Establishment of an Ultrasound Malignancy Risk Stratification Model for Thyroid Nodules Larger Than 4 cm

Abstract

BackgroundThe incidence and mortality of thyroid cancer, including thyroid nodules > 4 cm, have been increasing in recent years. The current evaluation methods are based mostly on studies of patients with thyroid nodules < 4 cm. The aim of the current study was to establish a risk stratification model to predict risk of malignancy in thyroid nodules > 4 cm.MethodsA total of 279 thyroid nodules > 4 cm in 267 patients were retrospectively analyzed. Nodules were randomly assigned to a training dataset (n = 140) and a validation dataset (n = 139). Multivariable logistic regression analysis was applied to establish a nomogram. The risk stratification of thyroid nodules > 4 cm was established according to the nomogram. The diagnostic performance of the model was evaluated and compared with the American College Radiology Thyroid Imaging Reporting and Data System (ACR TI-RADS), Kwak TI-RADS and 2015 ATA guidelines using the area under the receiver operating characteristic curve (AUC).ResultsThe analysis included 279 nodules (267 patients, 50.6 ± 13.2 years): 229 were benign and 50 were malignant. Multivariate regression revealed microcalcification, solid mass, ill-defined border and hypoechogenicity as independent risk factors. Based on the four factors, a risk stratified clinical model was developed for evaluating nodules > 4 cm, which includes three categories: high risk (risk value = 0.8-0.9, with more than 3 factors), intermediate risk (risk value = 0.3-0.7, with 2 factors or microcalcification) and low risk (risk value = 0.1-0.2, with 1 factor except microcalcification). In the validation dataset, the malignancy rate of thyroid nodules > 4 cm that were classified as high risk was 88.9%; as intermediate risk, 35.7%; and as low risk, 6.9%. The new model showed greater AUC than ACR TI-RADS (0.897 vs. 0.855, p = 0.040), but similar sensitivity (61.9% vs. 57.1%, p = 0.480) and specificity (91.5% vs. 93.2%, p = 0.680).ConclusionMicrocalcification, solid mass, ill-defined border and hypoechogenicity on ultrasound may be signs of malignancy in thyroid nodules > 4 cm. A risk stratification model for nodules > 4 cm may show better diagnostic performance than ACR TI-RADS, which may lead to better preoperative decision-making.

Development and validation of clinical predictive models for differentiated thyroid carcinoma and cervical lymph node metastasis: A retrospective cohort study

“…strati ed thyroid nodules by diameter and demonstrated that ACR-TIRADS had a better predictive performance in nodules > 4 cm in diameter. 27 Our results suggest that the model incorporating the abovementioned factors is superior to TIRADS alone in predicting nodules in terms of both calibration and discrimination, although no de nitive study has con rmed that indicators such as number and size have a clear causal relationship with the malignant trend of nodules. This result may be similar to the results of the study that is associated with radiomics.…”

Assessment of the statistical optimization strategies and clinical evaluation of an artificial intelligence-based automated diagnostic system for thyroid nodule screening

“…Additionally, tests of nodule-size stratification are often ignored during evaluation of AI systems (16)(17)(18)(19). However, there have been numerous studies showing size-dependent diagnostic difficulty for thyroid nodules by radiologists using conventional B-mode ultrasound-based TI-RADS criteria (20)(21)(22). It is important for radiologists to know the limitation of an AI system for certain categories of cases, such that they need to exercise more caution if an AI system shows considerably worse diagnostic accuracy than that achieved by the radiologists themselves.…”