Liu Dongming, Chen Jiu, Ge Honglin, Hu Xinhua, Yang Kun, Liu Yong, Hu Guanjie, Luo Bei, Yan Zhen, Song Kun, Xiao Chaoyong, Zou Yuanjie, Zhang Wenbin, Liu Hongyi
Department of Neurosurgery, The Affiliated Brain Hospital of Nanjing Medical University, Nanjing, China.
Institute of Neuropsychiatry, The Affiliated Brain Hospital of Nanjing Medical University, Fourth Clinical College of Nanjing Medical University, Nanjing, China.
Front Oncol. 2022 Jul 28;12:848846. doi: 10.3389/fonc.2022.848846. eCollection 2022.
Tumor infiltration of central nervous system (CNS) malignant tumors may extend beyond visible contrast enhancement. This study explored tumor habitat characteristics in the intratumoral and peritumoral regions to distinguish common malignant brain tumors such as glioblastoma, primary central nervous system lymphoma, and brain metastases. The preoperative MRI data of 200 patients with solitary malignant brain tumors were included from two datasets for training. Quantitative radiomic features from the intratumoral and peritumoral regions were extracted for model training. The performance of the model was evaluated using data ( = 50) from the third clinical center. When combining the intratumoral and peritumoral features, the Adaboost model achieved the best area under the curve (AUC) of 0.91 and accuracy of 76.9% in the test cohort. Based on the optimal features and classifier, the model in the binary classification diagnosis achieves AUC of 0.98 (glioblastoma and lymphoma), 0.86 (lymphoma and metastases), and 0.70 (glioblastoma and metastases) in the test cohort, respectively. In conclusion, quantitative features from non-enhanced peritumoral regions (especially features from the 10-mm margin around the tumor) can provide additional information for the characterization of regional tumoral heterogeneity, which may offer potential value for future individualized assessment of patients with CNS tumors.
中枢神经系统(CNS)恶性肿瘤的肿瘤浸润可能超出可见的对比增强范围。本研究探讨了肿瘤内和肿瘤周围区域的肿瘤生境特征,以区分常见的恶性脑肿瘤,如胶质母细胞瘤、原发性中枢神经系统淋巴瘤和脑转移瘤。纳入了来自两个数据集的200例孤立性恶性脑肿瘤患者的术前MRI数据用于训练。从肿瘤内和肿瘤周围区域提取定量放射组学特征用于模型训练。使用来自第三个临床中心的数据(n = 50)评估模型的性能。当结合肿瘤内和肿瘤周围特征时,Adaboost模型在测试队列中达到了最佳曲线下面积(AUC)为0.91,准确率为76.9%。基于最佳特征和分类器,二元分类诊断中的模型在测试队列中分别实现了0.98(胶质母细胞瘤和淋巴瘤)、0.86(淋巴瘤和转移瘤)和0.70(胶质母细胞瘤和转移瘤)的AUC。总之,肿瘤周围非增强区域的定量特征(尤其是肿瘤周围10毫米边缘的特征)可为区域肿瘤异质性的表征提供额外信息,这可能为未来中枢神经系统肿瘤患者的个体化评估提供潜在价值。