Insights in biomarkers complexity and routine clinical practice for the diagnosis of thyroid nodules and cancer

Introduction

Thyroid nodular disease continues to be a major challenge in clinical practice because the differential diagnosis between benign and malignant nodules remains fraught with many difficulties and doubts. Thyroid nodules are commonly found in the general population, with prevalence rates varying based on the method of detection: 2–6% by palpation, 19–35% by ultrasound, and 8–65% based on autopsy data (Tamhane & Gharib, 2016). Thyroid cancer (TC) represent about 1% of total cancers, being the most frequent endocrine malignancy with a frequency of more than 90% (Siegel et al., 2021; Miranda-Filho et al., 2021). Thyroid cancers are diagnosed in 1% to 5% of thyroid nodules (Grussendorf, Ruschenburg & Brabant, 2022), being more than 90% differentiated thyroid carcinoma (DTC), out of which papillary thyroid carcinoma (PTC) is the most prevalent (over 85%) (Li, Brito & Vaccarella, 2020). The 2022 5th edition of the WHO Classification of Endocrine and Neuroendocrine Tumors (WHO, 2022) classified thyroid tumors into two groups based on their cells of origin: the major group derived from follicular epithelial cells and another from calcitonin-secreting C cells (3–5%). Follicular cell-derived tumors, which represent the majority of thyroid neoplasms, are categorized into benign, low-risk, and malignant neoplasms. These tumors show a significant correlation between histological phenotype and genotype, allowing for a refinement of the 5th WHO Classification of Thyroid Tumors. This refinement aims to enhance the understanding and standardization of differentiated thyroid carcinomas. The classification is clinically relevant as it also serves as a prognostic risk classification (Table 1). The incidence of PTC is increasing, attributed to environmental factors and the higher sensitivity of complementary diagnostic techniques, although this increase is not accompanied by a change in mortality rate (Ferlay et al., 2019; Kitahara & Sosa, 2020). PTC generally has a favorable prognosis, but certain clinicopathological features and genetic alterations can affect its progression (Ren et al., 2018). Overdiagnosis of thyroid cancer is currently a global public health issue (Pizzato et al., 2022). Improving pre-surgical diagnostic tools is crucial for optimizing treatment and surgical options in thyroid nodular disease (Hlozek et al., 2022).

Ultrasound-guided fine needle aspiration cytology (US-FNAC) remains the gold standard for pre surgical diagnosis of thyroid nodular disease, though over 25% of nodules still do not have a definitive diagnosis (Ali & Cibas, 2018). The 2023 Bethesda System for Reporting Thyroid Cytopathology (TBSRTC) has updated the standardized reporting system for thyroid fine needle aspirations. However, diagnostic limitations persist in nondiagnostic (B-I) and indeterminate Bethesda categories (B-III and B-IV) (Ali et al., 2023). Each category presents a different risk of malignancy (ROM), with varying treatment and follow-up options (Bongiovanni et al., 2018; Basolo et al., 2023).

In recent years, significant advancements have been made in understanding the molecular mechanisms and signaling pathways of thyroid carcinogenesis. Thyroid neoplasms develop and progress based on the activation or inactivation of somatic mutations and changes in gene expression patterns. Key genetic alterations in thyroid cancer include point mutations in the BRAF, RAS, TERTp, RET, and TP53 genes, as well as fusion genes like RET/PTC, PAX8/PPAR-γ, and NTRK.

