ETE, defined as tumor spread outside of the thyroid gland and into the surrounding tissues, occurs in up to 30% of differentiated thyroid cancer cases (19–23). Minimal ETE, detectable only on histological examination, was not a risk factor for disease specific survival and disease related mortality. Gross ETE, or macroscopic ETE, predicted increased recurrence and mortality (24). Thus, the general consensus is to consider gross ETE as an absolute indication for total thyroidectomy and administration of postoperative radioactive iodine (25). Differentiating minimal from gross ETE is essential in the selection of candidates for AS, however, to date, there is no reliable data to evaluate the diagnostic accuracy of ultrasound (US) alone for gross ETE in PTMC. As shown in Table 1, several studies assessed diagnostic ability of US for ETE (minimal and gross) in PTC or PTMC (26–33). The sensitivity and specificity of US ranged from 25 to 100% and from 13 to 93%, respectively. The huge variation in accuracy of US among different studies may result from : different percentage of minimal and gross ETE; : different diagnostic criteria of US; and : different levels of experience of the US technicians. Furthermore, we extracted 9 cases of T4 PTC patients from 5 articles and found that only 1 patient was diagnosed correctly by US, as shown in Table 2 (30–37). That indicates US alone, which is dependent on the experience of the technician and interpreting physician, can't be used to reliably diagnose gross ETE in PTMC.

Tracheal and RLN invasiveness are the most commonly observed gross ETE. Consequently, the Kuma hospital elected to implement “no signs or symptoms of invasion to RLN or trachea” as their selection criteria for AS in PTMC (15). In 2005, a study from Kuma hospital demonstrated US could diagnose tracheal invasion of PTC with extremely favorable sensitivity, specificity, and accuracy of 91, 93, and 93%, respectively (27). Moreover, Ito from Kuma hospital diagnosed tracheal invasion from low-risk PTMC based on the angles between tumor and tracheal wall with 100% sensitivity and 94.5% specificity, while diagnosis of RLN invasion was based on whether the normal rim of the thyroid was clearly present in the direction of RNL with 100% sensitivity and 90.3% specificity. However, 841 (74%) low-risk PTMC patients in this study were diagnosed with help of plain neck computed tomography (CT) because of uncertainty in US imaging. A study enrolled 377 PTC patients demonstrated the combination of US and CT scan decreased the false negative and false positive rates, improving ETE prediction accuracy. In a subgroup of PTMC, the combination of US and CT features also increased positive predictive value (PPV) remarkably (31). Choi et al. demonstrated that contrast-enhanced CT imaging correctly diagnosed a PTC patient as T4, while US alone would have categorized the patient as T3. However, they indicated the combined use of contrast-enhanced CT imaging and US did not improve accuracy for the diagnosis of minimal ETE in PTMC patients (35).

Currently, there are very few studies reporting RLN invasion in PTMC, presumably due to the low incidence of RLN invasion in PTMC. Ito et al. found only 9 of 1,143 PTMC patients with RLN invasion, all 9 of whom had a tumor diameter of 7 mm or larger. Consequently, Ito et al. concluded tumors of < 7 mm in their largest diameter were unlikely to have RLN invasion. But PTMC was derived from abnormal follicular epithelial cell which meant it could be located anywhere within the thyroid. Small PTMCs (< 5 mm) which invade RLN were more likely located in the dorsal part of thyroid. Inaccurate identification for boundaries of small PTMCs and dorsal membrane of thyroid by US may lead to misdiagnosis of gross ETE. Due to limitations in US at the time of evaluation, ETE in these patients was incorrectly assessed. The evidence reminds us of the limited efficacy for US alone in predicting gross ETE and that not all PTMC patients are low risk.

