From the Center for Biomedical Image Computing and Analytics and Department of Radiology, University of Pennsylvania, 3710 Hamilton Walk, Philadelphia, PA 19104 (S.R., M.B., A.A.); School of Artificial Intelligence, Guilin University of Electronic Technology, Guangxi, China (A.C.); Comsats University Islamabad, Lahore Campus, Lahore, Pakistan (M.A.I.); and University Hospital of Old Age Psychiatry and Psychotherapy, University of Bern, Bern, Switzerland (A.A.).
Radiol Imaging Cancer. 2021 Jul;3(4):e200108. doi: 10.1148/rycan.2021200108.
Purpose To test the hypothesis that combined features from MR and digital histopathologic images more accurately predict overall survival (OS) in patients with glioma compared with MRI or histopathologic features alone. Materials and Methods Multiparametric MR and histopathologic images in patients with a diagnosis of glioma (high- or low-grade glioma [HGG or LGG]) were obtained from The Cancer Imaging Archive (original images acquired 1983-2008). An extensive set of engineered features such as intensity, histogram, and texture were extracted from delineated tumor regions in MR and histopathologic images. Cox proportional hazard regression and support vector machine classification (SVC) models were applied to MRI features only (MRI/svc), histopathologic features only (HistoPath/svc), and combined MRI and histopathologic features (MRI+HistoPath/svc) and evaluated in a split train-test configuration. Results A total of 171 patients (mean age, 51 years ± 15; 91 men) were included with HGG ( = 75) and LGG ( = 96). Median OS was 467 days (range, 3-4752 days) for all patients, 350 days (range, 15-1561 days) for HGG, and 595 days (range, 3-4752 days) for LGG. The MRI+HistoPath model demonstrated higher concordance index (C-index) compared with MRI and HistoPath models on all patients (C-index, 0.79 vs 0.70 [ = .02; MRI] and 0.67 [ = .01; HistoPath]), patients with HGG (C-index, 0.78 vs 0.68 [ = .03; MRI] and 0.64 [ = .01; HistoPath]), and patients with LGG (C-index, 0.88 vs 0.62 [P = .008; MRI] and 0.62 [P = .006; HistoPath]). In binary classification, the MRI+HistoPath model (area under the receiver operating characteristic curve [AUC], 0.86 [95% CI: 0.80, 0.95]) had higher performance than the MRI model (AUC, 0.68 [95% CI: 0.50, 0.81]; = .01) and the HistoPath model (AUC, 0.72 [95% CI: 0.60, 0.85]; = .04). Conclusion The model combining features from MR and histopathologic images had higher accuracy in predicting OS compared with the models with MR or histopathologic images alone. Survival Prediction, Gliomas, Digital Pathology Imaging, MR Imaging, Machine Learning
验证假设，即与 MRI 或组织病理学特征相比，MR 和数字组织病理学图像的组合特征更能准确预测胶质瘤患者的总生存期 (OS)。
从癌症影像学档案（原始图像采集于 1983 年至 2008 年）中获取诊断为胶质瘤（高级或低级胶质瘤 [HGG 或 LGG]）患者的多参数 MR 和组织病理学图像。从 MR 和组织病理学图像中勾画的肿瘤区域提取了大量的工程特征，如强度、直方图和纹理。Cox 比例风险回归和支持向量机分类（SVC）模型分别应用于 MRI 特征（MRI/svc）、组织病理学特征（HistoPath/svc）和 MRI 和组织病理学特征的组合（MRI+HistoPath/svc），并在分割训练-测试配置中进行了评估。
共纳入 171 例患者（平均年龄 51 岁±15；91 例男性），包括 HGG（ = 75 例）和 LGG（ = 96 例）。所有患者的中位 OS 为 467 天（范围：3-4752 天），HGG 患者为 350 天（范围：15-1561 天），LGG 患者为 595 天（范围：3-4752 天）。MRI+HistoPath 模型在所有患者（C 指数，0.79 比 0.70 [ =.02；MRI]和 0.67 [ =.01；HistoPath]）、HGG 患者（C 指数，0.78 比 0.68 [ =.03；MRI]和 0.64 [ =.01；HistoPath]）和 LGG 患者（C 指数，0.88 比 0.62 [P =.008；MRI]和 0.62 [P =.006；HistoPath]）上的一致性指数（C-index）均高于 MRI 和 HistoPath 模型。在二分类中，MRI+HistoPath 模型（受试者工作特征曲线下面积 [AUC]，0.86 [95%CI：0.80，0.95]）的性能优于 MRI 模型（AUC，0.68 [95%CI：0.50，0.81]； =.01）和 HistoPath 模型（AUC，0.72 [95%CI：0.60，0.85]； =.04）。
与仅使用 MRI 或组织病理学图像的模型相比，结合 MR 和组织病理学图像特征的模型在预测 OS 方面具有更高的准确性。
生存预测、神经胶质瘤、数字病理学成像、磁共振成像、机器学习