Numerous studies have aimed to identify genetic biomarkers to improve pre-surgical diagnosis, especially in indeterminate nodules, and to predict thyroid cancer prognosis, but definitive evidence is still lacking (Grimmichova et al., 2022; Nixon et al., 2017). The assessment of microRNA (miRNA) expression is another promising area in thyroid carcinogenesis research (Silaghi et al., 2021; Titov et al., 2020; Agarwal, Bychkov & Jung, 2021). miRNAs regulate about 30% of the human genome and play a crucial role as gene global regulators of expression (Wójcicka, Kolanowska & Jażdżewski, 2016). They influence intracellular regulatory processes involved in carcinogenesis, such as differentiation, proliferation, and apoptosis (Ludvíková, Kalfeřt & Kholová, 2015; Abdullah et al., 2019). Studies on miRNA expression profiles in human tumors have identified specific “signatures” associated with diagnosis, staging, prognosis, and treatment response (Jin et al., 2019; Marini, Luzi & Brandi, 2011). Additionally, correlations between the expression levels of selected miRNAs and specific genetic mutations have been linked to increased aggressiveness and poorer prognosis in thyroid cancer. Often, thyroid surgery is the only diagnostic approach for suspicious and malignant nodules. The inability to differentiate between benign and malignant nodules in a pre-surgical phase does not prevent unnecessary surgeries and potential complications (Hauch et al., 2014).

Artificial intelligence (AI) has emerged recently as a new area of interest in scientific research, being a complementary tool in the diagnosis and classification of thyroid nodules, mainly in image, in cytology and histology studies, but still with uncertain practical applicability (Ludwig et al., 2023; Sorrenti et al., 2022).

In the last 2 years, three articles were published with significant guidelines to the approach to thyroid nodular pathology and cancer, which included genetic and molecular biomarkers for pre-surgical diagnosis and prognosis: in 2022, the 5th edition of the WHO Classification of Endocrine and Neuroendocrine Tumors (WHO, 2022), and in 2023, the 3rd TBSRTC (Ali et al., 2023) and the European Thyroid Association guidelines for the management of thyroid nodules (Durante et al., 2023). However, in addition to the lack of consensus on which panels of genetic and molecular biomarkers to use, their implementation in clinical practice is not yet possible in the vast majority of countries due to technical and economic reason (Oczko-Wojciechowska et al., 2020; Moore et al., 2021). The rising incidence of thyroid neoplasm (Sung et al., 2021; Raposo et al., 2017; Vaccarella et al., 2016), coupled with the low incidence of aggressive thyroid carcinoma (Durante et al., 2018), urges the exploration of strategies to improve the diagnostic accuracy in a pre-surgical phase, particularly for indeterminate nodules, to prevent unnecessary surgeries. The aim of this review is to discuss the impact of biomarkers in the routine diagnostic practice of thyroid nodules and cancers, originated from follicular epithelial cells.

Methodology

The literature search for this review was performed through the digital platforms Pub Med (https://pubmed.ncbi.nlm.nih.gov/), Science Direct (https://www.sciencedirect.com/), Google Scholar (https://scholar.google.com/). Boolean operators (AND, OR) and predetermined keywords related with molecular biomarkers (genetic and miRNAs), thyroid nodules and cancer, ultrasound fine-needle aspiration cytology (US-FNAC), and Bethesda and WHO classification were applied. We selected 156 publications with significant contribution to this topic, being the majority 86 (55.1%) between January 2019 to March 2024, due to the greater volume of publications in the last 5 years resulting from increased scientific interest in the subject, 45 (28.8%) between January 2014 to December 2018 and 24 (15.4%) before 2014, including some publications of our group in those periods. The inclusion criteria used for the selection of article titles in our review were based in articles on the same specific topic, published in recognized scientific journals in digital platforms with high contribution to developing the proposed topic. We only included articles that had full text available and described the exact method and results regarding molecular biomarkers and thyroid nodules and cancer. We excluded articles that were not focus straightly on our topic, especially those articles that not emphases the impact of molecular biomarkers in refine the pre-surgical diagnosis of thyroid nodules.

