Zhang Fengyan, Bai Jingjing, Liu Botao, Yuan Miao, Fang Changxing, Yang Guoqiang, Qiao Ying
Department of Radiology, First Hospital of Shanxi Medical University, Taiyuan, Shanxi, China.
College of Medical Imaging, Shanxi Medical University, Taiyuan, Shanxi, China.
Cancer Biomark. 2025 Apr;42(4):18758592251322028. doi: 10.1177/18758592251322028. Epub 2025 Apr 28.
ObjectiveThis study aimed to develop and validate a radiomics nomogram based on 40 KeV images and iodine density maps derived from dual-layer spectral detector CT (DLSDCT) for predicting cervical lymph node (LN) metastasis in patients with papillary thyroid carcinoma (PTC).MethodsA total of 214 LNs from 143 patients with histopathologically confirmed PTC in our hospital were included in the study. The LNs were randomly divided into a training group (n = 150) and a validation group (n = 64) in a 7:3 ratio. Radiomics features were extracted from non-enhanced, arterial phase, and venous phase 40 KeV images, as well as arterial phase and venous phase iodine density maps. Recursive feature elimination (RFE) and logistic regression (LR) were used for feature selection and radiomics score construction. A multivariate logistic regression model was established, incorporating the radiomics score and CT image features. The receiver operating characteristic (ROC) curve was used to evaluate the model's performance. The Hosmer-Lemeshow test and calibration curve were used to assess the model's goodness of fit, while decision curve analysis (DCA) evaluated its clinical applicability.ResultsThe radiomics features consisted of 11 LN-related features that exhibited a good predictive effect. The radiomics nomogram, which included radiomics features, lymphatic hilum status, and significant enhancement in the arterial phase, demonstrated excellent calibration and discrimination in both the training set (AUC = 0.955; 95% confidence interval [CI]: 0.924-0.985) and the validation set (AUC = 0.928; 95% CI: 0.861-0.994). The decision curve analysis confirmed the clinical validity of our nomogram. The DeLong test comparing the radiomics-clinical nomogram with the clinical model yielded a -value of <0.001.ConclusionsThe radiomics nomogram, incorporating radiomics features and CT image features, serves as a non-invasive preoperative prediction tool with high accuracy in predicting cervical lymph node metastasis in patients with PTC.
本研究旨在开发并验证一种基于双层光谱探测器CT（DLSDCT）的40 keV图像和碘密度图的影像组学列线图，用于预测甲状腺乳头状癌（PTC）患者的颈部淋巴结（LN）转移。
本研究纳入了我院143例经组织病理学确诊为PTC患者的214个淋巴结。这些淋巴结以7:3的比例随机分为训练组（n = 150）和验证组（n = 64）。从非增强、动脉期和静脉期的40 keV图像以及动脉期和静脉期碘密度图中提取影像组学特征。采用递归特征消除（RFE）和逻辑回归（LR）进行特征选择和影像组学评分构建。建立了一个多因素逻辑回归模型，纳入影像组学评分和CT图像特征。采用受试者工作特征（ROC）曲线评估模型性能。使用Hosmer-Lemeshow检验和校准曲线评估模型的拟合优度，而决策曲线分析（DCA）评估其临床适用性。
影像组学特征由11个与淋巴结相关的特征组成，具有良好的预测效果。包含影像组学特征、淋巴门状态和动脉期显著强化的影像组学列线图在训练集（AUC = 0.955；95%置信区间[CI]：0.924 - 0.985）和验证集（AUC = 0.928；95% CI：0.861 - 0.994）中均表现出出色的校准和区分能力。决策曲线分析证实了我们列线图的临床有效性。将影像组学 - 临床列线图与临床模型进行比较的DeLong检验得出的P值<0.001。
结合影像组学特征和CT图像特征的影像组学列线图可作为一种非侵入性的术前预测工具，在预测PTC患者颈部淋巴结转移方面具有较高的准确性。
