Ultrasound-based radiomics for prediction of lateral neck lymph node metastasis in patients with medullary thyroid cancer

Introduction

Medullary thyroid cancer (MTC) originates from thyroid C cells and accounts for <5% of thyroid malignancies (1). However, it is responsible for up to 8.6% of all deaths related to thyroid cancer (2). MTC can be classified into sporadic MTC and hereditary MTC based on genetic characteristics, with the latter accounting for 25% of cases (3).

MTC is prone to early lymph node metastasis, with approximately 80% of patients experiencing metastasis to the central compartment (levels VI and VII) or lateral neck compartments (levels II–V) based on previous research findings (4). MTC is not sensitive to chemotherapy, and complete surgical removal of all neoplastic tissue is the only curative approach. International guidelines have established total thyroidectomy (TT) and central lymph node dissection (CND) as the standard treatment for MTC (5,6). However, there is still controversy over whether prophylactic lateral neck lymph node dissection (LLND) should be performed in patients with no preoperative evidence of lateral neck lymph node metastasis (LLNM), due to a lack of universally accepted standards (6).

Preoperative ultrasound (US) is the pivotal tool in the risk stratification of thyroid nodules. Previous studies reported that the malignant-looking US features of MTC may be associated with a poor prognosis (7,8). Studies reported that large size, irregular shape, spiculated margin, microcalcification, and subcapsular location on US were associated with high risk of LLNM in patients with MTC (8-10).

Radiomics is a rapidly evolving field of research. Radiomics generally aims to extract quantitative and ideally reproducible information from diagnostic images, including complex patterns that are difficult to recognize or quantify by the human eye. Many studies applied it to predict cervical lymph node metastasis in papillary thyroid cancer and achieved good prediction results (11-13). Currently, few studies are on the use of US radiomics to predict LLNM in patients with MTC.

Therefore, we analyzed the clinical and US features and extracted radiomics features from US images, aiming to develop a novel model to predict the risk of LLNM in patients with MTC. We present this article in accordance with the TRIPOD reporting checklist (available at https://gs.amegroups.com/article/view/10.21037/gs-2025-aw-473/rc).

Methods

Patients

The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the Medical Ethics Committee of Chinese PLA General Hospital (No. S2023-752-01) and individual consent for this retrospective analysis was waived. In this retrospective study, the preoperative clinical, US and radiomics features were analyzed to predict LLNM in patients with MTC. For this purpose, the patients were classified into two groups: LLNM(+) group, i.e., N1b patients in whom LLNM was identified at initial surgery; and LLNM(−) group, i.e., N0-N1a patients with undetectable calcitonin (Ctn) within the reference range and without any lymph node metastasis or recurrence during the median 62.3 months [interquartile range (IQR), 38.9–84.1 months] of follow-up after initial treatment. All patients underwent regular follow-ups for postoperative evaluation of the disease status with physical examination, serum Ctn measurements, and cervical US at 6- to 12-month intervals.

We retrospectively reviewed a consecutive series of 113 MTC patients over 18 years old, all having undergone TT with CND between January 2014 and June 2023 and all having US examinations performed prior to surgery at The First Medical Center, Chinese PLA General Hospital. All patients were followed up for at least 6 months. Two patients with initial N0–N1a disease who experienced recurrence after initial surgery were excluded. Fifteen patients were excluded because of US images of unsatisfactory quality. In addition, seven patients who had incomplete clinical data were also excluded. Finally, 89 patients were included. Of them, 58 patients underwent LLND during the initial surgery. Among the subjects, 54 patients (60.7%) were categorized as the LLNM(−) group and 35 patients (39.3%) were categorized as the LLNM(+) group.

The following demographic and clinical information were included: age, sex (male or female) and preoperative serum Ctn (pg/mL) and carcinoembryonic antigen (CEA, ng/mL).

