Preoperative prediction of extrathyroidal extension in papillary thyroid carcinoma: a nomogram integrating dual-plane ultrasound-based radiomics and clinical features

Introduction

Papillary thyroid carcinoma (PTC) is the most common type of thyroid malignancy, accounting for approximately 85–90% of all cases, and its incidence has increased steadily worldwide over recent decades (1). Although most PTCs have a favorable prognosis, a subset of patients develops extrathyroidal extension (ETE), which is associated with aggressive tumor behavior (2). According to the 8^th edition of the American Joint Committee on Cancer (AJCC) staging system, ETE is defined as the direct invasion of tumor beyond the thyroid capsule into perithyroidal soft tissues, strap muscles, or adjacent structures including the trachea, esophagus, and recurrent laryngeal nerve (3). Studies have demonstrated that ETE is associated with increased risks of cervical lymph node metastasis, distant metastasis, local recurrence, and reduced disease-specific survival, and is considered an independent prognostic factor in clinical staging (2,4,5). Patients without ETE may undergo thyroid lobectomy alone, whereas those with ETE typically require total thyroidectomy with or without central compartment lymph node dissection, which is associated with higher perioperative risks including hypoparathyroidism and recurrent laryngeal nerve injury (6).

High-frequency ultrasound is the preferred imaging method for preoperative evaluation of thyroid nodules and is recommended by the 2025 American Thyroid Association (ATA) guidelines for ETE assessment (6). However, the sensitivity of conventional ultrasound in detecting ETE remains unsatisfactory, with reported sensitivities as low as 30% (7,8). This is mainly because sonographic signs of ETE are often subtle and diagnostic accuracy relies heavily on operator experience, making some cases of ETE difficult to recognize with the naked eye (7,9). This has direct clinical consequences: underestimation of ETE may result in insufficient surgical resection, increasing the risk of reoperation and disease recurrence, whereas overestimation may lead to unnecessarily aggressive procedures (6). Therefore, an objective and quantitative preoperative tool is needed to improve ETE identification and guide individualized surgical planning.

Radiomics is a high-throughput technique that extracts numerous quantitative features from medical images, capturing texture, morphology, and statistical characteristics that are not visible to the naked eye (10). By converting imaging data into analyzable quantitative features combined with machine learning algorithms, radiomics has shown promising performance in the diagnosis and prognosis of various malignancies, including thyroid cancer (11,12). Several studies have explored radiomics-based models for the preoperative prediction of ETE and reported promising results (13). Nevertheless, current studies have several limitations: most used single-plane ultrasound images rather than dual-plane (transverse and longitudinal) imaging, and constructed models based solely on radiomics features without integrating clinical factors, which restricts clinical applicability (13-15). To address these limitations, this study aimed to establish a nomogram integrating dual-plane ultrasound radiomics and clinical features for the preoperative prediction of ETE in PTC, providing an objective and reproducible tool for individualized clinical decision-making. We present this article in accordance with the TRIPOD reporting checklist (available at https://gs.amegroups.com/article/view/10.21037/gs-2026-0304/rc).

Methods

The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. This retrospective study was approved by the Institutional Ethics Committee of Beijing Friendship Hospital, Capital Medical University (No. 2022-P2-346-01). Written informed consent was waived due to the retrospective design.

Patients

Consecutive patients with pathologically confirmed PTC who underwent thyroidectomy at Beijing Friendship Hospital, Capital Medical University between January 2020 and June 2024 were enrolled. The enrolment procedure is shown in Figure 1.

Figure 1 Flowchart of patient enrolment in this study. ETE, extrathyroidal extension; PTC, papillary thyroid carcinoma.

Inclusion criteria

(I) Patients with PTC confirmed by postoperative pathological diagnosis; (II) high-quality preoperative ultrasound images in transverse and longitudinal planes stored in Digital Imaging and Communications in Medicine (DICOM) format; (III) complete clinical and laboratory information.

Exclusion criteria

(I) Patients with history of thyroid surgery or ablation therapy; (II) patients with history of previous head and neck malignancies, surgery or radiation exposure; (III) non-PTC (follicular, medullary, anaplastic, or metastatic carcinoma); (IV) poor ultrasound image quality unsuitable for accurate ultrasound evaluation and region of interest (ROI) segmentation; (V) incomplete clinical or laboratory information.