Thyroid nodules—a global public health problem

Thyroid nodules (TNs) are becoming more prevalent in clinical practice in recent years, with prevalence in the general population varying from 2% to 65%, depending on diagnostic techniques (Tamhane & Gharib, 2016). Mu et al. (2022) presented a systematic review and meta-analysis about epidemiology of TNs, considering a prevalence of 24.83% (95% CI [21.44–28.55]). They showed that TNs had become more prevalent during 2012–2022 (29.29%) when compared with 2000–2011 (21.53%, p = 0.02). Despite the lack of evidence with iodine nutrition, their results showed a significant association between increasing TN prevalence and older age, gender (female preponderance) and overweight. The incidence of TC has increased in many countries all over the world, but mortality continuous stable at lower rates, probably associated to the over diagnosis of TNs (Pizzato et al., 2022; World Health Organization, 2024). The GLOBOCAN 2022 (World Health Organization, 2024) database showed the global distribution of TC incidence and mortality rates in world population. The incidence rates, using the age-standardized rate, were 13.6 per 100,000 women and 4.6 per 100,000 men, and mortality rates were 0.53 for women and 0.35 per 100,000, under one per 100,000 in most countries. The incidence rates in women differ by 15 times among regions of the world, whereas mortality rates were relatively similar across different settings. Incidence rates were five times higher in countries with high human development index, which may be related to the health care system and TC overdiagnosis in each country, and also to other unclear reasons (Lamartina et al., 2022; Li, Dal Maso & Vaccarella, 2020). TC was the fifth most frequent cancer in women and the thirteen in men in 2022. The numbers of TNs and TC are expected to rise during the next decades, and research is needed to quantify the determinants of these trends, as well as the potentially deleterious effects of overdiagnosis and overtreatment in the world population.

Fine needle aspiration cytology of thyroid neoplasm

The differential diagnosis of thyroid nodules integrates clinical evaluation, with a history and physical examination, laboratory tests, neck ultrasound (US), and cytology via ultrasound-fine needle aspiration (US-FNAC). Thyroid US evaluate some features suggestive of malignancy in nodules, such as hypoechogenicity, irregular/infiltrative border, microcalcifications, central vascularization, and a taller-than-wide shape in the transverse view, especially if cervical lymph node involvement is present. The diagnostic sensitivity and specificity of each features is variable and none of them alone, can reliably distinguish malignant from benign nodules (Kobaly, Kim & Mandel, 2022). Thyroid imaging reporting and data systems (TIRADS) available were created based in risk stratification systems, to standardize the criteria used in thyroid US. However, no consensus exists yet, and current efforts intent to develop an international TIRADS to unify criteria and recommendations (Dobruch-Sobczak et al., 2022). US-FNAC, a minimally invasive technique with low complications and reduced costs, is the gold standard method to evaluate thyroid nodules. Martin & Ellis (1930), in 1930, were the first to describe the use of needle biopsy for the diagnosis of thyroid nodules, using an 18-gauge–needle aspiration technique. Thereafter, other techniques using cut-needles were used, but not