US features and development of the conventional model

US examinations were performed using iU 22 or iU Elite (Philips Medical Systems, Bothell, WA, USA) or LOGIQ E9 (GE Healthcare, Wauwatosa, WI, USA), Acuson S2000 (Siemens Healthineers, Mountain View, CA, USA), Resona 7 (Mindray, Shenzhen, China), or VINNO 70 (Vinno Technology, Suzhou, China) instruments equipped with 4.0–15.0 MHz linear array transducers. Two senior readers, each with over five years of experience in thyroid imaging, independently analyzed the image features while being blinded to the clinical information, radiologic report, and postoperative pathology. Disagreements between two senior readers were resolved by a third expert with more than 20 years of experience in thyroid imaging. US features were analyzed according to the standard methodology from previously published reports (10,14,15). Nodule size was determined by measuring the maximum diameter of the MTC on US. The dimensions of the largest MTC lesion were utilized in patients with multifocal MTCs. Nodules with small cystic components, occupying less than 5% of the total volume, should be classified as solid nodules. Nodule echogenicity can be categorized as markedly hypoechoic, hypoechoic, isoechoic, and hyperechoic. Markedly hypoechoic nodules have echogenicity lower than the adjacent neck muscles, while hypoechoic nodules have echogenicity lower than the thyroid gland. The shape was defined as having an ovoid to round or taller than wide. The margin could be categorized as ill-defined, smooth, lobulated or irregular (specifically defined as protrusions into adjacent tissue, jagged, spiculated, or sharp angles). Calcifications were also evaluated. The subcapsular location was identified when a nodule was found adjacent to the thyroid capsule without any intervening thyroid parenchyma. Vascularity was evaluated by color Doppler flow imaging, which was classified into types 1–4 (type 1, no vascularity; type 2, perinodular vascularity only, i.e., circumferential vascularity at the nodule margin; type 3, mild intranodular vascularity with or without perinodular vascularity; type 4, marked intranodular vascularity with or without perinodular vascularity) (15). The suspicious lateral lymph nodes detected by US (USLN) were determined based on the results of original US report. In our hospital, USLN were defined based on any of the following suspicious features: focal or diffuse hyperechogenicity, cystic degeneration, calcification, or peripheral vascularity.

Clinical and US features that were significant in the univariate analysis (either Student’s independent test or chi-square test) were selected into the multivariate analysis, and the odds ratios (ORs) with 95% confidence intervals (CIs) were calculated. Significant features in the multivariate analysis or clinically relevant ones by experts were used to develop the conventional model. The receiver operating characteristic (ROC) curve analysis was used to determine the optimal cut-off points for patient age, tumor size, values of Ctn and CEA for the prediction of LLNM.

National Comprehensive Cancer Network (NCCN) and European Society for Medical Oncology (ESMO) assessment

The NCCN suggests that preoperative US detection of LLNM, or a tumor size >1 cm, or for patients with central lymph node metastasis, should be considered as indications for preventive LLND (16). ESMO suggests that patients with negative US results should be decided whether to undergo LLND based on their Ctn levels (5). Ipsilateral LLND is performed when Ctn is between 50 and 200 pg/mL, and bilateral LLND is performed when it exceeds 200 pg/mL. We evaluated the collected patients according to these guidelines and determined whether they should undergo LLND.

Image segmentation and radiomics feature extraction

The radiomics flowchart is demonstrated in Figure 1. The ITK-SNAP 3.8 software (https://www.itksnap.org/) was used to manually segment the thyroid nodules as regions of interest (ROIs) in both transverse and longitudinal sections. Radiomics features were extracted from ROI using PyRadiomics. Two operators were trained to segment ROIs before the study began. Intra-observer and inter-observer consistency were evaluated by a random group of 30 nodules segmented by the operators. One month after the first segmentation, two operators re-segmented the same group of images. The intraclass correlation coefficient (ICC) was used to assess the reproducibility and robustness of lesion segmentation and feature extraction. An ICC greater than 0.75 suggested a good agreement for the feature extraction.