LNM to the central and lateral compartments are common occurrences in PTC, and increase the rate of loco-regional recurrence and mortality, especially among old patients (38). Nearly 80% of PTC patients had micrometastatic lymph nodes on postoperative pathologic examination while 30% had clinical lymph nodes on initial presentation (39, 40). However, as shown in Table 3, the accuracy of preoperative US for diagnosing metastatic lymph nodes is low (26, 35, 36, 41–52). Appropriate selection of candidates for AS requires high sensitivity in order to prevent the enrollment of higher-risk PTMC patients. To predict central lymph node metastases (CLNM), sensitivity of US ranged from 22.6 to 55%, meaning nearly half of CLNM were not correctly diagnosed. This is perhaps due to the difficulty in detecting metastatic lymph nodes in the retropharynx, superior mediastinum, and deep trachea-esophageal groove. In contrast to CLNM, US sensitivity to detect lateral lymph node metastases (LLNM) was much better (62 to 100%). Of note, micrometastases are less important clinically compared to macrometastases. The benefit of treating incidentally identified micro-metastases are not well-demonstrated. Consequently, the American Thyroid Association (ATA) recommended fine needle aspiration (FNA) only for suspicious cervical lymph nodes larger than 8–10 mm (25). Among the articles we summarized in Tables 3, 5 studies focused on metastatic lymph nodes larger than 8–10 mm (26, 35, 47, 50, 52). However, the sensitivity of US for diagnosing CLNM remained low (26–53.2%). US didn't show any advantages in diagnostic ability for larger metastatic lymph nodes compared with the micrometastases.

Shown in Figure 1 and Table 4, standalone CT imaging, or CT combined with US remarkably increased CLNM and LLNM diagnostic sensitivity and accuracy (35, 42, 43, 48, 50). In Choi's study which focused on metastatic lymph nodes larger than 10 mm, combination of US and CT increased sensitivity of CLNM from 53.2 to 73%, and LLNM from 93.9 to 95.9% (35). A separate prospective study from United States demonstrated that the combination of US and CT increased sensitivity of detecting metastatic central and lateral lymph node significantly to 54, 97%, respectively. Accordingly, they suggested combination of US and CT could provide reliable preoperative macroscopic nodal metastasis information to design rational nodal surgery in PTC patients (50).

The limitations of US to detect thyroid nodule and cervical lymph nodes were operator-dependent, presumably due to difficultly in evaluating deep anatomic structures such as mediastinum, parapharyngeal, retropharyngeal and infraclavicular regions, acoustically shadowed by bone, calcification or air (66). As a result, the 2015 ATA guideline recommended preoperative CT as an adjunct to US for patients with large or invasive primary tumor or US suspected advanced disease (25). Nevertheless, is it possible to diagnose gross ETE with virtually 100% sensitivity, as reported by the Kuma hospital? Could we diagnose cervical macroscopic LNM with decent sensitivity only by US? As summarized above, it may be possible to approach this high sensitivity through the combination of diagnostic CT and US imaging, which demonstrated significant improvements in diagnostic performance compared to US alone.

Multiple studies demonstrated that radiologist-performed USs were less accurate and provided inadequate preoperative staging when compared to surgeon-performed USs (67–70). Nearly half of patients received incorrect initial surgery with high local recurrence when an operation decision was made only based on radiologist-performed USs. Denise Carneiro-Pla reported surgeon-performed US changed the therapeutic strategy of 45% thyroid cancer patients through the accurate identification of CLNM/LLNM and thyroid intrathoracic extension (69). Rosebel Monteiro demonstrated that metastatic lymph nodes were diagnosed more frequently by CT imaging than US (70.8 vs. 54%). Moreover, surgeon-performed US was only able to detect 45% of metastatic lymph nodes in a cohort comprised of patients with LLNM (67). In the Kuma hospital, US was performed by specially trained sonographers and retrospectively reviewed by surgeons (15). Thus, it is extremely important to note that the appropriate selection of low-risk PTMC patients for AS is limited by the experience, or inexperience, of diagnosing physicians. Addressing this issue means improvements in both imaging technologies and in the education of physicians play important roles in AS candidate selection.

