Zhang Xi-Wei, Wusiman Dilinaer, Zhang Ye, Yu Xiao-Duo, Miao Su-Sheng, Wang Zhi, Liu Shao-Yan, Li Zheng-Jiang, Sun Ying, Yi Jun-Lin, An Chang-Ming
Department of Head and Neck Surgical Oncology National Cancer Center/National Clinical Research Center for Cancer/Cancer Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College Beijing China.
Purdue Institute for Cancer Research, Purdue University West Lafayette Indiana USA.
World J Otorhinolaryngol Head Neck Surg. 2025 Mar 24;11(3):440-448. doi: 10.1002/wjo2.70001. eCollection 2025 Sep.
The aim of this study is to develop a multimodal MRI radiomics-based model for predicting long-term overall survival in hypopharyngeal cancer patients undergoing definitive radiotherapy.
We enrolled 207 hypopharyngeal cancer patients who underwent definitive radiotherapy and had 5-year overall survival outcomes from two major cancer centers in China. Pretreatment MRI images and clinical features were collected. Regions of interest (ROIs) for primary tumors and lymph node metastases (LNM) were delineated on T2 and contrast-enhanced T1 (CE-T1) sequences. Principal component analysis (PCA), support vector machine (SVM), and 5-fold cross-validation were used to develop and evaluate the models.
Multivariate Cox regression analysis identified age under 50 years, advanced T stage, and N stage as risk factors for overall survival. Predictive models based solely on clinical features (Model A), single radiomics features (Model B), and their combination (Model C) performed poorly, with mean AUC values in the validation set of 0.663, 0.772, and 0.779, respectively. The addition of multimodal LNM and CE-T1 radiomics features significantly improved prediction accuracy (Models D and E), with AUC values of 0.831 and 0.837 in the validation set.
We developed a well-discriminating overall survival prediction model based on multimodal MRI radiomics, applicable to patients receiving definitive radiotherapy, which may contribute to personalized treatment strategies.
本研究旨在开发一种基于多模态磁共振成像（MRI）影像组学的模型，用于预测接受根治性放疗的下咽癌患者的长期总生存率。
我们纳入了207例接受根治性放疗且有5年总生存结局的下咽癌患者，这些患者来自中国的两个主要癌症中心。收集了治疗前的MRI图像和临床特征。在T2加权像和对比增强T1加权像（CE-T1）序列上勾画原发肿瘤和淋巴结转移（LNM）的感兴趣区（ROI）。采用主成分分析（PCA）、支持向量机（SVM）和五折交叉验证来开发和评估模型。
多因素Cox回归分析确定年龄小于50岁、T分期晚期和N分期为总生存的危险因素。仅基于临床特征的预测模型（模型A）、单一影像组学特征的预测模型（模型B）及其组合模型（模型C）表现不佳，验证集的平均AUC值分别为0.663、0.772和0.779。添加多模态LNM和CE-T1影像组学特征显著提高了预测准确性（模型D和模型E），验证集的AUC值分别为0.831和0.837。
我们基于多模态MRI影像组学开发了一种具有良好鉴别能力的总生存预测模型，适用于接受根治性放疗的患者，这可能有助于制定个性化治疗策略。