Figure 1 The radiomics flowchart. We collected ultrasound images from patients with medullary thyroid cancer, and then regions of interest were segmented and radiomics features were extracted and selected. The selected features were integrated with clinical and ultrasound characteristics to develop the nomogram model. The model underwent subsequent validation and interpretation. LASSO, least absolute shrinkage and selection operator; ROC, receiver operating characteristic.

LLNM-related feature selection and radiomics model development

All radiomics features underwent Z-score transformation for standardization to enhance data comparability and reduce bias. Only features with good agreement (ICC >0.75) were included in the analysis. Univariate analysis was performed to identify features with P<0.05 for further study. Subsequently, the least absolute shrinkage and selection operator (LASSO) method was employed to select radiomics features with excellent prediction capabilities. The logistic regression method was used to develop the radiomics score (rad score) model to predict LLNM based on these features.

Development of the nomogram model

The nomogram model was established based on the clinical and US features used in the conventional model and the rad score model, thereby integrating clinical, US, and radiomics features.

Performance and interpretation of the models

The internal validation was performed using 1,000 bootstrap samplings, and the ROC curves were plotted to calculate the area under the ROC curve (AUC) for evaluating performance of the models. The diagnostic performance was classified as poor (AUC 0.5–<0.7), moderate (AUC 0.7–<0.9), and excellent (AUC ≥0.9). The model with better performance was selected as the final model. Calibration curves and decision curve analysis (DCA) curves were plotted to evaluate the performance and patient benefits. Furthermore, we interpreted the model using SHapley Additive exPlanations (SHAP) method.

Statistical analysis

Categorical variables were presented as frequencies and percentages, while continuous variables were described using mean ± standard deviation, or median (IQR) when required. The statistical analysis and development of prediction models were conducted using R software (version 4.3.1, https://www.r-project.org) and Python software (version 3.8). All the statistical significance levels were two-sided, with the P value less than 0.05.

Results

Clinical and US features

A total of 89 patients were included in the study, and all of them had a definitive pathological diagnosis of MTC after surgery. The median age was 47.0 (37.0, 59.0) years, with males accounting for 51.7%. The median maximum tumor diameter was 1.5 (0.9, 2.6) cm. LLNM occurred in 39.3% of the MTC patients. There were no significant differences in age and gender between patients with and without LLNM (Table 1). Large size and high levels of Ctn and CEA were more likely to be associated with LLNM. By ROC analysis, the optimal thresholds for tumor size, Ctn, and CEA were 1.0 cm, 539.7 pg/mL, and 12.6 ng/mL with their AUC of 0.635, 0.659, and 0.664, respectively. Subcapsular tumors were more common in patients with LLNM (Table 1). Type 3 blood flow signals were more frequently observed in patients with LLNM. Other US features were not significantly different between patients with and without LLNM. By preoperative US, LLNMs were missed in 37.1% of patients, while overdiagnosis occurred in 18.5%.

Radiomics features

A total of 1,288 radiomics features were extracted. A total of 1,147 radiomics features were retained due to their good repeatability (ICC >0.75). Twenty-three features selected by univariate analysis were included in the LASSO analysis. A penalty coefficient (λ=0.08) was determined, and five radiomics features with non-zero coefficients were selected to develop the rad score model (Figure 2).

Figure 2 Ultrasound radiomics features selection by the LASSO method. (A) The selection process of the optimum values of the parameters in the LASSO by 10-fold cross-validation method. (B) LASSO coefficient profiles of the radiomics features. (C) The five features with non-zero coefficients selected through LASSO. LASSO, least absolute shrinkage and selection operator.

Construction and evaluation of prediction models

The conventional model was established using Ctn values, subcapsular location, and USLN. Although Ctn values did not show a significant difference (P=0.07) in the multivariate analysis, we still incorporated it into the conventional model based on expert opinions and previous studies. The rad score model was constructed using the five selected radiomics features. The logistic regression method was used to build the nomogram model by integrating the rad score as a continuous variable and the significant clinical and US features (Figure 3).