Clinical information collection

The clinicopathological data were collected from preoperative ultrasound examinations, laboratory records, and postoperative pathological reports. Demographic and clinical variables included gender, age, serum thyroid-stimulating hormone (TSH), and anti-thyroglobulin antibody (anti-TgAb). Postoperative pathological reports provided confirmation of PTC diagnosis and ETE status. ETE, histologically confirmed on surgical specimens as tumor invasion beyond the thyroid capsule into perithyroidal soft tissue or adjacent structures, was the primary outcome variable (3). Patients who underwent thyroidectomy between January 2020 and December 2023 formed the training cohort, and those who underwent thyroidectomy between January 2024 and June 2024 served as the independent validation cohort.

Ultrasound image acquisition and feature assessment

Preoperative ultrasound examinations were performed using a broadband linear array transducer (5–12 MHz; iU Elite, Philips Medical Systems, Bothell, WA, USA) with the patient in the supine position and the neck adequately extended. The thyroid gland and bilateral cervical lymph nodes were systematically evaluated. For each nodule, the following features were recorded: maximum diameter, margin, calcification, aspect ratio, multifocality (defined as ≥2 thyroid lesions), suspicious lymph nodes, nodule-to-capsule contact and American College of Radiology Thyroid Imaging Reporting and Data System (ACR TI-RADS) category.

Tumor size was measured as the long-axis and short-axis diameters on the maximal transverse and longitudinal planes of the lesions, and ultrasound images of the lesions in both transverse and longitudinal planes were stored for subsequent analysis. Calcifications in the lesions included microcalcifications and macrocalcifications. All ultrasound images of the lesions were evaluated independently by two radiologists with at least 5 years of diagnostic experience in thyroid ultrasound who were blinded to the pathological results of ETE. If there was any disagreement, the two radiologists would make a decision by consensus.

To assess reproducibility, radiomics features were extracted twice by two independent observers in a randomly selected subset of 30 patients. Features with intraclass correlation coefficient (ICC) ≥0.75 for both interobserver and intraobserver agreement were retained for further analysis.

ROI segmentation and radiomics feature extraction

For each eligible nodule, ROIs were manually delineated along the tumor margin on both transverse and longitudinal planes by an experienced radiologist, with cystic, necrotic, and hemorrhagic components carefully excluded from segmentation. All segmentations were verified by a second senior radiologist. For patients with multiple nodules, the nodule with the highest ACR TI-RADS category was selected as the index lesion for further analysis.

Radiomics feature extraction was performed using PyRadiomics (version 3.0.1; https://pyradiomics.readthedocs.io) (16). Quantitative features were extracted from each plane, including shape descriptors, first-order statistics, and texture features derived from the gray-level co-occurrence matrix (GLCM), gray-level run length matrix (GLRLM), gray-level size zone matrix (GLSZM), gray-level dependence matrix (GLDM), and neighborhood gray tone difference matrix (NGTDM), as well as high-dimensional features derived from wavelet and Laplacian-of-Gaussian (LoG) filters at multiple scales.

Radiomics feature selection and Rad‑score construction

All features were subjected to Z-score normalization before proceeding with radiomics feature selection. A two-step feature selection procedure was applied. First, independent-samples t-test were performed to exclude features with no significant differences between the ETE-positive and ETE-negative groups (P>0.05). Second, least absolute shrinkage and selection operator (LASSO) logistic regression with five-fold cross-validation was used to identify the optimal subset of predictive features.

Three radiomics models were constructed using features derived from the transverse-, longitudinal-, and dual-plane ultrasound images, respectively. Rad-scores were generated from the optimal model identified among the three radiomics models, and calculated as a weighted linear combination of the selected features, with weights determined by LASSO regression. All analyses were performed using Python (version 3.9; https://www.python.org).

Clinical model and nomogram development

Clinical and ultrasound features with P<0.05 in univariate analysis were entered into multivariate logistic regression analysis with backward stepwise selection to identify independent predictors for ETE, and a clinical model was developed in the training cohort. The Rad-score was further incorporated with the independent clinical and ultrasound predictors into a multivariate logistic regression model to develop the combined model in the training cohort.