In terms of diagnostic accuracy, 3-dimensional (3D) US outperformed 2-dimensional (2D) US when compared to patients' final histopathological outcome (94). A single sweep of 3D US provided imaging for reconstruction and overcame the major limitations of 2D US. Kim et al. evaluated 91 thyroid nodules from 85 consecutive patients and compared sensitivity and specificity between 3D and 2D US. They found 3D US had significantly higher sensitivities than 2D in predicting ETE (94). In contrast, a separate study from South Korea reported 3D US with tomographic ultrasound imaging algorithms alone was not superior to real-time 2D US (95). This discrepancy is perhaps attributable to the differences that variable image reconstruction parameters have on US interpretation. Slapa et al. summarized the advantages of 3D US as follows: distinct separation between imaging acquisition and analysis, better remote consultation, less operator dependency, and increased diagnostic accuracy (96).

Recently, shear wave elastography (SWE) has emerged to diagnose and predict the pathologic prognostic factors of PTC using quantitative information about thyroid nodule elasticity. It is operator-independent and can display elastograms of estimated tissue stiffness. Yun et al. enrolled 208 PTC patients and found ETE was associated with the elasticity index determined by SWE, and quantification of the elasticity index could accurately predict pathologic ETE (97). Diagnostic accuracy of cervical lymph nodes was also significantly improved by SWE. Woo et al. reported the elasticity indices of SWE were significantly correlated with not only malignant lymph nodes, but also the number, size and ETE of involved lymph nodes. They concluded quantitative SWE could predict pathologic prognostic factors of cervical LNM (98). Azizi et al. evaluated 270 lymph nodes from 236 patients with both conventional US and SWE. Using single shear wave velocity cut off of 2.93 m/s, SWE could improve diagnostic sensitivity and specificity to 92.59 and 75.46%, respectively. Lymph node stiffness measured by SWE is reported to be an independent predictor of malignant lymph node (99). Xu from China also found predictive performance for CLNM in PTC was markedly improved with the combination of conventional US and SWE, which indicated SWE would be a useful tool for treatment planning (100).

Liu et al. evaluated cervical metastatic lymph nodes using dual-energy spectral CT and found venous phase λ_HU (slope of the spectral Hounsfield unit curve) was the best parameter for diagnosis with sensitivity, specificity of 62.0 and 91.1%, respectively. Compared to conventional CT, quantitative assessment with gemstone spectral CT parameters improved accuracy for detecting cervical metastatic lymph nodes of PTC (101). Considering MRI, several studies have reported the apparent diffusion coefficient (ADC) derived from diffusion-weighted imaging (DWI) could be used as a predictor for thyroid cancer aggressiveness (102–104). Hao et al. evaluated the predictive performance of ADC for ETE of PTCs in a cohort of 23 PTMC patients. PTCs with ETE had significant lower median ADC, 5th percentile ADC, and 25th percentile ADC while PTMCs had significant lower ADC only in 5th percentile ADC (102). Another study used DWI histogram analysis of whole tumor ADC to investigate the relationships between ADC parameters with histopathological features like LNM, ETE, Ki-67, and p53. They found ADC mean, ADC max, ADC median, ADC modus, ADC p75, and ADC p90 were all related significantly with p53, which was prognostic marker for thyroid cancer. Moreover, they identified an inverse correlation between ADC max, ADC p90, and Ki-67, which was regarded as predictor for disease progression during AS (105). Importantly, ADC histogram skewness and kurtosis were also identified to be parameters for predicting LNM (104). Meyer et al. demonstrated MRI texture analysis, which was a novel imaging technique derived from extensive data provided by conventional sequences, was a very useful tool to predict histopathological features in thyroid cancer although they only enrolled 13 thyroid cancer patients (4 PTC; 4FTC; and 5 ATC) (103).