Figure 3 Nomogram model to predict the LLNM in patients with medullary thyroid cancer. The nomogram is used by summing all points identified on the scale for each variable. The total points projected on the bottom scales indicate the probabilities of LLNM. Ctn, calcitonin; LLNM, lateral neck lymph node metastasis; rad score, radiomics score; USLN, suspicious lateral lymph nodes detected by ultrasound.

In the 1,000 bootstrap internal validations, the average AUC for the conventional model was 0.864 (95% CI: 0.863–0.864), 0.862 (95% CI: 0.861–0.863) for the rad score model, and 0.949 (95% CI: 0.949–0.950) for the nomogram model (Figure 4A). The average AUC of the nomogram model was significantly higher than the conventional and rad score models (P=0.008, 0.01, respectively). The calibration curve of nomogram model showed good agreements between the observed and predicted results (Figure 4B). The DCA suggested that the nomogram model conferred benefits to patients when the risk threshold was between 0.1 and 1.0 (Figure 4C).

Figure 4 Performance and SHAP values of models. (A) ROC curves of conventional model, rad score model, nomogram model, ESMO, and NCCN. (B) Calibration curves of nomogram model. (C) DCA curves of nomogram model. (D) Features included in the nomogram model and ranked by SHAP values, which capture the influence of each feature within the model. Each row represents a feature with dots indicating individual patient contributions to the outcome: red for higher risk and blue for lower risk. Ctn, calcitonin; DCA, decision curve analysis; ESMO, European Society for Medical Oncology; LLNM, lateral neck lymph node metastasis; NCCN, National Comprehensive Cancer Network; rad, radiomics; rad score, radiomics score; ROC, receiver operating characteristic; SHAP, SHapley Additive exPlanations; USLN, suspicious lateral lymph nodes detected by ultrasound.

We further compared the results with the two clinical guidelines. According to the ESMO, 91.00% of patients were expected to undergo LLND, with a predicted AUC of 0.623. According to the NCCN, 70.01% of patients were expected to undergo LLND, with a predicted AUC of 0.553. The nomogram model constructed based on clinical, US, and radiomics features outperformed the ESMO and NCCN suggestions, with a higher AUC (P<0.001) (Figure 4A).

Interpretability analysis

We used SHAP to illustrate the selected features and their impact on predicting LLNM in the nomogram model. The rad score model, subcapsular location, and USLN were found to be associated with a higher prediction probability of LLNM (Figure 4D).

We selected one case for individual explanation. The patient was a 68-year-old male with Ctn over 2,000 pg/mL who had no suspicious lymph nodes detected on the preoperative US. However, postoperative pathology revealed central and lateral neck lymph node metastases. Our model indicated a high risk of LLNM due to the elevated Ctn level, nodule’s subcapsular location, and high rad score (Figure 5).

Figure 5 A 68-year-old male patient with no suspicious lateral lymph nodes detected on the preoperative ultrasound. However, our model indicated a high risk of LLNM due to the elevated calcitonin level, nodule's subcapsular location, and high rad score. Ctn, serum calcitonin; LLNM, lateral neck lymph node metastasis; rad, radiomics; USLN, suspicious lateral lymph nodes detected by ultrasound.

Discussion

MTC is a relatively rare neuroendocrine cancer (3). However, it exhibits a higher rate of lymph node metastasis and mortality compared with well-differentiated thyroid cancer (17,18). Previous studies reported approximately 40% for LLNM in patients with MTC and its negative impact on prognosis (19). Therefore, the preoperative prediction of LLNM is significant and could inform the surgical extent. In this study, three clinical and US features were identified for the construction of conventional model: elevated preoperative Ctn, subcapsular location on US, and preoperative US detected suspicious lateral lymph node. This study might be the first study to use radiomics features to predict LLNM in patients with MTC. Five radiomics features were selected to develop the rad score model. A nomogram model was then established by combining these two models via a fusion method, which achieved satisfactory prediction performance.

