Department of Radiology, Peking Union Medical College Hospital, Peking Union Medical College, Chinese Academy of Medical Sciences, Shuaifuyuan No. 1, Wangfujing Street, Dongcheng District, Beijing, 100730, China.
Deepwise AI Lab, Deepwise Inc., Haidian Avenue No. 8, Sinosteel International Plaza, Beijing, 100080, China.
Eur Radiol. 2020 Dec;30(12):6749-6756. doi: 10.1007/s00330-020-06893-8. Epub 2020 Jun 29.
To build a CT-based radiomics model to predict the pathological grade of bladder cancer (BCa) preliminarily.
Patients with surgically resected and pathologically confirmed BCa and who received CT urography (CTU) in our institution from October 2014 to September 2017 were retrospectively enrolled and randomly divided into training and validation groups. After feature extraction, we calculated the linear dependent coefficient between features to eliminate the collinearity. F-test was then used to identify the best features related to pathological grade. The logistic regression method was used to build the prediction model, and diagnostic performance was analyzed by plotting receiver operating characteristic (ROC) curve and calculating area under the curve (AUC), sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV).
Out of 145 included patients, 108 constituted the training group and 37 the validation group. The AUC value of the radiomics prediction model to diagnose the pathological grade of BCa was 0.950 (95% confidence interval [CI] 0.912-0.988) in the training group and 0.860 (95% CI 0.742-0.979) in the validation group, respectively. In the validation group, the diagnostic accuracy, sensitivity, specificity, PPV, and NPV were 83.8%, 88.5%, 72.7%, 88.5%, and 72.7%, respectively.
CT-based radiomics model can differentiate high-grade from low-grade BCa with a fairly good diagnostic performance.
•CT-based radiomics model can predict the pathological grade of bladder cancer. •This model has good diagnostic performance to differentiate high-grade and low-grade bladder cancer. •This preoperative and non-invasive prediction method might become an important addition to biopsy.
初步建立基于 CT 的影像组学模型预测膀胱癌（BCa）的病理分级。
回顾性分析 2014 年 10 月至 2017 年 9 月在我院接受手术切除和病理证实的 BCa 且接受 CT 尿路造影（CTU）的患者，将患者随机分为训练组和验证组。在特征提取后，计算特征之间的线性相关系数以消除共线性。然后使用 F 检验来识别与病理分级相关的最佳特征。使用逻辑回归方法建立预测模型，通过绘制受试者工作特征（ROC）曲线并计算曲线下面积（AUC）、敏感度、特异度、阳性预测值（PPV）和阴性预测值（NPV）来分析诊断性能。
纳入的 145 例患者中，108 例为训练组，37 例为验证组。训练组中影像组学预测模型诊断 BCa 病理分级的 AUC 值为 0.950（95%置信区间[CI]0.912-0.988），验证组为 0.860（95%CI0.742-0.979）。在验证组中，诊断准确性、敏感度、特异度、PPV 和 NPV 分别为 83.8%、88.5%、72.7%、88.5%和 72.7%。
基于 CT 的影像组学模型可区分高级别和低级别 BCa，具有较好的诊断性能。
•基于 CT 的影像组学模型可预测膀胱癌的病理分级。•该模型具有良好的诊断性能，可区分高级别和低级别膀胱癌。•这种术前、非侵入性的预测方法可能成为活检的重要补充。
