基于MRI的口咽癌颈部淋巴结转移深度学习影像组学模型的构建

Construction of deep learning radiomics model for cervical lymph node metastasis of oropharyngeal carcinoma based on MRI

摘要:

目的 旨在构建基于MRI的CET1及T2WI序列口咽癌颈部淋巴结转移的深度学习影像组学（DLR）模型，并评估其临床应用价值。方法 本研究纳入2019年7月—2025年10月湖南省肿瘤医院头颈外科及中南大学湘雅医院耳鼻咽喉头颈外科手术确诊的口咽癌患者123例，并按7∶3的比例分为训练集和测试集。采用 Pyradiomics和ResNet101算法提取所有患者CET1及T2WI序列的影像组学（Rad）特征和深度学习特征。通过Spearman相关性系数和最小绝对收缩和选择算子筛选具有价值的特征，并构建Rad和DLR模型。采用曲线下面积（AUC）和决策曲线分析对模型的有效性进行评价，DeLong检验用于比较Rad及DLR模型AUC曲线的差异。结果 DLR模型在训练集和测试集中均表现出最佳的性能，在训练集中AUC为0.993(95%CI:0.982~ 1.000)，测试集中AUC为 0.934(95%CI:0.849~1.000)。结论 基于CET1及T2WI的DLR模型在术前预测口咽癌淋巴结转移方面具有较好的性能，可以为临床医生制订更精准的个体化治疗策略提供帮助。

Abstract:

Objective To construct a deep learning radiomics model based on magnetic resonance imaging (MRI) CET1 and T2WI sequences for predicting cervical lymph node metastasis in oropharyngeal carcinoma.Methods This study included 123 patients diagnosed with oropharyngeal cancer through surgery at the Department of Head and Neck Surgery of Hunan Cancer Hospital and the Department of Otorhinolaryngology Head and Neck of Xiangya Hospital, Central South University from July 2019 to October 2025. They were divided into a training set and a test set in a 7∶3 ratio. The Pyradiomics and ResNet101 algorithms were used to extract radiomics features and deep learning features from the MRI CET1 and T2WI sequences of all patients. Valuable features were screened using the Spearman correlation coefficient and the least absolute shrinkage and selection operator (LASSO) to construct a radiomics (Rad) model and a deep learning radiomics (DLR) model. The effectiveness of the models was evaluated using the area under the curve (AUC) and decision curve analysis (DCA). The DeLong test was used to compare the significance of differences in the AUC curves between the Rad and DLR models.Results The DLR model demonstrated the best performance in both the training set and the test set, with an AUC of 0.993 (95%CI： 0.982~1.000) in the training set and 0.934 (95%CI： 0.849~1.000) in the test set.Conclusions The DLR model based on CET1 and T2WI sequences shows good performance in predicting lymph node metastasis in oropharyngeal carcinoma before surgery, and can assist clinicians in formulating more precise and individualized treatment strategies.