Objective: To explore the values of ultrasonographic features in differentially diagnosing benign and malignant thyroid nodules, and attempt to establish a quantitative ultrasound classification system.
Methods: We retrospectively analyzed 20 ultrasound features of 926 thyroid nodules in 527 patients. Using logistic regression analysis, we obtained the probability function for predicting the malignancy in thyroid nodules and established a preliminary ultrasound classification system.
Results: The ages of patients with thyroid nodules was older than that of the patients with thyroid carcinoma (t = 6.496, P < 0.001). The correctness rate of ultrasonic diagnosis was 80.1%. The logistic multivariate regression analysis showed that among all ultrasonographic features, aspect ratio ≥ 1, obscure boundary, irregular margin, significant internal hypoecho, internal hypoecho, internal micro-calcifications, posterior echo attenuation, thyroid capsule invasion, abnormal adjacent lymph nodes, and ultrasonic elastography 5-point evaluation scores > 3 were contributing factors for thyroid carcinoma. The equation was P (us) = 1 /[1+e(-)Z], where z is the logit of malignant thyroid nodule, and taking P (us) > 0.50 as boundary value, the prediction accuracy rate was 84.1%. Using this model, 92.2% of the thyroid nodules were predicted as nodular goiter, and 69.4% of the thyroid carcinomas were correctly predicted. P (us) was stratified into four levels: Level 1: 0-16% malignancy; Level 2: 17%-50% malignancy; Level 3: 51%-70% malignancy; and level 4: 71%-100% malignancy.
Conclusions: The quantitative thyroid imaging reporting and data system developed in this study makes ultrasound reports more objective, normalized and standardized. It can be used to evaluate the malignancy risk of thyroid nodules and help to make right decision in clinics.