wide accepted due to the fear of malignant implants in the needle track, false-negative results, and severe complications. In the 1960s, Scandinavian investigators introduced small needle aspiration biopsy of the thyroid, which came into widespread use after the 1980s (Bäckdahl et al., 1987). The accuracy of FNAC was improved with the integration of thyroid US in the 2000s. Thyroid US is recommended in patients with a single or a multinodular goiter or in a patient suspected of having a nodule, but it is not to be used as a screening test for the general population (Gharib et al., 2010). The US-FNAC technique used is minimally invasive, being a reliable and usefulness diagnostic test for determining malignancy in a nodule, becoming a standard test, often performed by an endocrinologist (Godinho-Matos, Kocjan & Kurtz, 1992; Todsen et al., 2021; Jasim, Dean & Gharib, 2023). FNAC is also required if a suspicious lymph node (LN) was detected by US, and a washout for thyroglobulin measurement should be performed (Patel et al., 2020). The need to clarify the terminology concerning thyroid fine needle aspirations, and the communication among physicians, and other health care providers, determined the standardization of criteria. The Bethesda system for reporting thyroid cytopathology (TBSRTC) emerged in 2007, the 1st edition, and in 2016, the 2nd edition and in 2023, the 3rd edition (Ali & VanderLaan, 2023). The 3rd TBSRTC, integrating the 5th WHO classification of thyroid tumors (Ali et al., 2023), standardized and reporting system for thyroid fine needle aspirations, considered six diagnostic categories, with a different risk of malignancy each, as follow: (I) Nondiagnostic (ND); (II) Benign; (III) Atypia of undetermined significance (AUS); (IV) Follicular neoplasm (FN); (V) Suspicious for malignancy; and (VI) Malignant (Table 2). Bethesda II and VI cytological results have excellent performance when assessing the benign or malignant nature of a thyroid nodule, with a sensitivity and specificity, near to 100% in the diagnosis of thyroid cancer (Haugen et al., 2017). Diagnostic limitations are particularly significant in the nondiagnostic (B-I) and in indeterminate Bethesda categories (B-III and B-IV) (Basolo et al., 2023). The differential diagnose between benign or malignant thyroid nodules is obtained in 70–75% of cases by US-FNAC (Nikiforova et al., 2018), with an accuracy of 68.8%, including 25% of indeterminate nodules, and a positive predictive value (PPV) of 98.6% and a negative predictive value (NPV) of 55.9%, for adult population (WHO, 2022). The expected ROM for Bethesda categories was of 13–30% (B-III), 23–34% (B-IV), 67–83% (B-V) and 97–100% (B-VI) (Lebrun & Salmon, 2024). When the noninvasive follicular thyroid neoplasm with papillary like nuclear features (NIFTP) is excluded, the ROM evaluated changed to 16% (B-III), 23% (B-IV), 65% (B-V) and 94% (B-VI) (Gharib et al., 2010). Both the 2023 Bethesda Classification and the 2022 WHO Classification of Thyroid Tumors, consider the analysis of biomarkers to refine the diagnosis in a pre-surgical phase and to provide prognostic information in thyroid tumors. This contributes to reduce unnecessary surgeries, often associated with hypothyroidism and lifelong thyroid hormone replacement, as well as surgical complications, such as hypoparathyroidism (transient or permanent in 0.12–11%, respectively), and recurrent laryngeal nerve paralysis (Yazıcıog˘lu et al., 2021; Brisebois et al., 2021).