Braf, as a member of RAF kinase family, served as a growth signal transduction protein kinase. Braf^V600E composed nearly 90% of all somatic mutated Braf and played an important oncogenic role in thyroid tumorigenesis (106). The replacement of valine with glutamate at codon 600 resulted from the substitution of thymine with adenine at nucleotide 1799, then activated its serine/threonine kinase constitutively, leading to further activation of MAPK pathway (107). The downstream effectors of mutated Braf, such as Mek and Erk, will be phosphorylated and take part in thyroid tumorigenesis (106, 107). Moreover, Braf^V600E could promote tumor formation and aggressiveness by regulating the expression of other genes epigenetically, either through hyper- or hypomethylation. The interaction between Braf^V600E and epigenetic alterations, which downregulated tumor suppressor genes (like T1MP3, SLC5A8, DAPK1, RARβ2) and upregulated oncogenes (like HMGB2 and FDG1), increased tumor cell proliferation and invasion (108, 109).

In 2015, a meta-analysis was performed to investigate the correlation between Braf^V600E and clinical features for PTMC (110). In Li et al, the authors analyzed 3437 PTMC patients across 19 studies after searching PubMed, EMBASE, and the Cochrane library. They found that Braf^V600E mutation was associated with aggressive clinicopathological features like multifocality, ETE, LNM, and advanced stage of PTMC. Consequently, they suggested Braf^V600E could be used as a risk factor for the stratification and management of PTMC (110). Lee et al. predicted gross ETE of PTMC with 100% sensitivity through the use of tumor size, US features, and Braf^V600E mutation status. They categorized US features of the primary tumor into four groups: A: intraparechymal; B, tumor abutting the capsule < 50% of diameter; C: tumor abutting >50% of diameter; and D: tumor destroyed the capsule. In a subgroup of Braf^V600Enegative patients, a tumor size of 0.7 cm and US categorizations B and C were cut-off values for gross ETE, with 100% sensitivity, whereas US categorizations A and B as cutoff value had 100% sensitivity for predicting gross ETE in the Braf^V600E mutation positive patients (111). Besides clinical risk features, Chen et al. also found PTMCs with Braf^V600E mutation were more likely to recur (OR 2.09 [95% CI:1.31–3.33]) by a meta-analysis of 2,247 PTMC patients from 4 published studies and 2 institutional cohort primary data (112). Niemeier et al. developed a molecular-pathological score (including superficial tumor location, intraglandular tumor spread/multifocality, tumor fibrosis, and Braf status) to stratify PTMC into different risk groups and successfully predict recurrence rate. In the diagnosis of aggressive PTMC, the combination of histologic features and Braf status increased diagnostic sensitivity from 77 to 96% and specificity from 68 to 80% (113). With mounting evidence, revisions to the ATA guidelines in 2015 began to consider Braf^v600e status as a risk factor of structural disease recurrence in PTMC patients after initial therapy (25).

However, Miyauchi et al. in the Kuma hospital detected BRAF^V600E status in 11 PTMC patients without disease progression, 10 PTMC with tumor size progression, and 5 with novel LNM. The percentage of Braf^V600E was 64, 70, and 80% in each group, respectively (114). Consequently, the use of BRAF^V600E alone is insufficient to accurately stratify risk in PTMC patients. If using Braf^V600E alone as biomarker for selecting AS candidates, nearly 60% of PTMCs who may never have disease progression will be categorized wrongly. Considering high prevalence of Braf^v600e mutation among PTMC, Braf^V600E alone cannot be used as reliable biomarker for differentiating aggressive PTMC from indolent ones, and identifying potential disease progression cases from stable ones during AS. A possible reason may be that the oncogenic event driving PTMC aggressiveness requires additional mutations acting in conjunction with BRAF^V600E and the MAPK signaling pathway (115). Thus, the identification of additional genetic variants, which are less abundant than BRAF^V600E, could be important in predicting PTMC aggressiveness.

