基于常规MRI VASARI特征的脑胶质瘤Ki-67表达水平预测模型的建立与验证

Development and Validation of a Predictive Model for Ki-67 Expression Level in Gliomas Based on Conventional MRI VASARI Features

摘要: 目的：本研究旨在基于常规MRI VASARI特征构建并验证脑胶质瘤Ki-67表达水平的预测模型。方法：本回顾性研究共纳入241例脑胶质瘤患者，按照7:3的比例随机分为训练集(n = 168)和验证集(n = 73)。收集参与者的临床资料及术前MRI影像，提取相关VASARI影像学特征。通过单因素分析和多因素逐步回归分析筛选候选变量，建立Ki-67表达水平的Logistic回归预测模型。模型预测效能根据受试者工作特征(receiver operating characteristic, ROC)曲线下面积(area under curve, AUC)评估。采用Hosmer-Lemesshow检验模型拟合优度，并绘制校准曲线及决策曲线。结果：单因素与多因素回归分析结果显示，强化程度、肿瘤跨中线和弥散受限是Ki-67表达水平的独立预测因素。基于上述3项特征构建的预测模型在训练集和验证集中的AUC分别为0.886和0.848，Hosmer-Lemeshow检验表明模型拟合良好(训练集χ² = 3.727，p = 0.444，验证集χ² = 5.901，p= 0.207)。校准曲线显示模型的校准度较好，决策曲线分析表明模型具有较高的净收益。结论：基于常规MRI VASARI特征，我们成功构建了可有效预测脑胶质瘤Ki-67表达水平的回归模型，或可为脑胶质瘤的个体化治疗决策提供参考。

Abstract: Objective: This study aims to develop and validate a predictive model for Ki-67 expression levels in gliomas based on conventional MRI VASARI features. Methods: A total of 241 glioma patients were retrospectively included and randomly divided into a training set (n = 168) and a validation set (n = 73) at a ratio of 7:3. Clinical data and preoperative MRI images were collected, and relevant VASARI imaging features were extracted. Candidate variables were selected using univariate analysis and multivariate stepwise regression analysis to construct a Logistic regression model for predicting Ki-67 expression levels. The model’s predictive performance was evaluated using the area under the receiver operating characteristic (ROC) curve (AUC). The Hosmer-Lemeshow test was used to assess model goodness-of-fit, and calibration curves and decision curve analysis (DCA) were performed. Results: Univariate and multivariate regression analyses identified enhancement degree, midline crossing, and diffusion restriction as independent predictors of Ki-67 expression levels. The predictive model incorporating these three features achieved AUCs of 0.886 in the training set and 0.848 in the validation set. The Hosmer-Lemeshow test indicated a good model fit (training set: χ² = 3.727, p = 0.444; validation set: χ² = 5.901, p = 0.207). The calibration curve demonstrated good agreement between predicted and actual values, while decision curve analysis indicated a high net benefit. Conclusion: We successfully developed a regression model based on conventional MRI VASARI features to effectively predict Ki-67 expression levels in gliomas. This model may serve as a valuable tool for individualized treatment decision-making in glioma management.