Our study found that independent variables for FTN were pseudocapsule, fissure-filling enhancement, hyperintense on T2WI with enhancement, uniformity of contrast enhancement, and hyperenhancement in early phase. The nomogram for predicting FTN had good discrimination performance, calibration, and clinical utility. Tumor size, cystic degeneration, and restricted diffusion were independent variables for MFTN. Moreover, interobserver agreement was good for these variables. By using the aforesaid independent variables, we constructed RSS to predict MFTN. ROC and NRI analyses indicated that the RSS for predicting MFTN was more accurate than five main TIRADS classification systems.
The nomogram for predicting FTN exhibited good diagnosis performance in both training and validation cohorts. FTN patients might benefit from this nomogram by reducing the need for unnecessary invasive biopsy. In our study, pseudocapsule was an independent risk factor for predicting FTN, and present in 61% of FTN in the training cohort, which was a greater prevalence than in the non-FTN. The pseudocapsule is a fibrous connective tissue ring formed by tumor compression on the peripheral parenchyma, which appears as halo sign on ultrasonography. Li et al. [26] reported that about 74% FTN had a pseudocapsule. Our study showed that few non-FTN presented fissure-filling enhancement with only 1.9%, whereas about 42.9% FTN presented fissure-filling enhancement. Loose follicular structures, fibrous stroma, and hemorrhage were found in the internal area of FTN, with few vessels, whereas densely distributed follicular structures and cells surrounded the loose area, with abundant vessels. These histological features can explain the pathological mechanism of fissure-filling enhancement. Hyperintense on T2WI with enhancement and hyperenhancement in early phase were important factors that predicted FTN. The loose area inside the FTN contains more liquid components, which is the pathological mechanism of hyperintense on T2WI with enhancement. The characteristic of dense area inside FTN is that most of the cells are obviously crowded and/or accompanied by microfollicular formation [27]. These crowded cells contain proliferative-positive cells and, along with neovascularity, lead to hyperenhancement in early phase.
It is difficult to distinguish between benign and malignant FTN, a limitation that presents a clinical, radiological, and histological challenge. Investigators have made attempts to preoperatively distinguish FTC from FTA by, e.g., serum-based analysis [28], radiomics and machine learning based on ultrasonography or CT images [29–31], and the application of ultrasonography features [26, 32–34]. Ultrasonography, however, is limited because its diagnostic performance is heavily dependent on physician experience, and interobserver variability is not negligible [35]. In our study, we developed a RSS to differentiate between MFTN and BFTN based on multiparametric MRI features. The RSS exhibited satisfactory ability to predict MFTN, with an AUC value of 0.902.
Tumor size was a significant factor for predicting MFTN. We defined tumor size as a dichotomous categorical variable with a cut-off of 4 cm. Mu et al. [36] reported that tumor size was an independent predictor for MFTN. Our findings agree with previous studies [37]. Cystic degeneration was another independent risk factor that predicted MFTN. Cystic degeneration was a protective factor for malignancy in non-FTN [38, 39], but a risk factor for malignancy in FTN. Ou et al. [34]reported that cystic degeneration was associated with FTC, which agrees with our results. Other studies based on ultrasonography features did not show the same result, which may be due to different definitions of cystic degeneration or interobserver variability.
In oncology, diffusion-weighted imaging (DWI) is an emerging technique for diagnosis, monitoring, and outcome prediction of malignancies [40]. The DWI enables quantification of water diffusivity via an apparent diffusion coefficient (ADC), which is inversely correlated with the density of cells in the tumor tissue [41]. The ADC has been used as a diagnostic tool to identify benign and malignant thyroid nodules [42]. However, despite the merits of the ADC, quantitative ADC values are not routinely reported in clinical practice, and reproducibility of ADC measurements among different MRI scanners is still challenging. Restricted diffusion was defined as the enhanced lesion areas that appeared hyperintense on DWI and hypointense on ADC. The assessment of restricted diffusion, in comparison with quantitative ADC values, is a feasible and simple method. In our study, restricted diffusion was the most important independent predictor. In terms of pathology, MFTN is composed of small, compact follicular cells, which impose restrictions on water molecule motion.
Our study had limitations. First, the study was a single-center retrospective investigation; thus, the results could have been skewed because of selection bias. Second, the MFTN sample size was small, and data stratified by subtypes of MFTN were not analyzed; thus, a larger MFTN sample size is needed to confirm the accuracy of the RSS in benign and malignant FTN. Third, although the MRI-based imaging features identified by our study are practical and easy to obtain for clinical use, subjectivity was inevitable and may have affected the results. Lastly, the review of static ultrasound images and reports can result in some bias compared with real-life clinical practice. However, the AUC of five ultrasonography-based risk stratification systems to predict MFTN ranged from 0.515 to 0.643 in our present study, which was comparable to findings of Lin et al. [21] who reported an AUC of 0.573 to 0.611.
In conclusion, the nomogram using MRI-based imaging features have achieved favorable diagnostic performance in the preoperative prediction of FTN, which may reduce unnecessary invasive biopsy. The RSS for predicting MFTN may help clinicians to optimize therapeutic decision-making.