Department of Radiology, Asan Medical Center, University of Ulsan College of Medicine, Seoul, Republic of Korea.
Department of Radiology, Asan Medical Center, University of Ulsan College of Medicine, Seoul, Republic of Korea.
Clin Radiol. 2021 Aug;76(8):627.e13-627.e21. doi: 10.1016/j.crad.2021.03.001. Epub 2021 Mar 21.
To develop and validate a radiomics-based model for predicting response to neoadjuvant chemotherapy (NAC) using baseline computed tomography (CT) images in patients with muscle-invasive bladder cancer (MIBC).
A radiomics signature for predicting pathological complete response (pCR) was developed using radiomics features selected by a random forest classifier on baseline CT images, and imaging predictors were identified in the training set (87 patients). By incorporating imaging predictors and radiomics signature, an imaging-based model was constructed using multivariate logistic regression analysis and validated in an independent validation set consisting of 48 patients with CT from outside institutions. The performance and clinical usefulness of the imaging-based model for predicting pCR were evaluated using area under the receiver operating characteristic curve (AUC) and decision curve analysis. Using a cut-off determined in the training set, the positive likelihood ratios of the imaging-based model were calculated and compared with imaging and histological predictors.
The radiomics signature was developed based on six stable radiomics features. An imaging-based model incorporating radiomics signature, tumour shape, tumour size, and clinical stage showed good performance for predicting pCR in both the training (AUC, 0.85; 95% confidence interval [CI], 0.78-0.93) and validation (AUC, 0.75; 95% CI, 0.60-0.86) sets, providing a larger net benefit in decision curve analysis. The imaging-based model showed a higher positive likelihood ratio (1.91) for pCR than imaging and histological predictors (1.33-1.63).
The radiomics-based model using baseline CT images may predict the response of patients with MIBC to NAC.
利用基线计算机断层扫描（CT）图像，开发并验证一种基于放射组学的模型，以预测肌层浸润性膀胱癌（MIBC）患者新辅助化疗（NAC）的反应。
利用随机森林分类器从基线 CT 图像中选择放射组学特征，建立预测病理完全缓解（pCR）的放射组学特征模型，并在训练集（87 例患者）中识别出影像学预测因子。通过结合影像学预测因子和放射组学特征，利用多元逻辑回归分析构建基于影像学的模型，并在由来自外部机构的 CT 组成的独立验证集中进行验证。通过接受者操作特征曲线（ROC）下面积（AUC）和决策曲线分析来评估基于影像学的模型预测 pCR 的性能和临床实用性。使用在训练集中确定的截断值，计算基于影像学的模型的阳性似然比，并与影像学和组织学预测因子进行比较。
基于六个稳定的放射组学特征建立了放射组学特征模型。基于影像学的模型，将放射组学特征、肿瘤形态、肿瘤大小和临床分期相结合，在训练集（AUC：0.85；95%置信区间[CI]：0.78-0.93）和验证集（AUC：0.75；95%CI：0.60-0.86）中均具有良好的预测 pCR 的性能，在决策曲线分析中提供了更大的净获益。基于影像学的模型的 pCR 阳性似然比（1.91）高于影像学和组织学预测因子（1.33-1.63）。
利用基线 CT 图像的放射组学模型可预测 MIBC 患者对 NAC 的反应。
