基于CT影像组学列线图鉴别甲状腺良性与恶性滤泡性肿瘤的价值

The value of diagnostic nomogram based on CT radiomics for the preoperative differentiation between benign and malignant thyroid follicular neoplasms

目的探讨基于CT影像组学列线图鉴别甲状腺滤泡性肿瘤良生与恶性的价值。方法回顾性收集2016年1月至2018年12月复旦大学附属肿瘤医院经手术病理证实的200例甲状腺滤泡性肿瘤患者的临床资料及CT图像,其中甲状腺滤泡癌(FTC)46例、甲状腺滤泡腺瘤(FTA)154例。采用随机数表法随机分为训练集(n=140)和验证集(n=60)。采用LIFEx软件提取增强CT图像的48个影像组学特征。根据最小绝对收缩和选择算子回归进行特征降维、筛选及模型建立,在此基础上绘制列线图。利用受试者操作特征曲线及曲线下面积(AUC)评估列线图预测甲状腺良性与恶性滤泡性肿瘤的效能,通过校准曲线对列线图进行内部及外部验证,最后应用决策曲线分析评估列线图的临床应用价值。结果经筛选得到4种特征用于建立预测甲状腺良性与恶性滤泡性肿瘤的列线图,分别为灰度区域矩阵(GLZLM)-区域长度不均匀性、GLZLM-低灰度区域因子、传统指数-HU单元Q3值、传统指数-HU单元平均值。列线图在训练集中区分FTC和FTA的AUC为0.863(95%CI 0.746~0.932),准确度为87.9%,灵敏度为67.9%,特异度为91.1%;验证集中AUC为0.792(95%CI 0.658~0.917),准确度为75.0%,灵敏度为66.7%,特异度为90.5%。校正曲线结果显示列线图预测值与病理结果之间具有良好的一致性,决策曲线分析表明列线图在临床上具有良好的应用价值。结论CT影像组学模型对于鉴别甲状腺良性与恶性滤泡性肿瘤具有较好的效能,基于此的列线图可准确、直观地预测甲状腺滤泡性肿瘤患者的恶性概率。

Objective To investigate the value of nomogram constructed by CT-based radiomics for differentiating benign and malignant thyroid follicular neoplasms.Methods Totally 200 post-surgery patients with pathologically confirmed thyroid follicular neoplasms in Fudan University Shanghai Cancer Center from January 2016 to December 2018 were retrospectively analyzed.Among the patients,46 were follicular thyroid carcinoma(FTC)and 154 patients were follicular thyroid adenoma(FTA).The patients were randomly divided into a training set(n=140)and validation set(n=60)using a random number table.CT signs and radiomics features of each patient were analyzed within the LIFEx package.A predictive model was developed by the least absolute shrinkage and selection operator regression to build a nomogram based on selected parameters.The predictive effectiveness of differentiating benign and malignant thyroid follicular neoplasms was evaluated by the area under receiver operating characteristic curve(AUC).Calibration plots were formulated to evaluate the reliability and accuracy of the nomogram based on internal(training set)and external(validation set)validity.The clinical value of the nomogram was estimated through the decision curve analysis.Results The prediction nomogram was built with 4 selected parameters,including grey level zone length matrix(GLZLM)-gray-level zone length matrix_zone length non-uniformity,GLZLM-gray-level zone length matrix_low gray-level zone emphasis,CONVENTIONAL_HUQ3,CONVENTIONAL_HUmean.In training and validation sets,the AUCs for differentiating FTC and FTA were 0.863(95%CI 0.746-0.932),0.792(95%CI 0.658-0.917),accuracy were 87.9%and 75.0%,sensitivity were 67.9%and 66.7%,specificity were 91.1%and 90.5%,respectively.The calibration curves indicated good consistency between actual observation and prediction for differentiating the malignancy.Decision curve analysis demonstrated the nomogram was clinically useful.Conclusions The CT radiomics mode shows the certain value and great potential to identify benign or malignant thyroid follicular neoplasms and the nomogram can accurately and intuitively predict the malignancy potential in patients with thyroid follicular neoplasms.