Telomerase reverse transcriptase (TERT) is the catalytic protein subunit of telomerase, which can maintain chromosomal integrity and genome stability (116). Malignant cancer cells, which were replicative immortal, required activation of telomerase and regulation of other growth controlling genes, pathways and molecular by TERT (117). First reported in 2013, TERT promoter mutation (C228T and C250T) in thyroid cancer has progressed rapidly in recent 5 years (118). Many studies have demonstrated TERT mutation was associated with more aggressive clinicopathological features of thyroid cancer, such as male gender, ETE, LNM, advanced stage, distant metastasis, recurrence, and mortality (119–122). Two meta-analyses in 2016 investigated clinicopathological significance of TERT promoter mutations in PTC and found the average prevalence of TERT promoter mutation was around 10%. Additionally, PTC patients with TERT promoter mutation displayed more aggressive histopathological features (121, 122). Kim et al. developed an effective risk stratification system using TERT promoter mutation status that reliably predicted structural recurrence and mortality in DTC patients (119).

Of note, the co-occurrence of Braf and TERT promoter mutations enhanced the predictive ability for prognosis of PTC. Moon et al. performed a meta-analysis including 13 studies with 4,347 PTC patients and found the co-occurrence of Braf and TERT promoter mutations was more significantly associated with aggressive clinicopathological features than either mutation alone (123). Accordingly, they believed these two mutations had a synergistic effect on prognosis and were useful in risk stratification of PTC. Liu et al. categorized 1,051 PTC patients according to mutation status of Braf and TERT promoter and demonstrated deaths per 1,000-person years in PTC patients with neither mutation, Braf ^V600E alone, TERT mutation alone, or both mutations were 0.80, 3.08, 6.62, and 29.86, respectively. Simple 4-genotype classification can predict disease-specific mortality accurately (124). Recently, this synergistic effect of BRAF and TERT promoter has been demonstrated as Braf^V600E → MAPK pathway → FOS → GABP → TERT signaling/transcription axis in human cancers (125). Firstly, mutated Braf^V600E activated MAPK pathway, which phosphorylated FOS to be an active transcription factor for activating the GABPB promotor. Then increased expression of GABPB and formation of GABPA-GABPB complex activated the mutant TERT promoter. In this axis, phosphorylated FOS played important oncogenic bridging role between Braf^V600E and TERT promoter mutations (125).

However, de Biase et al. detected TERT promoter mutations with next-generation sequencing in 431 PTMC patients assembled from six different institutions. They found the prevalence of TERT promoter mutations among PTMC was only 4.7%, less than the 10% reported in PTC patients. Moreover, the presence of TERT promoter mutations was not associated with unfavorable clinicopathological features (126). Also in Miyauchi's study, no PTMC patients undergoing AS were positive for TERT promoter mutations, even in a subgroup of patients with increased tumor sizes and/or novel lymph node appearance (114). Therefore, with regard to its low prevalence in PTMC, TERT promoter mutations are unlikely to be reliable molecular markers of tumor aggressiveness/progression.

MicroRNA is defined as a group of small endogenous, single stranded non-coding RNAs of 19–25 nucleotides that can exclusively regulate their proprietary mRNA expression (127). The miRNA-221-222 cluster, downstream of the MAPK pathway, played an important role in tumorigenesis and aggressiveness for PTC (128). Located on the X chromosome, miRNA-221-222 cluster was regulating PTC formation and invasion through negative regulation of p27 (129). Multiple studies have demonstrated that upregulated miR-221-222 cluster was associated with more unfavorable clinicopathological features, treatment resistance, increased recurrence rate, and worse prognosis (130–133). Because of that, the miRNA-221-222 cluster was considered as a potential biomarker for aggressive PTC. Additionally, miRNA-146b is another well-studied and overexpressed microRNA in PTC. Its expression level was positively associated with tumor aggressiveness and poor prognosis (131, 133). Study has shown miRNA-146b functioned in PTC through binding with the 3′UTR region of retinoic acid receptor beta (RARβ) (134). Moreover, advanced PTC patients could receive benefit from retinoic acid (a RARβ ligand) treatment. Retinoic acid treatment resulted in tumor shrinkage and increased radioiodine uptake in 38% and 26% of patients, respectively (135). These studies suggested that miRNA-146b might play important role in thyroid cancer initiation and progression. In addition to the two microRNAs discussed above, there are also other microRNAs which have been identified to be associated with tumor aggressiveness (especially ETE, LNM and distant metastasis) including miRNA135-b, 146-a, 146-5p and several others (Table 6) (129, 136–155).