Previous studies indicated that old age and male gender might be associated with worse prognosis in MTC patients (20-22). However, there were no statistical differences in the age and gender between the LLNM(−) and LLNM(+) groups in this study. These results were consistent with the findings of OH and Luo et al. (10,23). Elevated Ctn levels were associated with LLNM, which was consistent with prior studies (24-26). However, upon interpreting our model through SHAP values, we found that a single increase in Ctn level contributed less to model prediction compared to other predictors. Integrating Ctn levels with other clinical and US features was anticipated to enhance the prediction accuracy for LLNM.

Although previous studies found that large size was more likely to be associated with poor prognosis and LLNM (10,23,27), our study failed to find an association between them. According to SHAP values, subcapsular location on preoperative US contributed significantly to the model, which suggested a close association between this indicator and LLNM. Proximity to the capsule may be closely related to extrathyroidal extension of the tumor. Previous studies reported that extrathyroidal extension in MTC was closely linked with poor prognosis and occurrence of LLNM (27-29). Therefore, for subcapsular nodules on US, a more aggressive surgical approach might be suggested.

To the best of our knowledge, this might be the first study to evaluate the use of US radiomics for predicting LLNM in patients with MTC. By integrating radiomics features, our nomogram model demonstrated better prediction performance compared to the conventional model based on clinical and US features. DCA indicated that the nomogram model yielded maximum benefits across a wide range of threshold probabilities, suggesting its superior clinical applicability and practicality.

Some researchers argue for routine LLND in MTC patients due to the inability of these tumors to take up iodine, rendering radioactive iodine ablation ineffective against lymph node metastases (17). Meticulous lymph node dissection has been reported to result in a higher biochemical cure rate. Surgical clearance through lymph node dissection is considered the only effective strategy for eliminating these deposits. However, other scholars contend that the dissection of lymph nodes may not improve overall survival in patients with MTC (10,30). While LLND is generally safe when performed by experienced surgeons, there is a potential for increased morbidity (2,31). Complications include, but are not limited to injury to important nerves like the spinal accessory nerve, and lymphatic leakage during the procedure (32). Our model could be used to predict the LLNM, and the preoperative prediction of LLNM could inform the extent of surgery, leading to personalized and improved patient care.

There are some limitations in this study. First, this study was of retrospective design, which could lead to biased results in the model. Second, as the training set only included individuals from the Asian population, its applicability to other regions may be limited. Third, due to the low incidence of MTC leading to a small sample size, we did not set up an independent external validation in order not to waste valuable training data. Although 1,000 bootstrap internal validations provided a good assessment of model generalizability, a large-scale, multicenter external validation is needed.

Conclusions

A nomogram model was established by combining clinical, US, and radiomics features, and achieved satisfactory performance for the prediction of LLNM in patients with MTC.

Acknowledgments

None.

Footnote

Reporting Checklist: The authors have completed the TRIPOD reporting checklist. Available at https://gs.amegroups.com/article/view/10.21037/gs-2025-aw-473/rc

Data Sharing Statement: Available at https://gs.amegroups.com/article/view/10.21037/gs-2025-aw-473/dss

Peer Review File: Available at https://gs.amegroups.com/article/view/10.21037/gs-2025-aw-473/prf

Funding: This research was supported by the Shandong Provincial Medical and Health Science and Technology Projects (No. 202409020788).

