摘要
目的 探讨基于多模态CT特征联合纹理分析及临床信息的模型在评估甲状腺结节良恶中的应用价值.方法 回顾性分析 134 枚经病理证实的甲状腺结节CT图像及临床信息,比较良、恶性结节的CT形态学特征、能谱CT定量参数(动脉期及静脉期标准化碘浓度、能谱曲线斜率、标准化有效原子序数)、影像组学纹理特征及临床危险因素.分类变量采用多因素Logistic回归分析,连续变量采用非参数检验(Mann-Whitney U检验),基于常规影像学特征、能谱CT定量参数、组学纹理特征、临床危险因素建立多个临床模型,通过绘制受试者工作特征曲线评估模型的诊断效能.结果 CT常规影像学特征中,保留了包膜完整度、平扫咬饼征、钙化、结节长径及动脉增强率用于构建模型,能谱CT参数中动脉期的标准化碘浓度(NIC)、能谱曲线斜一率(λHu)1、λHU2及静脉期中的λHu1、λHu2 有统计学意义(P<0.05).影像组学纹理特征中,最终留下 2 个特征用于构建模型.临床风险因素中,年龄及促甲状腺激素水平(TSH)差异具有统计学意义(P<0.05).在各模型中,联合模型具有最高的鉴别诊断效能,其曲线下面积为 0.958,灵敏度为 0.887,特异度为 0.926.结论 基于多模态CT特征联合纹理分析及临床信息的模型在评估甲状腺良恶性方面有较高的诊断价值.
Abstract
Objective To explore the application value of a model based on multimodal CT features combined with radiomics and clinical information in the evaluation of benign and malignant thyroid nodules.Methods The CT images and clinical information of 134 thyroid nodules confirmed by pathology were retrospectively analyzed.CT morphological features,spectral CT quantitative parameters(normalized iodine concentration,the slope of spectral Hu curve,normalized effective atomic number),radiomics features,and clinical risk factors were compared.Multivariate Logistic regression analysis was used for categorical variables,and a non-parametric test(Mann-Whitney U test)was used for continuous variables.Multiple models were established based on conventional imaging features,spectral CT quantitative parameters,radiomics features,and clinical risk factors.The diagnostic efficacy of the model was evaluated by the receiver operating characteristic.Results Some CT conventional imaging features were constructed of the model,such as the complete capsule,the cookie bite sign in the plain scan,calcification,nodule length,and arterial enhancement rate.The spectral CT parameters of NIC,λ1Hu and λ2Hu in the arterial phase,and λ1Hu and λ2Hu in the venous phase were statistically significant(P<0.05).In the radiomics features,two features were finally left to construct the model.Among the clinical risk factors,age and thyroid stimulating hormone level(TSH)were statistically significant(P<0.05).Among the models,the combined model had the highest differential diagnostic efficiency,with an area under the curve of 0.958,a sensitivity of 0.887,and a specificity of 0.926.Conclusion The model based on multimodal CT features combined with radiomics and clinical information has a high diagnostic value in evaluating benign and malignant thyroid nodules.
维普
万方数据