Subsequently, the predictive performance of this combined model was validated in the independent validation cohort. A nomogram for the preoperative prediction of ETE in PTC was established based on the combined model.

The Hosmer-Lemeshow goodness-of-fit test and calibration curves were generated to evaluate the calibration and predictive performance of the model in both the training and independent validation cohorts. Decision curve analysis (DCA) was performed to assess the clinical net benefit of the prediction models. Model performance was evaluated using the area under the receiver operating characteristic (ROC) curve (AUC), sensitivity, specificity, and accuracy.

Statistical analysis

Continuous variables were compared between groups using the Mann-Whitney U test for non-normally distributed data. Categorical variables were analyzed with the Chi-squared test or Fisher’s exact test. Univariate analysis was performed to identify candidate predictors associated with ETE. Multivariate logistic regression analysis was further conducted to identify independent predictors. DeLong’s test was used to compare the AUCs of different prediction models. All statistical analyses were performed using Python (version 3.9; https://www.python.org). A two-sided P<0.05 was considered statistically significant.

Results

Clinical characteristics

The study flowchart is shown in Figure 2. A total of 538 patients with pathologically confirmed PTC who underwent thyroidectomy between January 2020 and December 2023 were initially screened. After applying the exclusion criteria, 483 patients were enrolled as the training cohort (368 females and 115 males; median age 43.0 years). An independent validation cohort of 103 patients (85 females and 18 males; median age 44.0 years) was enrolled from 116 consecutive patients screened between January 2024 and June 2024 using the same exclusion criteria. The clinical, pathological, and ultrasonographic characteristics of both cohorts are summarized in Table 1. The rate of ETE was 47.4% (229/483) in the training cohort and 50.5% (52/103) in the validation cohort (P=0.57). No significant differences were observed between the two cohorts in any baseline characteristic, including TSH, TgAb, maximum diameter, aspect ratio, margin, calcification, multiple nodules, suspicious lymph nodes on ultrasound or nodule-to-capsule contact (all P>0.05), confirming comparability between cohorts.

Figure 2 The workflow of this study. LASSO, least absolute shrinkage and selection operator; long., longitudinal; ROC, receiver operating characteristic; ROI, region of interest; TgAb, thyroglobulin antibody; trans., transverse; TSH, thyroid-stimulating hormone.

Clinical and ultrasound feature analysis

Univariate analysis was performed to evaluate associations between clinical and ultrasound features and ETE status, with detailed results shown in Table 2. Maximum diameter and nodule-to-capsule contact were all significantly associated with ETE positivity (all P<0.001). On multivariate logistic regression, nodule-to-capsule contact [odds ratio (OR) =3.202; 95% confidence interval (CI): 1.909–5.829; P<0.001] and maximum diameter (OR =6.641; 95% CI: 4.201–11.584; P<0.001) were identified as independent predictors of ETE, as presented in Table 2.

Radiomics feature selection and Rad-score development

In this study, 1130 radiomics features were extracted from the transverse and longitudinal plane images of lesions, yielding 2260 radiomics features in total. After screening with independent-sample t-tests, 72 radiomics features with significant differences (P<0.05) were retained. Subsequent radiomics feature selection was performed using LASSO regression with five-fold cross-validation, and five features from the transverse plane and five features from the longitudinal plane were selected (Figure 3). All 10 selected radiomics features demonstrated excellent reproducibility (ICC ≥0.75).

Figure 3 Radiomics feature selection using LASSO regression. (A) Regularization path of coefficient trajectories as a function of −log₁₀(λ). (B) Five-fold cross-validation error curve; the dashed red line indicates the optimal λ. (C) Coefficients of the 10 selected radiomics features; blue and red bars indicate positive and negative coefficients, respectively. LASSO, least absolute shrinkage and selection operator.

