Jiang Mingchen, Sun Yiyao, Yang Chunna, Wang Zekun, Xie Ming, Wang Yan, Zhao Dan, Ding Yuqi, Zhang Yan, Liu Jie, Chen Huanhuan, Jiang Xiran
School of Intelligent Medicine, China Medical University, No. 77 Puhe Road, Shenyang, Liaoning, 110122, People's Republic of China.
Department of Medical Imaging, Cancer Hospital of China Medical University, Liaoning Cancer Hospital and Institute, Shenyang, Liaoning, 110042, People's Republic of China.
Radiol Med. 2025 Feb;130(2):190-201. doi: 10.1007/s11547-024-01934-4. Epub 2024 Nov 22.
Early and accurate identification of the metastatic tumor types of brain metastasis (BM) is essential for appropriate treatment and management.
A total of 450 patients were enrolled from two centers as a primary cohort who carry 764 BMs originated from non-small cell lung cancer (NSCLC, patient = 173, lesion = 187), small cell lung cancer (SCLC, patient = 84, lesion = 196), breast cancer (BC, patient = 119, lesion = 200), and gastrointestinal cancer (GIC, patient = 74, lesion = 181). A third center enrolled 28 patients who carry 67 BMs (NSCLC = 24, SCLC = 22, BC = 10, and GIC = 11) to form an external test cohort. All patients received contrast-enhanced T1-weighted (T1CE) and T2-weighted (T2W) MRI scans at 3.0 T before treatment. Radiomics features were calculated from BM and brain-to-tumor interface (BTI) region in the MRI image and screened using least absolute shrinkage and selection operator (LASSO) to construct the radiomics signature (RS). Volume of peritumor edema (VPE) was calculated and combined with RS to create a joint model. Performance of the models was assessed by receiver operating characteristic (ROC).
The BTI-based RS showed better performance compared to BM-based RS. The combined models integrating BTI features and VPE can improve identification performance in AUCs in the training (LC/NLC vs. SCLC/NSCLC vs. BC/GIC, 0.803 vs. 0.949 vs. 0.918), internal validation (LC/NLC vs. SCLC/NSCLC vs. BC/GIC, 0.717 vs. 0.854 vs. 0.840), and external test (LC/NLC vs. SCLC/NSCLC vs. BC/GIC, 0.744 vs. 0.839 vs. 0.800) cohorts.
This study indicated that BTI-based radiomics features and VPE are associated with the metastatic tumor types of BM.
早期准确识别脑转移瘤（BM）的转移瘤类型对于恰当的治疗和管理至关重要。
从两个中心招募了450例患者作为原发性队列，这些患者携带764个源自非小细胞肺癌（NSCLC，患者 = 173例，病灶 = 187个）、小细胞肺癌（SCLC，患者 = 84例，病灶 = 196个）、乳腺癌（BC，患者 = 119例，病灶 = 200个）和胃肠道癌（GIC，患者 = 74例，病灶 = 181个）的脑转移瘤。第三个中心招募了28例携带67个脑转移瘤（NSCLC = 24个，SCLC = 22个，BC = 10个，GIC = 11个）的患者组成外部测试队列。所有患者在治疗前接受3.0T的对比增强T1加权（T1CE）和T2加权（T2W）MRI扫描。从MRI图像中的脑转移瘤和脑-肿瘤界面（BTI）区域计算影像组学特征，并使用最小绝对收缩和选择算子（LASSO）进行筛选以构建影像组学特征（RS）。计算瘤周水肿体积（VPE）并与RS相结合以创建联合模型。通过受试者操作特征（ROC）评估模型的性能。
基于BTI的RS表现优于基于BM的RS。整合BTI特征和VPE的联合模型在训练队列（LC/NLC对SCLC/NSCLC对BC/GIC，0.803对0.949对0.918）、内部验证队列（LC/NLC对SCLC/NSCLC对BC/GIC，0.717对0.854对0.840）和外部测试队列（LC/NLC对SCLC/NSCLC对BC/GIC，0.744对0.839对0.800）的AUC中可提高识别性能。
本研究表明基于BTI的影像组学特征和VPE与脑转移瘤的转移瘤类型相关。
