration （Hosmer-Lemeshow P > 0.05）， and the calibration plot for the combined model showed strong agreement between predicted probabilities and actual AD diagnoses confirmed by aortic CTA. Conclusions Integrating AI-derived probabilities with interpretable non-contrast CT imaging features significantly enhances AD detection accuracy. This combined model offers a robust， non-invasive tool to aid in the rapid diagnosis and clinical triage of suspected AD.