文献速递头颈甲状腺与浅表技术与安全CT

基于增强CT栖息地成像预测甲状腺乳头状癌隐匿性中央区淋巴结转移

Habitat imaging based on enhanced CT for predicting occult central lymph node metastasis in papillary thyroid carcinoma.
PubMed 文献 · Eur Radiol · 2026-09-19

本研究回顾性分析两个医学中心219例甲状腺乳头状癌患者的术前增强CT图像与临床参数,提取动脉期和静脉期肿瘤内栖息地及3mm瘤周放射组学特征。采用五种机器学习方法建模并整合临床因素构建联合模型,训练队列AUC达0.932,外部验证队列AUC达0.878。该联合模型可为PTC患者术前风险分层提供无创辅助信息。

要点

  • 动脉期与静脉期CT肿瘤内栖息地及瘤周3mm区域放射组学特征联合临床指标对隐匿性中央区淋巴结转移具有较高预测效能。
  • 联合模型在外部验证队列AUC为0.878,提示跨中心推广性尚可,但需注意回顾性设计的偏倚。
  • 年龄、TSH和PLT被识别为独立预测因子,报告时可结合这些临床参数辅助判断隐匿性转移风险。

核心结论

  • 整合增强CT瘤内栖息地、瘤周放射组学与临床特征的联合模型为PTC患者术前无创预测隐匿性中央区淋巴结转移提供了潜在方法。
  • 联合模型在训练队列和外部验证队列的AUC分别为0.932和0.878,显示较优的预测能力。

研究设计与样本

  • 回顾性研究纳入两个医学中心共219例甲状腺乳头状癌患者,其中训练队列来自中心1共154例,外部验证队列来自中心2共65例。
  • 术前增强CT图像包含动脉期和静脉期,分别从肿瘤内栖息地及3mm瘤周区域提取放射组学特征。
  • 采用五种机器学习方法建模,以AUC评估性能并选择最佳模型,最终整合临床因素构建联合模型。

关键影像表现/征象

  • 基于逻辑回归的肿瘤内栖息地放射组学模型在动脉期训练及外部验证队列AUC分别为0.820和0.805,静脉期分别为0.884和0.850。
  • 3mm瘤周区域放射组学特征与肿瘤内栖息地特征均对预测隐匿性中央区淋巴结转移具有贡献。

诊断标准与关键数值

  • 基于逻辑回归的肿瘤内栖息地模型在静脉期训练队列AUC为0.884,外部验证队列AUC为0.850。
  • 基于逻辑回归的瘤周放射组学模型在动脉期训练队列AUC为0.820,外部验证队列AUC为0.805。
  • 联合模型训练队列AUC为0.932,外部验证队列AUC为0.878。
  • 年龄、TSH和PLT被识别为独立预测因子。

临床应用与阅片要点

  • 日常阅片时对PTC患者增强CT动脉期和静脉期图像中的肿瘤内栖息地及瘤周3mm区域应予以关注,这些区域的放射组学特征与隐匿性中央区淋巴结转移相关。
  • 报告中纳入年龄、TSH和PLT等临床指标有助于提升对隐匿性中央区淋巴结转移的风险判断。
  • 联合模型的预测结果可作为术前风险分层的辅助信息,支持个体化手术方案制定。

局限性

  • 本研究为回顾性设计,存在选择偏倚与信息偏倚的可能。
  • 外部验证队列样本量为65例,规模有限。
  • 机器学习模型的跨中心泛化能力仍需更大样本的多中心前瞻性研究验证。

以上解读依据论文摘要(全文未开放获取或许可不允许翻译转载)。

英文原文(摘要)· 点击展开

OBJECTIVES: This study aimed to develop a combined model based on clinical, intratumoral habitat, and peritumoral radiomic features to predict occult central lymph node metastasis (OCLNM) in patients with papillary thyroid carcinoma (PTC).

MATERIALS AND METHODS: This retrospective study analysed preoperative enhanced CT images and clinical parameters from 219 PTC patients from two medical centers. The patients were divided into a training cohort from Center 1 (n = 154) and an external validation cohort from Center 2 (n = 65). Habitat radiomics features from the tumor and 3-mm peritumoral region were extracted from both arterial phase (AP) and venous phase (VP) CT images, respectively. Five machine learning (ML) methods were employed to construct models. The area under the curve (AUC) was used to evaluate the model performance, thereby selecting the best-performing ML model. Ultimately, a combined model was developed by integrating clinical factors with the optimal AP and VP habitat models.

RESULTS: Age, TSH, and PLT were identified as independent predictors. Both the intratumoral habitat and peritumoral radiomics models based on LR exhibited excellent performance in the training and external validation cohorts (AUC: 0.820 and 0.805 in AP; 0.884 and 0.850 in VP). The combined model demonstrated superior predictive capability, achieving AUCs of 0.932 in the training cohort and 0.878 in the external validation cohort.

CONCLUSION: The combined model integrating intratumoral habitat and peritumoral radiomics from enhanced CT images with clinical features provides a potential, non-invasive preoperative method for predicting OCLNM in PTC patients and may provide supplementary information for preoperative risk stratification.

KEY POINTS: Question Accurate preoperative diagnosis of OCLNM in PTC is crucial for guiding dissection but remains an urgent clinical challenge. Findings A combined model integrating CT-based intratumoral habitats, peritumoral radiomics, and clinical factors achieved superior performance (AUC: 0.878) in predicting occult metastases across two centers. Clinical relevance This non-invasive model may serve as an adjunctive tool for preoperative risk stratification, providing supplementary information to support individualized surgical planning in patients with PTC.

来源:Eur Radiol,查看原文。版权归原作者所有。

引用

  1. Zhao, Duan, Ke, et al. Habitat imaging based on enhanced CT for predicting occult central lymph node metastasis in papillary thyroid carcinoma. Eur Radiol. 2026. PMID: 42763332. doi:10.1007/s00330-026-12806-y

供执业医师工作参考,不替代临床判断;数值与分级以指南原文为准。