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://gs.amegroups.com/article/view/10.21037/gs-2025-aw-473/coif). The authors have no conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the Medical Ethics Committee of Chinese PLA General Hospital (No. S2023-752-01) and individual consent for this retrospective analysis was waived.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Gild ML, Clifton-Bligh RJ, Wirth LJ, et al. Medullary Thyroid Cancer: Updates and Challenges. Endocr Rev 2023;44:934-46. [Crossref] [PubMed]
2. Zhang J, Dionigi G, Kim HY, et al. Dynamic risk stratification-guided management of medullary thyroid carcinoma: integrating surgical precision with RET-targeted therapies and molecular surveillance. Int J Surg 2025;111:7072-86. [Crossref] [PubMed]
3. Fugazzola L. Medullary thyroid cancer - An update. Best Pract Res Clin Endocrinol Metab 2023;37:101655. [Crossref] [PubMed]
4. Azar FK, Lee SL, Rosen JE. Medullary thyroid cancer: an update for surgeons. Am Surg 2015;81:1-8.
5. Filetti S, Durante C, Hartl D, et al. Thyroid cancer: ESMO Clinical Practice Guidelines for diagnosis, treatment and follow-up†. Ann Oncol 2019;30:1856-83. [Crossref] [PubMed]
6. Valderrabano P, Klippenstein DL, Tourtelot JB, et al. New American Thyroid Association Sonographic Patterns for Thyroid Nodules Perform Well in Medullary Thyroid Carcinoma: Institutional Experience, Systematic Review, and Meta-Analysis. Thyroid 2016;26:1093-100. [Crossref] [PubMed]
7. Zhao J, Yang F, Wei X, et al. Ultrasound features value in the diagnosis and prognosis of medullary thyroid carcinoma. Endocrine 2021;72:727-34. [Crossref] [PubMed]
8. Ning CP, Kim EK. Sonographic predictors of aggressive behavior in medullary thyroid carcinomas. Asian J Surg 2022;45:291-8. [Crossref] [PubMed]
9. Yun G, Kim YK, Choi SI, et al. Medullary thyroid carcinoma: Application of Thyroid Imaging Reporting and Data System (TI-RADS) Classification. Endocrine 2018;61:285-92. [Crossref] [PubMed]
10. Oh HS, Kwon H, Song E, et al. Preoperative Clinical and Sonographic Predictors for Lateral Cervical Lymph Node Metastases in Sporadic Medullary Thyroid Carcinoma. Thyroid 2018;28:362-8. [Crossref] [PubMed]
11. Liu W, Zhang S, Shi J, et al. A multi-task deep learning framework for intraoperative diagnosis of thyroid cancer metastasis using whole slide images. Int J Med Inform 2026;206:106176. [Crossref] [PubMed]
12. Liu Y, Lai F, Lin B, et al. Deep learning to predict cervical lymph node metastasis from intraoperative frozen section of tumour in papillary thyroid carcinoma: a multicentre diagnostic study. EClinicalMedicine 2023;60:102007. [Crossref] [PubMed]
13. Dong L, Han X, Yu P, et al. CT Radiomics-Based Nomogram for Predicting the Lateral Neck Lymph Node Metastasis in Papillary Thyroid Carcinoma: A Prospective Multicenter Study. Acad Radiol 2023;30:3032-46. [Crossref] [PubMed]
14. Tessler FN, Middleton WD, Grant EG. Thyroid Imaging Reporting and Data System (TI-RADS): A User's Guide. Radiology 2018;287:29-36. [Crossref] [PubMed]
15. Ha EJ, Chung SR, Na DG, et al. 2021 Korean Thyroid Imaging Reporting and Data System and Imaging-Based Management of Thyroid Nodules: Korean Society of Thyroid Radiology Consensus Statement and Recommendations. Korean J Radiol 2021;22:2094-123. [Crossref] [PubMed]