Three radiomics models were constructed: the transverse model based on five transverse plane image features, the longitudinal model based on five longitudinal plane image features, and the dual-plane model incorporating all 10 features from both planes. The dual-plane model yielded the highest AUC (0.720; 95% CI: 0.671–0.769), outperforming both the transverse (AUC =0.648; 95% CI: 0.593–0.701) and longitudinal (AUC =0.630; 95% CI: 0.576–0.684) models (Table 3), and was therefore used to construct the final Rad-score. The Rad-score calculation formula was as follows:

Rad-score=−0.1153+0.6382×Trans_Wavelet-HLL_firstorder_Kurtosis+0.4100×Long_Wavelet-LLL_firstorder_Skewness+0.3673×Trans_LoG-sigma-2-0_NGTDM_Contrast+0.3447×Long_Wavelet-HLL_GLSZM_SmallAreaLowGrayLevelEmphasis+0.3303×Trans_Wavelet-HHL_firstorder_Mean+0.2976×Long_Wavelet-LLL_GLRLM_LowGrayLevelRunEmphasis+0.2819×Trans_Original_NGTDM_Complexity−0.2980×Long_Wavelet-LHH_firstorder_Mean−0.3090×Trans_Original_GLSZM_LowGrayLevelZoneEmphasis−0.4347×Long_LoG-sigma-1-0_GLSZM_LargeAreaHighGrayLevelEmphasis[1]

Model performance and nomogram development

In the independent validation cohort, the combined model outperformed both the radiomics and clinical models across all evaluation metrics (Table 4, Figure 4). It achieved the highest AUC of 0.858, which was significantly higher than the clinical model (AUC =0.763; P=0.03) and the radiomics model (AUC =0.704; P=0.02). The combined model achieved a sensitivity of 83.0% and a specificity of 80.6%, indicating a more balanced diagnostic performance compared with the other models.

Figure 4 ROC curves of the radiomics model, clinical model and combined model in the training (A) and independent validation (B) cohorts. AUC, area under the receiver operating characteristic curve; CI, confidence interval; ROC, receiver operating characteristic.

Based on the combined model, a nomogram integrating the Rad-score, maximum diameter, and nodule-to-capsule contact was established to facilitate individualized ETE risk prediction in PTC patients (Figure 5).

Figure 5 Nomogram for preoperative prediction of ETE incorporating the Rad-score, maximum diameter, and nodule-capsule contact. ETE, extrathyroidal extension.

Calibration curves in both the training and the independent validation cohorts demonstrated good consistency between predicted and actual probabilities, with the combined model exhibiting the best fit to the ideal line (Figure 6). The Hosmer-Lemeshow test confirmed satisfactory calibration for all models (all P>0.05).

Figure 6 Calibration curves and DCA of the three prediction models. (A,B) Calibration curves in the training and validation cohorts, respectively. (C,D) DCA in the training and validation cohorts, respectively. DCA, decision curve analysis.

DCA showed that the combined model provided the highest net benefit within a threshold range of 0.1–0.8 in both cohorts, surpassing both the radiomics and clinical models as well as the treat-all and treat-none strategies (Figure 6).

Discussion

ETE reflects the aggressive biological behavior of PTC and has direct implications for clinical decision-making. According to the 2025 ATA guidelines, preoperative neck ultrasound is recommended for all patients with PTC, whereas CT and MRI are reserved for cases with clinical suspicion of advanced or invasive disease (6). However, conventional ultrasound has limited ability to objectively assess ETE before surgery and evaluation is largely subjective and dependent on operator experience (7,8). Underestimation of ETE may result in insufficient surgical resection, increasing the risk of reoperation and disease recurrence, whereas overestimation may lead to unnecessarily aggressive procedures (6). To address this, we developed a combined nomogram incorporating the Rad-score, maximum diameter, and nodule-to-capsule contact, which achieved an AUC of 0.858 in the independent validation cohort, outperforming both the radiomics model (AUC =0.704) and the clinical model (AUC =0.763). It also showed good calibration and favorable net benefit on DCA across a clinically relevant threshold range. Conventional ultrasound assessment by radiologists in this cohort yielded an AUC of 0.612, with a specificity of 51.7%, notably lower than that of the combined nomogram. A meta-analysis of 12 studies and 5,337 PTC cases reported a summary C-index of 0.82 (95% CI: 0.78–0.86) for ultrasound-based radiomics in ETE identification (17). Zhu et al. demonstrated that radiomics features outperformed conventional ultrasound findings in predicting ETE in PTC, with the most efficient radiomics model achieving an AUC of 0.813 in the validation cohort (18). The present model showed comparable performance.

