为了在术前预测膀胱癌（BCa）的病理分级，我们构建和评估了基于计算机断层扫描（CT）的深度学习放射学组学图（DLRN）。我们对接受手术切除的688名BCa患者进行了回顾性招募（训练组469人，外部测试组219人）。我们从三相CT图像（包括皮质髓质相[C相]、肾造影相[N相]和排泄相[E相]）提取了手工制作的放射学（HCR）特征和深度学习（DL）特征。我们使用11个机器学习分类器构建了预测模型，并将放射学特征与临床因素相结合，构建了一个DLRN。我们利用曲线下面积（AUC）、校准曲线和决策曲线分析（DCA）来评估模型的性能和临床效用。基于HCR和DL组合特征的支持向量机（SVM）分类器模型是最佳放射学特征，训练组和外部测试组的AUC值分别为0.953和0.943。临床模型在训练组和外部测试组的AUC值分别为0.752和0.745。DLRN在两个数据组上都表现良好（训练组：AUC = 0.961；外部测试组：AUC = 0.947），并优于临床模型和最优放射学特征。所提出的基于CT的DLRN在区分高低分级BCa上表现出良好的诊断能力。© 2023。国际癌症成像学协会（ICIS）。

To construct and assess a computed tomography (CT)-based deep learning radiomics nomogram (DLRN) for predicting the pathological grade of bladder cancer (BCa) preoperatively.We retrospectively enrolled 688 patients with BCa (469 in the training cohort, 219 in the external test cohort) who underwent surgical resection. We extracted handcrafted radiomics (HCR) features and deep learning (DL) features from three-phase CT images (including corticomedullary-phase [C-phase], nephrographic-phase [N-phase] and excretory-phase [E-phase]). We constructed predictive models using 11 machine learning classifiers, and we developed a DLRN by combining the radiomic signature with clinical factors. We assessed performance and clinical utility of the models with reference to the area under the curve (AUC), calibration curve, and decision curve analysis (DCA).The support vector machine (SVM) classifier model based on HCR and DL combined features was the best radiomic signature, with AUC values of 0.953 and 0.943 in the training cohort and the external test cohort, respectively. The AUC values of the clinical model in the training cohort and the external test cohort were 0.752 and 0.745, respectively. DLRN performed well on both data cohorts (training cohort: AUC = 0.961; external test cohort: AUC = 0.947), and outperformed the clinical model and the optimal radiomic signature.The proposed CT-based DLRN showed good diagnostic capability in distinguishing between high and low grade BCa.© 2023. International Cancer Imaging Society (ICIS).