16. Haddad RI, Bischoff L, Ball D, et al. Thyroid Carcinoma, Version 2.2022, NCCN Clinical Practice Guidelines in Oncology. J Natl Compr Canc Netw 2022;20:925-51. [Crossref] [PubMed]
17. Jin LX, Moley JF. Surgery for lymph node metastases of medullary thyroid carcinoma: A review. Cancer 2016;122:358-66. [Crossref] [PubMed]
18. Pelizzo MR, Mazza EI, Mian C, et al. Medullary thyroid carcinoma. Expert Rev Anticancer Ther 2023;23:943-57. [Crossref] [PubMed]
19. Zhou TH, Zhao LQ, Zhang Y, et al. The Prediction of Metastases of Lateral Cervical Lymph Node in Medullary Thyroid Carcinoma. Front Endocrinol (Lausanne) 2021;12:741289. [Crossref] [PubMed]
20. Costa CS, Souteiro P, Paredes S, et al. Male gender as a poor prognostic factor in medullary thyroid carcinoma: behavior or biological difference? Minerva Endocrinol (Torino) 2023;48:395-400. [Crossref] [PubMed]
21. Abou Azar S, Tobias J, Applewhite M, et al. Medullary Thyroid Cancer: Single Institute Experience Over 3 Decades and Risk Factors for Recurrence. J Clin Endocrinol Metab 2024;109:2729-34. [Crossref] [PubMed]
22. Oh MY, Jung KY, Choi H, et al. Distinct Impacts of Clinicopathological and Mutational Profiles on Long-Term Survival and Recurrence in Medullary Thyroid Carcinoma. Endocrinol Metab (Seoul) 2024;39:877-90. [Crossref] [PubMed]
23. Luo Z, Hong Y, Yan C, et al. Nomogram for preoperative estimation risk of cervical lymph node metastasis in medullary thyroid carcinoma. Front Oncol 2022;12:883429. [Crossref] [PubMed]
24. Asarkar A, Chang BA, Nathan CO. What Is the Extent of Neck Dissection in Medullary Thyroid Carcinoma? Laryngoscope 2021;131:458-9. [Crossref] [PubMed]
25. Du Y, Shen C, Song K, et al. Updated Thresholds of Basal Calcitonin Level and Extent of Lymph Node Metastasis in Initially Treated Medullary Thyroid Cancer. JAMA Otolaryngol Head Neck Surg 2025;151:761-7. [Crossref] [PubMed]
26. Lin X, Huo J, Su H, et al. Risk factors for cervical lymph node metastasis in the central or lateral cervical region in medullary thyroid carcinoma: a systematic review and meta-analysis. Eur Arch Otorhinolaryngol 2024;281:547-61. [Crossref] [PubMed]
27. Le MK, Kawai M, Odate T, et al. Metastatic Risk Stratification of 2526 Medullary Thyroid Carcinoma Patients: A Study Based on Surveillance, Epidemiology, and End Results Database. Endocr Pathol 2022;33:348-58. [Crossref] [PubMed]
28. Moura MM, Cabrera RA, Esteves S, et al. Correlation of molecular data with histopathological and clinical features in a series of 66 patients with medullary thyroid carcinoma. J Endocrinol Invest 2021;44:1837-46. [Crossref] [PubMed]
29. Kotwal A, Erickson D, Geske JR, et al. Predicting Outcomes in Sporadic and Hereditary Medullary Thyroid Carcinoma over Two Decades. Thyroid 2021;31:616-26. [Crossref] [PubMed]
30. Ansley W, Kamyab A, Noden L, et al. Does the extent of neck surgery based on preoperative calcitonin level influence survival in medullary thyroid carcinoma: a retrospective tertiary centre experience. Ann R Coll Surg Engl 2025;107:199-205. [Crossref] [PubMed]
31. Khokhar M, Milas M. Management of Nodal Disease in Thyroid Cancer. Surg Clin North Am 2019;99:611-32. [Crossref] [PubMed]
32. Yu J, Deng Y, Liu T, et al. Lymph node metastasis prediction of papillary thyroid carcinoma based on transfer learning radiomics. Nat Commun 2020;11:4807. [Crossref] [PubMed]