| 李洋,杨晓红,黄丽霞.基于多模态超声的 Nomogram模型术前预测甲状腺结节性质的临床研究[J].安徽医药,2025,29(7):1353-1358. |
| 基于多模态超声的 Nomogram模型术前预测甲状腺结节性质的临床研究 |
| A clinical study of preoperative prediction of thyroid nodule properties by Nomogram model based on multimodal ultrasound |
| |
| DOI：10.3969/j.issn.1009-6469.2025.07.017 |
| 中文关键词: 甲状腺结节 肿瘤分期 肿瘤分级 组织学类型肿瘤 多模态超声 Nomogram模型 弹性成像 |
| 英文关键词: Thyroid nodule Neoplasm staging Neoplasm grading Neoplasms by histologic type Multimode ultrasound No-mogram model Elastography |
| 基金项目:保定市科技计划项目( 2441ZF020) |
|
| 摘要点击次数: 833 |
| 全文下载次数: 213 |
| 中文摘要: |
| 目的基于多模态超声术前预测甲状腺结节性质，并构建 Nomogram模型。方法选取 2020年 1月至 2023年 1月于保定市第一医院就诊并行甲状腺结节切除术的病人 225例。经计算机生成随机数字表法按 2∶1分为建模集( 150例)、检验集( 75例)。病人均于术前行多模态超声检查，建模集根据术后病理结果将其分为良性组(125例， 151个结节)、恶性组( 25例， 35个结节)。比较建模集两组常规超声、超声造影( CEUS)及弹性成像特征；采用多因素逐步 logistic回归模型分析建模集多模态超声参数与甲状腺结节性质的相关性；建立建模集术前预测甲状腺结节性质的 Nomogram模型，用受试者操作特征曲线( ROC曲线)、 Calibration曲线验证模型的效能。结果与良性组［ 67.55%、31.13%、33.11%、34.44%、51.66%、9.93%、47.68%、0.00%、 25.83%、39.07%、9.93%、(27.76±4.10)s、(1.95±0.24)分、(33.69±5.47)%、(56.58±12.24)s、(1 956.32±412.27)%s］比较，恶性组极低或低回声、边缘不规则或分叶状、纵横比 >1、点状强回声、造影剂分布不均匀、向心性增强、低增强、增强后边界模糊无法分辨、造影后结节体积增大、晚于周边甲状腺实质同步增强、早于周边甲状腺实质同步消退构成比、达峰时间( TP)、弹性成像评分［ 97.14%、74.29%、51.43%、57.14%、91.43%、40.00%、88.57%、5.71%、54.29%、65.71%、28.57%、(37.20±5.62)s、(3.36±0.51)分］升高( P<0.05)，结节及周围正常组织的峰值强度( PI)、平均渡越时间( MTT)、时间 -强度曲线下面积( AUCt)［( 22.48±6.95)%、(45.30±15.19)s、(1 127.18±321.07)%s］降低( P<0.05)。多因素逐步 logistic回归模型显示，极低或低回声、造影后结节体积增大、 TP、MTT、弹性成像评分与甲状腺结节性质密切相关( P<0.05)。将上述风险因素作为预测变量，建立建模集 Nomogram模型。 ROC曲线显示， Nomogram模型预测建模集术后甲状腺结节恶性的 AUC为 0.95，灵敏度为 91.43%，特异度为 90.07%，检验集的 AUC及其 95%CI为 0.92(0.84，0.96)灵敏度为 83.33%，特异度为 84.00%，建模集、检验集预测建模集术后甲状腺结节恶性的 AUC均大于美国放射学会甲状腺影告与数据系统分级；经 Hosmer-Lemeshow检验，建模集、检验集的 Calibration曲线均像报，差异无统计学意义( χ2=0.46、0.41，P=0.321、0.384)显示校正曲线与理想曲线是拟合的(P>0.05)。采用 Bootstrap法进行内部验证， C-index指数 0.91、0.90，95%CI为( 0.76，0.85)、(0.，84，0.97)区分度良好。结论多模态超声可综合各项敏感指标，提高超声对甲状腺结节性质的诊断效能，据此建立的 Nomogram模型预测，效能良好。 |
| 英文摘要: |
| Objective To preoperatively predict the nature of thyroid nodules using multimodal ultrasound and to construct a Nomo-gram model.Methods A total of 225 patients who underwent thyroid nodule resection at Baoding No. 1 Hospital from January 2020 toJanuary 2023 were enrolled. They were randomly assigned into a modeling set (150 cases) and a validation set (75 cases) in a 2:1 ratiousing a computer-generated random number table. All patients underwent preoperative multimodal ultrasound examinations. Based onpostoperative pathological results, the modeling set was further divided into a benign group (n=125,151 nodules) and a malignant group (n=25,35 nodules). Conventional ultrasound, contrast-enhanced ultrasound (CEUS), and elastography features were compared betweenthe two groups in the modeling set. Multivariate