基于CT影像组学列线图鉴别甲状腺乳头状癌与结节性甲状腺肿的价值

The Value of Diagnostic Nomogram Based on CT Radiomics for the Preoperative Differentiation between Thyroid Papillary Carcinoma and Nodular Goiter

目的探讨基于CT影像组学列线图鉴别诊断甲状腺乳头状癌(PTC)与结节性甲状腺肿(NG)的临床价值。方法回顾性分析经病理证实的甲状腺结节患者的临床资料及CT图像。术前2周内行CT平扫及增强扫描;PTC 113例,NG 119例;以7∶3的比例随机分层抽样划分成训练集(n=162)和测试集(n=70)。从CT平扫及双期增强图像中提取甲状腺结节相关征象和影像组学特征,通过临床影像征象和影像组学特征筛选,训练集与测试集均构建4个独立模型并计算影像组学评分,基于影像组学评分和临床模型构建联合模型,并基于联合模型绘制列线图。通过受试者工作特征曲线、曲线下面积(AUC)、连续净重分类改善度(NRI)及综合判别改善度(IDI)等多个指标评估各模型对PTC与NG的鉴别诊断效能,利用校准曲线直观的评估列线图的可靠性与准确性,使用决策曲线评估列线图的临床实用价值。结果平扫、动脉期、静脉期分别保留了6、4、3个影像组学特征,分别构建单期相影像组学模型。单期相中,平扫模型鉴别诊断效能最优,6个独立预测因子用于构建临床模型,基于联合模型的列线图对PTC与NG具有最高的鉴别诊断效能(训练集:AUC为0.980;测试集:AUC为0.928)。除测试集临床模型外,列线图与其余各独立模型间AUC值差异均具有统计学意义(P<0.05),训练集和测试集列线图较平扫模型和临床模型均具有正向改善力(NRI>0,IDI>0)。校准曲线显示列线图预测结果与病理结果间有较好的一致性,决策曲线表明列线图具有更高的临床实用价值。结论基于甲状腺临床影像征象联合影像组学模型构建的列线图对于鉴别诊断PTC与NG具有较高的临床价值。

Objective To explore the classification performance of nomogram constructed from CT signs combined with radiomics for thyroid papillary carcinoma(PTC)and nodular goiter(NG).Methods The clinical data and CT images of patients with thyroid nodules confirmed by pathology in our hospital from January 2019 to April 2022 were analyzed retro⁃spectively.CT plain scan and enhanced scan were performed within 2 weeks before operation;PTC 113 cases,NG 119 ca⁃ses;The patients were divided into training cohort(162 cases)and test cohort(70 cases)at a ratio of 7∶3 using random stratified sampling.Relevant signs and iconography features of thyroid nodules were extracted from CT plain scan and two⁃phase enhanced images.Through screening of clinical image signs and iconography features,four independent models were constructed in the training set and the test set,and calculating the radiomics score.The combined model was constructed based on the radiomics score and the clinical model,and to draw the nomogram based on the combined model.The perform⁃ance of models in the differential diagnosis of PTC and NG was evaluated by multiple indicators,such as receiver operating characteristic curve(ROC),area under the curve(AUC),net reclassification improvement(NRI),integrated discrimination improvement(IDI),and so on.Calibration plots were formulated to evaluate the reliability and accuracy of the nomograms,and the clinical value of nomograms were evaluated by using decision curves.Results 6,4 and 3 features were retained to construct plain‐scanned model,arterial⁃phase and venous⁃phase models respectively.In single phase,The plain⁃scanned model has the best differential diagnosis performance.Six independent predictors were used to construct the clinical models.Nomograms based on combined models had the highest differential diagnostic efficacy for PTC and NG(training set:AUC was 0.980;test set:AUC was 0.928).Except for the clinical models in the test set,the differences of AUC values between the nomograms and other independent models were significant statistically(P<0.05).The nomogram of training set and test set has positive improvement compared with plain scan model and clinical model(NRI>0,IDI>0).The calibration curves showed that the predicted results of nomograms were in good agreement with the pathological results,and the decision curves showed that nomograms had the higher clinical value.Conclusion The nomogram of CT signs combined with radiomics might be a potential method for distinguishing PTC and NG with good performance.