Departments of Radiology, Jiangxi Tumor Hospital, Nanchang, Jiangxi, China (X.Z., Y.S., C.G., L.L.).
Departments of Radiology, The First Affiliated Hospital of Nanchang University, Nanchang, Jiangxi, China (Y.H., M.W., X.L.).
Acad Radiol. 2024 Apr;31(4):1560-1571. doi: 10.1016/j.acra.2023.09.010. Epub 2023 Oct 19.
To assess the predictive ability of intratumoral and peritumoral multiparametric magnetic resonance imaging (MRI)-based radiomics signature (RS) for preoperative prediction of Ki-67 proliferation status in glioblastoma. MATERIALS AND METHODS: A total of 205 patients with glioblastoma at two institutions were retrospectively analyzed. Data from institution 1 (n = 158) were used to develop the predictive model, and as an internal test dataset, data from institution 2 (n = 47) constitute the external test dataset. Feature selection was performed using spearman correlation coefficient, univariate ranking method, and the least absolute shrinkage and selection operator algorithm. RSs were established using a logistic regression algorithm. The predictive performance of the RSs was assessed using calibration curve, decision curve analysis (DCA), and area under the curve (AUC).
In the RSs based on single-parametric (contrast-enhanced T1-weighted image, T2-weighted image, or apparent diffusion coefficient maps), the AUCs of intratumoral, peritumoral, and combined area (intratumoral and peritumoral) were 0.60-0.67, with no significant difference among them. The RSs that using multiparametric features (integrating the previously mentioned three sequences) showed improved AUC compared to the single-parametric RSs; AUC reached 0.75-0.89. Among them, the multiparametric RS based on radiomics features of the combined area (Multi-Com) exhibited the highest performance, with an internal test dataset AUC of 0.89 (95% confidence interval (CI) 0.75-1.00) and an external test dataset AUC of 0.88 (95% CI 0.78-0.97). The calibration curve and DCA display RS (Multi-Com) have good calibration ability and clinical applicability.
The multiparametric MRI-based RS combining intratumoral and peritumoral features can serve as a noninvasive and effective tool for preoperative assessment of Ki-67 proliferation status in glioblastoma.
评估基于肿瘤内和肿瘤周围多参数磁共振成像（MRI）的放射组学特征（RS）预测胶质母细胞瘤术前 Ki-67 增殖状态的能力。
回顾性分析了来自两个机构的 205 名胶质母细胞瘤患者的数据。机构 1（n=158）的数据用于开发预测模型，机构 2（n=47）的数据作为内部测试数据集。使用 Spearman 相关系数、单变量排序法和最小绝对收缩和选择算子算法进行特征选择。使用逻辑回归算法建立 RS。使用校准曲线、决策曲线分析（DCA）和曲线下面积（AUC）评估 RS 的预测性能。
在基于单参数（增强 T1 加权像、T2 加权像或表观扩散系数图）的 RS 中，肿瘤内、肿瘤周围和联合区域（肿瘤内和肿瘤周围）的 AUC 为 0.60-0.67，无显著差异。使用多参数特征（整合上述三种序列）的 RS 与单参数 RS 相比，AUC 有所提高；AUC 达到 0.75-0.89。其中，基于联合区域放射组学特征的多参数 RS（Multi-Com）表现最佳，内部测试数据集 AUC 为 0.89（95%置信区间 0.75-1.00），外部测试数据集 AUC 为 0.88（95%置信区间 0.78-0.97）。校准曲线和 DCA 显示 RS（Multi-Com）具有良好的校准能力和临床适用性。
基于多参数 MRI 的 RS 结合肿瘤内和肿瘤周围特征，可作为胶质母细胞瘤术前评估 Ki-67 增殖状态的一种非侵入性有效工具。