stepwise logistic regression analysis was performed to evaluate the correlation betweenmultimodal ultrasound parameters and thyroid nodule malignancy in the modeling set. A Nomogram model for preoperative predictionof thyroid nodule malignancy was established, and its performance was validated using receiver operating characteristic (ROC) curvesand calibration curves.Results In comparison with the benign group [67.55%, 31.13%, 33.11%, 34.44%, 51.66%, 9.93%, 47.68%,0.00%, 25.83%, 39.07%, 9.93%, (27.76±4.10) s, (1.95±0.24) points, (33.69± 5.47) %, (56.58±12.24) s, (1956.32±412.27) %s], the ma-lignant group showed significantly higher proportions of very low or low echogenicity, irregular or lobulated margins, aspect ratio >1, punctate echogenic foci, heterogeneous contrast distribution, centripetal enhancement, hypo-enhancement, indistinct post-enhancement boundaries, post-contrast nodule volume increase, delayed enhancement synchronization compared to surrounding thyroid parenchyma,and early washout synchronization compared to surrounding thyroid parenchmia, along with prolonged time to peak (TP) and elevatedelastography scores [97.14%, 74.29%, 51.43%, 57.14%, 91.43%, 40.00%, 88.57%, 5.71%, 54.29%, 65.71%, 28.57%, (37.20±5.62) s,(3.36±0.51) points] (P<0.05). Conversely, peak intensity (PI), mean transit time (MTT), and time-intensity curve area under the curve(AUCt) of nodules and surrounding normal tissue [(22.48±6.95) % , (45.30±15.19) s, (1127.18±321.07) % s] were reduced (P<0.05). Multivariate stepwise logistic regression identified very low/low echogenicity, post-contrast nodule volume increase, TP, MTT, and elas-tography scores as significant predictors of thyroid nodule malignancy in the modeling set (P<0.05). A Nomogram model incorporatingthese predictors demonstrated excellent performance: ROC analysis revealed an AUC of 0.95 with 91.43% sensitivity and 90.07% spec-ificity in the modeling set, and 0.92 AUC [95% CI:(0.84,0.96)] with 83.33% sensitivity and 84.00% specificity in the validation set, out-performing the American College of Radiology Thyroid Imaging Reporting and Data System (ACR TI-RADS) classification. Hosmer-Lemeshow tests indicated good calibration in both sets (χ2=0.46, P=0.321, χ2=0.41, P=0.384), the calibration curve showing good agreement with the ideal curve (P>0.05). Bootstrap validation showed robust discrimination with C-indices of 0.91 [95% CI:(0.76,0.85)] and 0.90 [95%CI:(0.84,0.97)] for modeling and validation sets, respectively.Conclusion Multimodal ultrasound integration of sensi-tive parameters enhances diagnostic accuracy for thyroid nodule characterization. The established Nomogram model demonstrates supe-rior predictive performance for malignancy assessment. |
|
查看全文
查看/发表评论 下载PDF阅读器 |
| 关闭 |
|
|
|
