Abstract: Objective　To establish a thyroid nodule quantification scoring system based on nodular CT signs deduced from the thyroid image reporting and data system，and to explore its diagnostic efficiency for thyroid malignant nodules.Methods　A retrospective study was performed on patients with thyroid nodules who underwent CT examination，and whose diagnoses were confirmed by pathology，at the Fifth Affiliated Hospital of Sun Yat-Sen University.The training set included 85 patients （a total of 134 nodules） who were examined between October 2012 and August 2015；the validation set included 25 patients （a total of 61 nodules） who were examined between September 2015 and June 2016.The differences in 11 CT signs between benign and malignant nodules in the training set were analyzed.The OR values of significantly different CT signs were calculated and scored，and the overall score of each nodule was calculated.The ROC curve for the diagnosis of malignant nodules was plotted，and the best cutoff value was determined.A quantification scoring system was constructed，and its diagnostic value for thyroid nodules was evaluated.Results　Among the 11 CT signs，the number of nodules，nodule shape，nodule boundary，nodule edge，structural composition，calcification type，enhancement mode，thyroid capsule integrity，lymph node metastasis between benign and malignant thyroid nodules，were significantly different（P<0.05）.The AUC of the ROC curve of the scoring system in the diagnosis of malignant nodules was 0.921〔95%CI（0.871，0.971）〕.The best cut off value was 9 points.Pathological examination was used as the gold standard.Categories 1，2，and 3 in the CT quantification scoring system were deemed benign nodules，and categories 4 and 5 were deemed malignant nodules.The sensitivity，specificity，positive predictive value，negative predictive value and accuracy of the scoring system were 86.67%（13/15），89.13% （41/46），72.22% （13/18），95.35%（41/43） and 88.52%（54/61）．The AUC of the ROC curve was 0.872〔95%CI（0.742，1.000）〕.Conclusion　The thyroid nodule quantification scoring system based on nodular CT signs has a high diagnostic efficiency for differentiating benign and malignant thyroid nodules.

Key words: Thyroid nodule；Tomography, spiral computed；Thyroid imaging reporting and data system

摘要： 目的　参照超声检查的甲状腺影像报告与数据系统（TI-RADS），尝试建立以结节CT征象为基础的甲状腺结节CT量化评分分类系统，并探讨其对甲状腺恶性结节的诊断效能。方法　回顾性选择中山大学附属第五医院行CT检查并经病理检查证实的甲状腺结节患者为研究对象，其中2012年10月—2015年8月诊治的85例甲状腺结节患者（共134个结节）为训练组，2015年9月—2016年6月诊治的25例甲状腺结节患者（共61个结节）为验证组。分析训练组甲状腺良、恶性结节11项CT征象的差异，计算差异有统计学意义的CT征象的OR值并赋值评分，算出每个结节赋值评分后绘制其诊断甲状腺恶性结节的受试者工作特征（ROC）曲线，得出最佳诊断界值，构建甲状腺结节CT量化评分分类系统。分析该分类系统对验证组甲状腺结节的分类诊断价值。结果　11项CT征象中，甲状腺良、恶性结节的结节数目、结节形态、结节边界、结节边缘、结构成分、钙化类型、强化方式、甲状腺包膜完整性、淋巴结转移情况比较，差异有统计学意义（P<0.05）。结节赋值评分诊断恶性结节的ROC曲线下面积（AUC）为0.921〔95%CI（0.871，0.971）〕，最佳诊断界值为9分。以病理检查结果为金标准，以甲状腺结节CT量化评分分类系统中1、2、3类为良性结节，4、5类为恶性结节为标准，诊断甲状腺恶性结节的灵敏度、特异度、阳性预测值、阴性预测值、准确率分别为86.67%（13/15）、89.13%（41/46）、72.22%（13/18）、95.35%（41/43）、88.52%（54/61），AUC为0.872〔95%CI（0.742，1.000）〕。结论　基于结节CT征象构建的甲状腺结节CT量化评分分类系统对诊断甲状腺良、恶性结节有较高的诊断效能。

关键词: 甲状腺结节；体层摄影术, 螺旋计算机；影像报告与数据系统