Biomarkers in thyroid neoplasm

The study and identification of genetic and molecular alterations in thyroid may contribute with preoperative biomarkers to precise the ROM, and with postoperative molecular testing to provide prognostic information and potential targeted molecular alterations, particularly useful for therapy of radioiodine resistant cases, and contributing to a personalized and effective treatment of the patients (Acquaviva et al., 2018; Baloch et al., 2022).

Biomarkers in clinical practice

Only in 2022, the 5th WHO classification of thyroid neoplasms (WHO, 2022) and in 2023 the 3rd TBSRTC (Ali et al., 2023) included genetic and molecular biomarkers for pre-surgical diagnosis, refining the diagnostic categories and their risk of malignancy, and for prognosis of thyroid nodules, with corresponding management strategies. Molecular tests are especially recommended for indeterminate cytological FNAs, contributing to management strategies (Durante et al., 2023). Currently, many molecular platforms exist for thyroid FNAs molecular analysis, ranging from laboratory-developed tests, like the 7-gene mutational panel used in some European countries (Bardet et al., 2021), to commercialized tests such as Afirma, ThyroSeq, ThyGenX/ThyraMIR, and RosettaGX Reveal. These platforms use different targets and present various negative predictive values (NPV) (81–100%) and positive predictive values (PPV) (29–81%) (Gharib et al., 2010; Song et al., 2020). The SWEETMAC study (2021) (Bardet et al., 2021), using the 7-panel mutation testing in indeterminate thyroid nodules, presented an accuracy of 88%, sensitivity of 48%, specificity of 95%, PPV of 67%, and NPV of 91%. The results increased the overall risk of cancer from 16% to 67% in mutated nodules and, by diminishing it to 9% in wild type, confirmed the relevance of the 7-panel mutation testing in the diagnosis of indeterminate nodules. Tang et al. (2023) presented a study using a 5-gene panel test on FNAC samples from 759 nodules, reporting sensitivity, specificity, PPV, NPV, and accuracy of 96.83%, 100%, 100%, 42.86%, and 96.90%, respectively. They concluded that the 5-gene panel may represent a valid adjunct technique for distinguishing thyroid nodules. Despite the European and American thyroid associations (Durante et al., 2023; Haugen et al., 2017) consider the inclusion of molecular tests in US-FNAC, no consensus is yet achieved on the best molecular panel to use in the diagnosis of thyroid nodules. Available molecular tests have been developed in United States of America, and it is important to assess whether such tests can be used in other populations (Grimmichova et al., 2022). The most significant disadvantage of commercially available tests is related to their cost, and the lack of highly specialized reference laboratories in many countries (Titov et al., 2020; Agarwal, Bychkov & Jung, 2021). Other limitations in the studies reviewed on biomarkers in the diagnosis of thyroid nodular disease were mainly related to their retrospective nature and the small size of the samples, as well as the available laboratory techniques for studying molecular biomarkers in cytology and histology samples. Many doubts persist regarding the reliability of cyto-histological correlation, depending on the technique used and the quality of the cytological and histological samples. Most limitations in cytology samples were caused by the small amount of cellular material, reducing the number of neoplastic cells, and the heterogeneity and multifocality of tumors (Matos et al., 2024). The quality of hystology samples stored in pathology archives for many years was affected by the degradation of nucleic acids due to formalin fixation (Penna et al., 2017). US-FNAC is the central minimally invasive method for the study of thyroid nodular disease in a pre-surgical phase. An acceptable matched cyto-histological genetic profile agreement for molecular alterations was attained in a study of our group (Matos et al., 2024). Using 209 cases of DTC, our results indicate a good level of consistency between the two types of samples in the genetic analysis. BRAF and HRAS were the genes showing substantial cyto-histological molecular agreement, whereas moderate agreement was found for NRAS and TERTp genes. RAS genes were found to be the most frequently mutated in the indeterminate nodules, in accordance with Censi et al. (2017). The BRAF and/or TERTp mutations were only present in malignant cases, with a PPV of 100%, in both cytology and histology samples. While the Genetic profile’s ability to exclude malignancy was limited (low sensitivity and NPV), it effectively confirmed malignancy in cytology samples, which we could consider as “rule-in” tests, in accordance with 2015 ATA guidelines (PPV > 95%) (Haugen et al., 2017), reducing repeated US-FNACs. A good cyto-histological genetic profile agreement for molecular alterations was achieved between the two types of samples in indeterminate nodules, reinforcing the role of molecular analyses before surgery in those cases. The study of miRNA expression can be useful in a precocious and accurate diagnosis, as well as in the evaluation of progression of TC (Yoruker et al., 2016; Galuppini et al., 2022). The overexpression of miRNAs is correlated with tumor aggressiveness, high tumor node metastasis (TMN) stage, ETE, recurrence, and metastases, representing a real signature of TC, particularly of PTC (Celano et al., 2017; Pamedytyte et al., 2020), that can be analyzed in cyto-histology samples or in blood (circulating miRNAs) (Park et al., 2021; Pitoia, 2024).

