【摘要】 目的 构建基于增强CT影像组学模型用于术前预测膀胱尿路上皮癌（BUC）的组织学分级。资料与方法 回顾性分析2016年11月—2020年9月广州市第一人民医院127例经病理证实的高级别尿路上皮癌和低级别尿路上皮癌，随机分为训练集89例和验证集38例。在平扫、动脉期、静脉期CT图像上提取病灶影像组学特征。使用Logistic回归构建影像组学预测模型，绘制受试者工作特征曲线，并计算曲线下面积评价模型预测BUC组织学分级的效能，采用决策曲线评价模型鉴别高级别尿路上皮癌和低级别尿路上皮癌的净获益。结果 基于每例患者的三期CT图像共提取出5 202个影像组学特征，经过特征筛选后最终得到20个特征用于构建预测模型。模型在训练集及验证集中诊断BUC组织学分级的曲线下面积分别为0.90(95%CI0.83～0.96)和0.93(95%CI 0.85～1.00)。结论 基于增强CT影像组学构建的模型术前预测BUC组织学分级具有良好的诊断效能。更多还原

【Abstract】 Purpose To develop an enhanced CT-based radiomics model for predicting histological grading in patient with bladder urothelial carcinoma(BUC) before surgery. Materials and Methods This retrospective study involved 127 patients pathologically confirmed with high-grade urothelial carcinoma and low-grade urothelial carcinoma from November 2016 to September 2020 in Guangzhou First People’s Hospital. All patients were randomly divided into two groups, including training set(n=89) and validation set(n=38). Radiomics features of each lesion were extracted from the unenhanced, arterial and venous phase, respectively. A predictive model based on radiomics for predicting histological grading of BUC was established via Logistics regression. Predicting performance was evaluated by drawing receiver operator characteristic curve and calculating area under the curve. The decision curve was plotted to evaluate the net benefits of the predicting model for differentiating high-grade urothelial carcinoma from low-grade urothelial carcinoma. Results Of the 5 202 features extracted from three phases of CT images, a total of 20 features were eventually selected to develop the predictive model. The area under the curve of model to predict histological grading of BUC was 0.90(95% CI 0.83-0.96) in training set and 0.93(95% CI 0.85-1.00) in validation set, respectively.Conclusion Radiomics model based on enhanced CT has the favorable performance in predicting histological grading of BUC before surgery.更多还原