In this study, 10 radiomics features were selected from both the transverse and longitudinal planes, including first-order statistics, NGTDM, GLSZM, and GLRLM features. First-order features such as skewness and kurtosis reflect the intensity distribution and intratumoral heterogeneity of the tumor. Skewness has been associated with heterogeneous necrotic or calcified foci, which are commonly observed in more aggressive tumors (19). NGTDM, GLSZM, and GLRLM features capture local intensity variation, structural irregularity, and heterogeneous zonal and run patterns within the tumor, which may reflect subtle changes at the tumor margin and surrounding capsule region that are difficult to identify on visual inspection. LoG-filtered features enhance sensitivity to structural detail across different spatial scales, which may facilitate the detection of subtle capsule disruption. The dual-plane design improved performance over either plane alone (AUC: 0.720 vs. 0.648 for transverse and 0.630 for longitudinal; Table 3). This improvement has a clear anatomical rationale. ETE in PTC is inherently directional: anterior extension typically involves the strap muscles, whereas posterior extension may involve the trachea, esophagus, or recurrent laryngeal nerve (3,20). The transverse plane is more sensitive to lateral and posterior capsular changes, while the longitudinal plane better captures craniocaudal extension along the thyroid lobe (9). Importantly, this directional information is not merely of radiological interest but has direct surgical implications: the direction and extent of capsular involvement determine whether the surgeon needs to perform shaving of the strap muscles, window resection of the trachea, or careful dissection of the recurrent laryngeal nerve (6,20). By extracting textural features from both planes, the dual-plane radiomics approach captures complementary spatial information that a single imaging plane cannot fully capture, providing a more comprehensive preoperative characterization of tumor-capsule relationships. Fan et al. reported that a combined longitudinal and transverse plane radiomics model significantly outperformed single-plane models (P<0.05) in differentiated thyroid carcinoma, which is consistent with the present findings (12).

The radiomics model alone achieved an AUC of only 0.704 in the validation cohort, lower than that of the combined model (AUC =0.858). This finding, however, is not unexpected and reflects a fundamental characteristic of radiomics in the context of ETE prediction. Radiomics features are primarily derived from the intratumoral region and capture internal textural heterogeneity, intensity distribution, and microstructural patterns (10,11). By contrast, ETE is defined by the spatial relationship between the tumor and the thyroid capsule, specifically whether and to what extent the tumor has invaded beyond the capsular boundary (3). This anatomical information, including the tumor’s proximity to the capsule and its maximum diameter, cannot be fully captured by intratumoral texture analysis alone. The clinical model (AUC =0.763), which incorporates maximum diameter and nodule-to-capsule contact, directly encodes this spatial information that radiomics alone cannot provide. Tumor size has been established as one of the most important risk factors for ETE in PTC. Shin et al. showed in 2,902 PTC patients that tumor size was significantly associated with both ETE and recurrence-free survival (21). Several previous studies showed that the clinical significance of ETE varies according to primary tumor size (22,23). Wang et al. identified tumor diameter >10 mm as an independent predictor of ETE (24). Nodule-to-capsule contact has also been confirmed as one of the strongest preoperative sonographic predictors of ETE (25,26). The two approaches are therefore complementary: radiomics captures the internal characteristics of the tumor, while clinical features reflect its anatomical relationship to the capsule (27). The final nomogram includes only three accessible variables, the Rad-score, maximum diameter, and nodule-to-capsule contact, all of which can be obtained from a standard preoperative ultrasound examination, facilitating preoperative risk stratification and surgical planning. A high predicted probability of ETE may prompt consideration of more aggressive surgical strategies, such as total thyroidectomy with central compartment lymph node dissection.