The usefulness of molecular biomarkers in prognosis

The 5th WHO Classification of thyroid tumors considered new tumor entities that are clinically and morphologically borderline with a very low risk of distant metastases (Baloch et al., 2022). Low-risk follicular neoplasms represent a clinical challenge due to a lack of consensus among observers concerning whether tumor morphological characteristics indicate malignancy, such as the presence of capsular and vascular invasion, and/or extra thyroidal extension (ETE) (Nishino & Nikiforova, 2018; Turk et al., 2019). Additionally, low-risk neoplasm are not associated with specific diagnostic molecular biomarkers and are considered “molecular indeterminate” tumors. RAS mutations are found in a broad spectrum of thyroid neoplasm, from benign to malignant follicular-patterned tumors (Baloch et al., 2022). Tumors with RAS mutations are associated with a good prognosis (Tidyman & Rauen, 2016; Gilani et al., 2022). However, there is evidence that RAS mutations may be an early event in tumorigenesis, potentially leading to the development of the malignancy and dedifferentiation of the cells (Radkay et al., 2014). Detecting RAS mutations in thyroid samples in cytology FNAs, particularly in indeterminate nodules, provides improvement on risk stratification with evidence for neoplasia and has the potential for patient management guidance (Radkay et al., 2014; Howell, Hodak & Yip, 2013). However, it does not help distinguish between benign and malignant lesions. BRAF and RAS are mutually exclusive events, with no concomitant mutations expected (Park et al., 2013). The specificity of BRAF mutation in the diagnosis of PTC is nearly 100%, but its sensitivity is below 35% (Xing, 2005; Matos et al., 2024). Different studies have shown an association between BRAFV600E and ETE, tumor multifocality, lack of tumor capsule, and the presence of lymph node metastases, all of which predict a high propensity for local aggressiveness (Torregrossa et al., 2016; Xing et al., 2014). Although BRAF mutations were significantly associated with several clinicopathological features (Liu et al., 2018), these mutations are also highly prevalent in low-risk, infracentimeter PTCs, and their prognostic role remains debated (Soares & Sobrinho-Simões, 2011; Scheffel, Dora & Maia, 2022). Several authors demonstrated that TERTp mutations are associated with worse pathological features (older age, tumor size and stage), ETE, vascular invasion, aggressive behavior, treatment resistance, distant metastases, poor prognosis, and a higher risk of disease-specific mortality (Melo et al., 2014; Panebianco et al., 2019; Penna et al., 2018; Bournaud et al., 2019). A synergistic impact of TERTp mutations with BRAF mutations has been suggested, leading to a worse prognosis (e.g., distant metastases, cancer dedifferentiation), particularly in older patients (Estrada-Flórez et al., 2019). TERTp mutations also impact resistance to radioactive treatment as described in these tumors (Abdullah et al., 2019; Volante et al., 2009). Moreover, the presence of TERTp, PLEKHS1 promoter or TP53 mutations (in association with either BRAF or RAS mutations) was significantly associated with radioiodine refractory disease (Jung et al., 2020). In advanced TC, molecular testing must be considered since molecular alterations are targetable by specific drugs, such as RET, NTRK1-3, BRAF, ALK, ROS and mTOR (Gharib et al., 2010; Salvatore, Santoro & Schlumberger, 2021). Several associations between the expression levels of miR-146b, miR-221 and miR-222 with the presence of BRAF mutations in PTCs have been described by many authors, including our group (Matos et al., 2024). Yang & Choi (2020) showed that miR-146b expression was significantly associated with the presence of BRAF mutations. Lee et al. (2014), in a study of 100 PTCs patients, showed that tumors where the BRAF V600E mutation was detected concomitantly with miR-146b overexpression had shorter patient survival. In another study, upregulation of miR-146b was associated with tumor aggressiveness and poor clinicopathological features, such as extrathyroid and capsule invasion, and the presence of lymph nodes or distant metastases (Sistrunk et al., 2019). Sun et al. (2013) showed that miR-221, miR-222 and miR-146b, together with miR-181, were upregulated in PTCs patients with BRAF mutations. These findings suggest a connection or influence of the presence of BRAF mutations with miRNAs expression in the development of TC.

