基于动脉瘤形态学及影像组学特征模型预测大脑中动脉镜像动脉瘤破裂的价值

目的：探讨基于动脉瘤形态学及影像组学特征Logistic回归模型预测大脑中动脉镜像动脉瘤的价值。方法：回顾性分析2010年5月至2020年10月温州医科大学附属第一医院收治的38例大脑中动脉镜像动脉瘤患者（一侧破裂，一侧未破裂）。在破裂和未破裂的镜像动脉瘤的配对中，使用头颅CT血管造影（CTA）测量了7个形态学参数，并使用Pyradiomics提取了12个影像组学衍生形态学参数。采用单因素分析筛选动脉瘤破裂的相关因素，根据筛选出的特征构建形态学、影像组学及混合模型以预测大脑中动脉瘤是否破裂，采用受试者工作特征曲线（ROC曲线）及曲线下面积（AUC）对模型的预测效能进行评估。结果：采用单因素分析后，筛选出7个形态学参数（动脉瘤大小、垂直高度、动脉瘤高度、瘤颈、AR、SR及形状）及6个影像组学参数（Maximum3DDiameter、Maxium2DDiameterSlice、Maximum2DDiameterColumn、Maximun2DDiameterRow、SurfaceArea及SurfaceVolumeRatio）。ROC曲线分析结果表明，混合模型的AUC为0.85，高于形态学模型（AUC=0.83）及影像组学模型（AUC=0.71）。结论：基于动脉瘤形态学及影像组学特征构建的Logistic回归模型可能有助于独立于患者个体特征而评估大脑中动脉镜像动脉瘤破裂的风险。

The predictive value of morphological and radiomics-derived features in the rupture of middle cerebral artery mirror aneurysms

bjective: To investigate the predicting value of morphological and radiomics-derived feature logistic model in the rupture of middle cerebral artery mirror aneurysm. Methods: Totally 38 patients with middle cerebral artery mirror aneurysms (one side ruptured and one side unruptured) in the First Affiliated Hospital of Wenzhou Medical University from May 2010 to October 2020 were retrospectively analyzed. In the pairing of ruptured and unruptured mirror aneurysms, 7 morphological parameters were measured using CTA, and 12 radiomics-derived morphological parameters were extracted using Pyradiomics. Univariate analysis was used to select the related factors of aneurysm rupture. Logistic regression model for predicting the rupture of the middle cerebral artery mirror aneurysms was established based on morphological features, radiomics-derived features and morphological features combined with radiomics-derived features. Receiver operating characteristic curve and area under curve were used to evaluate the predictive performance of the model. Results: After univariate analysis, 7 morphological parameters (including aneurysm size, perpendicular height, aneurysm height, aneurysm neck size, aspect radio, size radio and aneurysm shape) and 6 radiomics-derived parameters (including Maximum3DDiameter, Maxium2DDiameterSlice, Maximum2DDiameterColumn, Maximun2DDiameterRow, SurfaceArea and SurfaceVolumeRatio) were selected. The results of the ROC curve were as follows: The AUC of combined model was 0.85, which was higher than that of morphological model (AUC=0.83) and radiomics model (AUC=0.71).Conclusion: Logistic regression model based on morphological and radiomics-derived features may be helpful in assessing the risks of middle cerebral artery aneurysm rupture independent of individual patient’s characteristics.