This study has several limitations. First, the retrospective single-center design introduces unavoidable selection bias, and the relatively small independent validation cohort limits the robustness of the findings and their generalizability. Multicenter prospective studies with larger sample sizes are needed to confirm the model’s applicability across different institutions and patient populations. Second, the model relies only on greyscale ultrasound. Incorporating other modalities such as elastography, contrast-enhanced ultrasound (CEUS), or three-dimensional ultrasound may provide more spatial and functional information on tumor-capsule relationships. Future studies should explore integrating these multimodal features with radiomics. Thirdly, combining radiomics with molecular markers such as BRAF mutation status may provide a more biologically informed basis for preoperative ETE risk assessment, which is a promising direction for further research. Finally, although a preliminary clinical workflow has been proposed, the practical implementation of the nomogram into routine clinical practice requires further rigorously designed prospective studies for evaluation and standardization.

Conclusions

In conclusion, the nomogram integrating the Rad-score derived from dual-plane ultrasound radiomics with clinical and ultrasound predictors provides an objective and comprehensive approach to preoperative ETE prediction in PTC. This tool has the potential to assist clinicians in individualized risk stratification and surgical decision-making for PTC patients.

Acknowledgments

The authors thank the Beijing Li Huanying Medical Foundation for supporting Quan Wen’s overseas academic visit. The authors also gratefully acknowledge Prof. Azra Alizad, MD from the Translational Ultrasound Research Laboratory, Department of Radiology, Mayo Clinic (Rochester, MN, USA) for hosting Dr. Quan Wen as a visiting scholar and providing valuable academic opportunities in ultrasound imaging research.

Footnote

Reporting Checklist: The authors have completed the TRIPOD reporting checklist. Available at https://gs.amegroups.com/article/view/10.21037/gs-2026-0304/rc

Data Sharing Statement: Available at https://gs.amegroups.com/article/view/10.21037/gs-2026-0304/dss

Peer Review File: Available at https://gs.amegroups.com/article/view/10.21037/gs-2026-0304/prf

Funding: None.

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://gs.amegroups.com/article/view/10.21037/gs-2026-0304/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 Institutional Ethics Committee of Beijing Friendship Hospital, Capital Medical University (No. 2022-P2-346-01). Written informed consent was waived due to the retrospective design.