Future prospects

During the development of this review, we added evidence that molecular markers can be of significant value to daily clinical practice. For that to be realized, however, additional studies and developments are needed to support that practice with robust indicators. The implementation of prospective studies of biomarkers in cytology and histology samples, with larger sample sizes and adapted to different populations, is mandatory to achieve a comprehensive understanding of molecular biomarkers and thyroid tumor features and behavior. The improvement of new techniques, such as liquid biopsy, a non-invasive method that detects and analyzes biological molecules released from the tumor into the bloodstream, is crucial. Different technologies can identify circulating tumor cells, circulating free nucleic acids (cDNA, cfRNA, and miRNA), and tumor-derived extracellular vesicles (EVs) in the serum of cancer patients (Alix-Panabières & Pantel, 2021), thus providing valuable molecular information (van Dessel et al., 2018). Circulating free-DNA and circulating miRNAs are more stable compared to cfRNA, and easier to investigate (Pös et al., 2018; Vitale et al., 2021). A set of miRNAs (e.g., miRNA-375, 34a, 145b, 221, 222, 155, Let-7, and 181b) have been evaluated in serum and proposed as diagnostic biomarker to distinguish PTCs from benign nodules and identify cancer at an early stage (Mahmoudian-Sani et al., 2017). In a prospective observational study (Rosignolo et al., 2017), the serum levels of 754 miRNAs were measured in 11 PTC patients before and after surgery. Among these, miRNA-146 and miRNA-221 were validated as serum tumor biomarkers during post-surgical follow-up, showing a significant correlation with disease recurrence, even in patients with low thyroglobulin levels. In fact, the use of liquid biopsy in thyroid cancer can have a significant role, as it can contribute to accurate patients diagnosis, prognosis, and monitoring tumor recurrence or treatment response (Romano et al., 2021). However, the use of liquid biopsy in TC presents several challenges, mainly depending on the sensitivity and specificity of the different methods, and tumor subtypes and stages. Other technical limitations in the use of liquid biopsy relate to the quantity of circulating material retrieved, especially in the context of early-stage disease, and issues of standardization, reproducibility, and validity (Geeurickx & Hendrix, 2020). Artificial Intelligence (AI) may be the next step in better understanding molecular biomarkers in thyroid nodular disease by integrating clinical and imaging features with cyto-histological and molecular diagnosis (Bini et al., 2021). Interest in the application of AI in thyroidology began in the 1990s, to improve the care of patients with thyroid nodules and cancer, as well as functional and autoimmune thyroid disease (Toro-Tobon et al., 2023). Ludwig et al. (2023), suggest its application at various stages in the diagnosis and treatment of thyroid nodules. Application of AI in the diagnostic of thyroid disease includes thyroid ultrasound (Cao et al., 2023), cytopathology (Wang et al., 2022) or histopathology studies (Siller et al., 2022), and nuclear medicine techniques such as thyroid scintigraphy (Qiao et al., 2021; Currie & Iqbal, 2022) or SPECT (single-photon emission computed tomography) (Ma et al., 2019). AI models may also be used in surgery for preoperative and intraoperative decision-making (Madani & Feldman, 2019).

Conclusions

The advances in the knowledge of molecular biomarkers for the diagnosis and prognosis of thyroid nodules and cancer reinforce the potential clinical utility of molecular testing in cytological assessments using US-FNA or blood. These advances may anticipate the genetic and molecular profiles of the tumors and their biological behavior. However, the complexity of biomarkers, the lack of consensus on the best panels to use (aiming for better risk stratification), the need for high technology and specialized laboratories, and the high costs not supported by national health care systems in many countries determine that the implementation of biomarkers in clinical practice is not yet a reality. More studies of molecular biomarkers are needed to achieve an international consensus on their applicability for diagnosis, risk stratification and prognosis of thyroid nodules, as well as for their introduction in routine clinical practice. The improvement of new techniques, such as liquid biopsy, will be complementary to tissue biopsy but is still not yet applicable in clinical practice. AI can bring significant benefits for patients, including the reduction of unnecessary FNACs, better surgery options, and reduced costs. However, despite the legal responsibility issues associated with the use of AI, more studies are needed to refine its accuracy in distinguishing between benign and malignant thyroid tumors. By integrating molecular knowledge with clinical evaluation and image features in the work flow of thyroid nodules diagnosis, clinicians may improve the accuracy of their decisions, reducing the number of FNACs, particularly in indeterminate nodules, and may optimize treatment strategies, from surgical options to active surveillance.