目的 探讨冠脉CT血管造影（CCTA）斑块人工智能（AI）定量参数与CT血流储备分数（FFR-CT）的相关性。 方法 收集怀疑冠心病患者共84例，男52例，女32例，年龄27 ~ 81岁，平均（58.1 ± 11.9）岁，均行CCTA检查。将图像传输至数坤软件并对冠脉血管标记、斑块分析、计算相应FFR-CT值。斑块AI参数包括长度、体积、最小管腔面积（MLA）、最狭窄程度（MLD）、脂类体积及占比、纤维脂类体积及占比、纤维体积及占比、钙化体积及占比。FFR-CT≤0.8定义为可能存在血流动力学异常或心肌缺血。评估斑块参数与FFR-CT相关性，采用二元logistic回归筛选冠脉血流储备异常（FFR-CT ≤ 0.8）的独立危险因素，采用ROC曲线及曲线下面积（AUC）评价其诊断效能，计算灵敏度、特异度及模型预测准确率。 结果 84例患者主要症状为胸痛（39例，占46.4%）与胸闷（27例，占32.1%）。Spearman相关分析结果显示FFR-CT与MLA呈正相关（r = 0.49，P < 0.000 1），与长度、体积、脂类体积、纤维脂类体积、纤维体积及钙化体积呈负相关（r = -0.44、-0.56、-0.40、-0.36、-0.42、-0.40，P < 0.05），其中与MLD呈中等程度负相关（r = -0.60，P < 0.000 1）。单因素logistic回归分析结果显示斑块长度、体积、MLA、MLD、脂类体积、纤维脂肪体积、纤维体积、钙化体积等参数是FFR-CT ≤ 0.8的独立危险因素（P < 0.05）。经调整后的多因素logistic回归分析结果显示，MLD是FFR ≤ 0.8的独立危险因素（OR= 1.082，95%CI：1.034 ~ 1.133，P = 0.001），预测模型为logit（P） = 0.079X₁ - 4.052，X₁为MLD值，预测准确率为85.2%。斑块长度、MLD、脂类体积、纤维体积及钙化体积诊断冠脉血流动力学异常（FFR-CT ≤ 0.8）的AUC分别为0.796、0.886、0.711、0.754、0.698，灵敏度与特异度分别为47.83%、73.91%、73.90%、52.17%、60.87%与92.11%、73.68%、60.53%、84.21%、89.47%，5种指标联合诊断的AUC为0.906，灵敏度与特异度为73.91%、71.05%。 结论 冠脉斑块AI参数与FFR-CT存在不同程度相关性，MLD是FFR-CT ≤ 0.8的独立危险因素，具有较高的诊断效能。

Objective To investigate the relationship between coronary artery plaque AI quantitative parameter and FFR-CT in coronary computed tomography angiography. Methods A total of 84 patients suspected of having CAD ［52 males and 32 females， aged 27 to 81 years with a mean age of （58.1 ± 11.9） years］ were enrolled in this study. All patients underwent coronary computed tomography angiography. The CCTA data was processed using shukun（SK） software for labeling and analysis of the coronary arteries， as well as obtaining quantitative parameters of coronary artery plaque AI and corresponding FFR-CT values. The quantitative parameters included plaque length， total volume， minimum lumen area （MLA）， minimal lumen degree （MLD）， lipid composition volume and proportion， fibrous-lipid composition volume and proportion， fibrous composition volume and proportion， calcified composition volume and proportion. Coronary artery hemodynamic abnormality or myocardial ischemia was defined as an FFR-CT value ≤ 0.8. Correlational analysis was performed to evaluate the association between AI plaque quantitative parameters and FFR-CT values. Univariate and multivariate binary logistic regression analyses were conducted to identify independent risk factors for predicting FFR-CT ≤ 0.8. The predictive performance of the model based on AI plaque quantitative parameters was assessed using receiver operating characteristic （ROC） curve analysis and calculation of the area under the curve （AUC）. Sensitivities， specificities， diagnostic test accuracy rates were also calculated. Results The predominant symptoms observed in the cohort of 84 patients were chest pain （n = 39， 46.4%） and distress （n = 27， 32.1%）. Spearman analysis results revealed a weak positive correlation between FFR-CT and MLA （r = 0.49， P < 0.000 1）， while weak negative correlations were found for plaque length， total volume， lipid composition volume， fibrous-lipid composition volume， fibrous composition volume， and calcified composition volume （r = -0.44， -0.56， -0.40， -0.36， -0.42， -0.40； all P < 0.05）. Additionally， MLD exhibited a moderate negative correlation with FFR-CT （r = -0.60， P < 0.000 1）. In the univariate binary logistic regression analysis， several variables including plaque length， total volume， MLA， MLD， lipid composition volume， fibrous-lipid composition volume， fibrous composition volume， and calcified composition volume were found to be independently associated with FFR-CT ≤ 0.8 （All P < 0.05）. The adjusted multivariate binary logistic regression analysis model revealed that MLD was the sole independent predictor （OR= 1.082， 95%CI：1.034 ~ 1.133， P = 0.001）. The logistics regression model expression was logit（P）=0.079X₁ - 4.052， where X₁ represents the value of MLD and achieved a predictive accuracy of 85.2%. The ROC AUC of plaque length， total volume， MLA， MLD， lipid composition volume， fibrous-lipid composition volume， fibrous composition volume and calcified composition volume were 0.796， 0.886， 0.711， 0.754 and 0.698 respectively， and the coresponding sensitivities and specificities were 47.83%， 73.91%， 73.90%， 52.17%， 60.87% and 92.11%， 73.68%， 60.53%， 84.21%， 89.47%. The five indexes combined diagnostic model possessed the largest AUC of 0.906， and 73.91%， 71.05% of sensitivity and specificity. Conclusion The AI quantitative parameters of coronary artery plaque exhibited varying degrees of correlation with FFR-CT， while MLD emerged as the sole independent predictor of FFR-CT ≤ 0.8， demonstrating high diagnostic efficiency.