Multiparametric magnetic resonance imaging in preoperative assessment of follicular thyroid neoplasm

Abstract

Abstract Objectives To examine multiparametric magnetic resonance imaging for differentiating follicular thyroid neoplasm (FTN) from non-FTN and malignant FTN (MFTN) from benign FTN (BFTN). Methods Seven hundred two thyroid nodules, postoperatively confirmed by pathology, were retrospectively investigated and divided into two cohorts: training (n = 482) and validation (n = 220). The 133 FTNs were split into two groups: BFTN (n = 116) and MFTN (n = 17). Univariate and multivariate logistic regression analysis were used to identify independent predictors of FTN and MFTN. An nomogram for FTN and a risk score system for MFTN were constructed based on the results of multivariable analysis. Nomogram’ performance was evaluated based on discrimination, calibration, and clinical utility. The diagnostic performance of the risk score system for MFTN was compared with the performance of the Thyroid Imaging Reporting and Data System (TIRADS). Results The nomogram, which incorporated independent predictors, demonstrated good discrimination and calibration for differentiating FTN and non-FTN both in the training cohort (AUC = 0.947, Hosmer-Lemeshow P = 0.698) and the validation cohort (AUC = 0.927, Hosmer-Lemeshow P = 0.088). Tumor size, restricted diffusion, and cystic degeneration were risk factors for differentiating MFTN from BFTN. The AUC of the risk score system for MFTN prediction was 0.902 (95% CI 0.811–0.993), and the sensitivity, specificity, accuracy, and positive and negative predictive values of the risk score system at the optimal cutoff value were 76.5%, 94%, 91.8%, 65%, and 96.5%, respectively, which was better performance than five TIRADS. Conclusions The models based on MRI features had favorable diagnostic performance for preoperative prediction of FTN and MFTN. These models may aid in reducing unnecessary invasive biopsy or surgery.