多参数磁共振组学预测甲状腺乳头状癌淋巴结转移的价值

Radiomics based on multiparametric MRI for preoperative prediction of lymph node metastasis in papillary thyroid carcinoma

目的:探讨基于术前原发肿瘤的多参数磁共振图像所构建的组学模型在预测甲状腺乳头状癌颈部淋巴结转移中的价值。方法:回顾性分析129例病理证实的甲状腺乳头状癌术前磁共振资料。从T_(2)WI、DWI、多期T_(1)WI增强图像中提取影像组学特征,按照7:3比例随机将129例患者为训练组和验证组,在训练集中采用最小冗余最大关联度(mRMR)和最小绝对收缩选择算子(LASSO)筛选最优影像组学特征,构建组学特征模型。采用受试者操作特征(ROC)曲线评价多参数磁共振组学模型术前预测颈部淋巴结转移的预测效能,并在验证集中验证。结果:分别从T_(2)WI、DWI、多期T_(1)WI增强提取396个影像组学特征,经预处理及筛选最终获得18个最佳组学特征,根据最优影像组学特征及相应权重系数建立影像组学预测模型并计算影像组学评分,训练组和验证组中影像组学评分淋巴结转移组高于无淋巴结转移组。训练组的AUC值为0.92,敏感度为83.3%,特异度为88.9%,验证组的AUC值为0.80,敏感度为78.9%,特异度为77.8%。结论:基于多参数磁共振组学模型可术前预测甲状腺乳头状癌颈部淋巴结转移,有望成为一种术前无创评估甲状腺乳头状癌颈部淋巴结转移的方法。

Objective:To evaluate the diagnostic performance of a radiomics model based on preoperative multiparametric MRI in preoperatively predicting lymph node metastasis in patients with papillary thyroid carcinoma(PTC).Methods:Data of 129 patients with histopathologically confirmed PTC were retrospectively reviewed in this study.Radiomics features were extracted from T_(2)-weighted imaging(T_(2)WI),diffusion weighted imaging(DWI)and T_(1)-weighted multiphase contrast enhancement images of primary thyroid tumor,respectively.All patients were randomly assigned to training and validation cohorts at the ratio of 7:3.Minimum redundancy maximum relevance(mRMR)and least absolute shrinkage and selection operator(LASSO)were used to select an optimized features'subset to construct the radiomics signature in training cohort.Receiver operating characteristic(ROC)curve was drawn to evaluate the efficacy of the radiomics signature for predicting performance in the training and validation cohorts,respectively.Results:Eighteen features was selected from the combined images.The Rad-score of patients with lymph node metastasis(LNM)were significantly higher than that of patients with non-LNM in both the training cohort and the validation cohort.According to the analysis of ROC curve,the area under the curve(AUC)of training cohort was 0.92,with the sensitivity of 83.3%and the specificity of 88.9%.The validation cohort also showed a good predictive ability,with the AUC of 0.80,the sensitivity of 78.9%and the specificity of 77.8%.Conclusion:Radiomics model based on multiparametric MRI images could be used to predict LNM in patients with PTC,which may expected to be an non-invasive method to evaluate LNM in PTC.