The upregulation of miRNA-221-222 cluster and miRNA-146b in BRAF^V600E positive tumors, was suggested to be attributable to activation via the NF-κB pathway (156, 157). In Braf^V600E PTMC patients, it remains unknown what molecular events trigger disease progression during AS. Would it be possible to increase our ability to predict PTMC disease progression by screening FNA biopsies for clinically actionable somatic mutations and/or the expression of miRNAs?

Compared with inherent instability of mRNA, circulating miRNA is subjected to nuclease activity and resistant to environment. Because of that, miRNA, which can be readily detected in bloodstream, is believed as a potential ideal candidate serum biomarker for PTC (158). Yu et al. detected serum miRNA expression by Solexa sequencing and found increased miR-151-5p, detected in the serum, was associated with LNM of PTC (159). However, the evidence of using circulating miRNA to predict disease progression of PTMC during AS was absent. In addition, a prospective observational pilot study found circulating myeloid-derived suppressor cells, which were detected preoperatively by novel flow cytometry-based immunoassay, were positively associated with a higher TNM stage and disease recurrence (160). Lubitz et al. reported they only detected 63% circulating Braf^V600E mutation by novel RNA-based blood assay compared with conventional tissue assays on surgical specimens. They concluded detecting circulating Braf^V600E could be a surrogate for conventional FNA detection (161). In contrast, a separate study found only 37.3% of PTC patients with locally advanced and metastasis were detected to have circulating Braf^V600E mutation. These patients didn't get any benefits from analysis of circulating tumor DNA (162). Accordingly, there are several challenges about the application of serum circulating biomarkers for PTMC which include: ① Molecular FNA diagnostics with biomarkers have high concordance with pathological results. In contrast, serum circulating biomarkers demonstrate only partial concordance with FNA determined pathology. Consequently, circulating biomarkers from blood are not superior to FNA biopsies in predicting aggressiveness. ② All studies about detecting serum circulating biomarkers enrolled cancer patients with advanced stage or distant metastasis. However, the serum circulating biomarkers identified in high-risk patients may not be detectable in low-risk PTMC patients. ③ Genetic background and alternations in circulating cells may be different with those in the primary tumor. Some cancer cells derived from the primary tumor may undergo changes that facilitate blood vessel invasion and then turn to circulating cells. ④ Other malignant tumors shared the same circulating miRNA or DNA with thyroid cancer. Differentiating where these circulating biomarkers came from is difficult.

Besides genetic alternations, LncRNAs, which is defined as a class of RNAs containing over 200 nucleotides, play important roles in tumor progression (163). Kim et al. reported LOC100507661 expression was positively related with LNM and Braf^V600E mutation in PTC patients (164). High expression of HOTAIR in thyroid cancer was associated with larger tumor size, more metastatic lymph nodes, and poorer outcome after a meta-analysis of TCGA and GEO databases (165). PTC patients with high expression of HIT000218960 had more multifocality, LNM and advanced TNM stage (166). Down-regulation of LINC00271 was identified as an independent risk factor for ETE, LNM, TNM stage and recurrence (167). Other LncRNAs, which related with aggressiveness of PTC, were also identified and summarized in Table 7 (164, 166–183).

Epigenetic changes, particularly methylation of DAPK, REC8, TIMP3, CDH1, FGFR2 were also reported to be associated with aggressive behavior of PTC (184). Whether we can predict the aggressiveness of PTMC using these biomarkers derived from PTC patients remains to be investigated.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.