Wang Qi-Guo, Li Mei, Deng Guang-Xiu, Huang Hai-Qing, Qiu Qin, Lin Jian-Jun
Department of Medical Ultrasound, the First People's Hospital of Qinzhou, Qinzhou, China.
Department of Medical Ultrasound, the People's Hospital of Chongzuo, Chongzuo, China.
Quant Imaging Med Surg. 2025 May 1;15(5):4641-4654. doi: 10.21037/qims-24-1796. Epub 2025 Apr 28.
Conventional ultrasound (US) has been routinely used for differential diagnosis of thyroid nodules, but its discriminatory performance remains unsatisfactory. This study aimed to develop and validate a prediction nomogram model based on conventional US and contrast-enhanced ultrasound (CEUS) features for differentiating malignant from benign thyroid nodules.
A total of 815 thyroid nodules with surgical pathology results and complete conventional US and CEUS data were retrospectively collected from the First People's Hospital of Qinzhou between January 2019 and July 2023. The nodules were grouped into a training cohort (n=571) and a validation cohort (n=244) at a 7:3 ratio. Independent risk factors of malignancy were selected by stepwise multivariate logistic regression analysis, and a prediction nomogram model was subsequently constructed. The diagnostic performance of the model was evaluated by the area under the receiver operating characteristic curve (AUC) in both the training and validation cohorts. The unnecessary fine-needle aspiration biopsy (FNAB) rate was calculated.
Multivariate logistic regression analysis identified irregular margin, aspect ratio >1, and microcalcification from conventional US images, as well as hypo-enhancement intensity and ring enhancement from CEUS images, as independent predictors for malignancy. The AUC, sensitivity, specificity, and accuracy of the prediction nomogram model were 0.947 [95% confidence interval (CI): 0.928-0.966], 90.4%, 88.8%, and 89.8% in the training cohort, and 0.957 (95% CI: 0.928-0.986), 94.5%, 86.4%, and 91.8% in the validation cohort, respectively. Using the prediction model, the unnecessary FNAB rates reduced from 29.6% to 6.1% in the training cohort and from 29.3% to 6.7% in the validation cohort compared to the Chinese Thyroid Imaging Reporting and Data System. Decision curve analysis demonstrated good clinical utility of the nomogram model.
The prediction nomogram model incorporating conventional US and CEUS features could effectively distinguish between malignant and benign thyroid nodules and reduce unnecessary FNAB rates.
传统超声（US）一直被常规用于甲状腺结节的鉴别诊断，但其鉴别性能仍不尽人意。本研究旨在开发并验证一种基于传统超声和对比增强超声（CEUS）特征的预测列线图模型，用于区分甲状腺恶性结节和良性结节。
回顾性收集了2019年1月至2023年7月期间钦州市第一人民医院的815个具有手术病理结果以及完整传统超声和CEUS数据的甲状腺结节。这些结节按7:3的比例分为训练队列（n = 571）和验证队列（n = 244）。通过逐步多因素逻辑回归分析选择恶性肿瘤的独立危险因素，随后构建预测列线图模型。在训练队列和验证队列中，通过受试者操作特征曲线（AUC）下面积评估模型的诊断性能。计算不必要的细针穿刺活检（FNAB）率。
多因素逻辑回归分析确定，传统超声图像中的边界不规则、纵横比>1和微钙化，以及CEUS图像中的低增强强度和环状增强是恶性肿瘤的独立预测因素。预测列线图模型在训练队列中的AUC、敏感性、特异性和准确性分别为0.947 [95%置信区间（CI）：0.928 - 0.966]、90.4%、88.8%和89.8%，在验证队列中分别为0.957（95% CI：0.928 - 0.986）、94.5%、86.4%和91.8%。与中国甲状腺影像报告和数据系统相比，使用预测模型后，训练队列中的不必要FNAB率从29.6%降至6.1%，验证队列中从29.3%降至6.7%。决策曲线分析表明列线图模型具有良好的临床实用性。
结合传统超声和CEUS特征的预测列线图模型能够有效区分甲状腺恶性结节和良性结节，并降低不必要的FNAB率。