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. Li M, Dal Maso L, Pizzato M, et al. Evolving epidemiological patterns of thyroid cancer and estimates of overdiagnosis in 2013-17 in 63 countries worldwide: a population-based study. Lancet Diabetes Endocrinol 2024;12:824-36. [Crossref] [PubMed]
2. Bortz MD, Kuchta K, Winchester DJ, et al. Extrathyroidal extension predicts negative clinical outcomes in papillary thyroid cancer. Surgery 2021;169:2-6. [Crossref] [PubMed]
3. Amin MB, Greene FL, Edge SB, et al. The Eighth Edition AJCC Cancer Staging Manual: Continuing to build a bridge from a population-based to a more "personalized" approach to cancer staging. CA Cancer J Clin 2017;67:93-9.
4. Youngwirth LM, Adam MA, Scheri RP, et al. Extrathyroidal Extension Is Associated with Compromised Survival in Patients with Thyroid Cancer. Thyroid 2017;27:626-31. [Crossref] [PubMed]
5. Shi W, Wang M, Dong L, et al. Extrathyroidal extension or tumor size of primary lesion influences thyroid cancer outcomes. Nucl Med Commun 2023;44:854-9. [Crossref] [PubMed]
6. Ringel MD, Sosa JA, Baloch Z, et al. 2025 American Thyroid Association Management Guidelines for Adult Patients with Differentiated Thyroid Cancer. Thyroid 2025;35:841-985. [Crossref] [PubMed]
7. Lamartina L, Bidault S, Hadoux J, et al. Can preoperative ultrasound predict extrathyroidal extension of differentiated thyroid cancer? Eur J Endocrinol 2021;185:13-22. [Crossref] [PubMed]
8. Lee YC, Jung AR, Sohn YM, et al. Ultrasonographic features associated with false-negative and false-positive results of extrathyroidal extensions in papillary thyroid microcarcinoma. Eur Arch Otorhinolaryngol 2018;275:2817-22. [Crossref] [PubMed]
9. Qi Q, Huang X, Zhang Y, et al. Ultrasound image-based deep learning to assist in diagnosing gross extrathyroidal extension thyroid cancer: a retrospective multicenter study. EClinicalMedicine 2023;58:101905. [Crossref] [PubMed]
10. Lambin P, Rios-Velazquez E, Leijenaar R, et al. Radiomics: extracting more information from medical images using advanced feature analysis. Eur J Cancer 2012;48:441-6. [Crossref] [PubMed]
11. Gillies RJ, Kinahan PE, Hricak H. Radiomics: Images Are More than Pictures, They Are Data. Radiology 2016;278:563-77. [Crossref] [PubMed]
12. Fan F, Li F, Wang Y, et al. Integration of ultrasound-based radiomics with clinical features for predicting cervical lymph node metastasis in postoperative patients with differentiated thyroid carcinoma. Endocrine 2024;84:999-1012. [Crossref] [PubMed]
13. Yuan SS, Zhang XR, Yu XQ, et al. Prediction model for extrathyroidal extension in thyroid papillary carcinoma based on ultrasound radiomics. Sci Rep 2025;15:36200. [Crossref] [PubMed]
14. Wan F, He W, Zhang W, et al. Preoperative prediction of extrathyroidal extension: radiomics signature based on multimodal ultrasound to papillary thyroid carcinoma. BMC Med Imaging 2023;23:96. [Crossref] [PubMed]
15. Zheng T, Xu Z, Hu L, et al. Diagnostic performance and generalizability of a clinical-ultrasound radiomics model for predicting extrathyroidal extension in thyroid carcinoma: a retrospective study. Ann Med 2026;58:2650862. [Crossref] [PubMed]
16. van Griethuysen JJM, Fedorov A, Parmar C, et al. Computational Radiomics System to Decode the Radiographic Phenotype. Cancer Res 2017;77:e104-7. [Crossref] [PubMed]
17. Liu Y, Xiang L, Liu FY, et al. Accuracy of Radiomics in the Identification of Extrathyroidal Extension and BRAF(V600E) Mutations in Papillary Thyroid Carcinoma: A Systematic Review and Meta-analysis. Acad Radiol 2025;32:1385-97. [Crossref] [PubMed]
18. Zhu H, Luo H, Li Y, et al. The superior value of radiomics to sonographic assessment for ultrasound-based evaluation of extrathyroidal extension in papillary thyroid carcinoma: a retrospective study. Radiol Oncol 2024;58:386-96. [Crossref] [PubMed]
19. He ZX, Zhang XP, Yang JS. Comprehensive linkage between molecular biology and imaging radiomics for thyroid nodules. World J Radiol 2025;17:111005. [Crossref] [PubMed]
20. Enomoto K, Inohara H. Surgical strategy of locally advanced differentiated thyroid cancer. Auris Nasus Larynx 2023;50:23-31. [Crossref] [PubMed]
21. Shin CH, Roh JL, Song DE, et al. Prognostic value of tumor size and minimal extrathyroidal extension in papillary thyroid carcinoma. Am J Surg 2020;220:925-31. [Crossref] [PubMed]
22. Park J, Kang IK, Bae JS, et al. Clinical Significance of Tumor Size in Gross Extrathyroidal Extension to Strap Muscles (T3b) in Papillary Thyroid Carcinoma: Comparison with T2. Cancers (Basel) 2022;14:4615. [Crossref] [PubMed]
23. Xu M, Xi Z, Zhao Q, et al. Causal inference between aggressive extrathyroidal extension and survival in papillary thyroid cancer: a propensity score matching and weighting analysis. Front Endocrinol (Lausanne) 2023;14:1149826. [Crossref] [PubMed]
24. Wang H, Zhao S, Yao J, et al. Factors influencing extrathyroidal extension of papillary thyroid cancer and evaluation of ultrasonography for its diagnosis: a retrospective analysis. Sci Rep 2023;13:18344. [Crossref] [PubMed]
25. Guo D, Chen C, Zheng Y, et al. Ultrasound feature-based nomogram model for predicting extrathyroidal extension in papillary thyroid carcinoma. BMC Cancer 2025;25:1185. [Crossref] [PubMed]
26. Lee CY, Kim SJ, Ko KR, et al. Predictive factors for extrathyroidal extension of papillary thyroid carcinoma based on preoperative sonography. J Ultrasound Med 2014;33:231-8. [Crossref] [PubMed]
27. Feng JW, Liu SQ, Qi GF, et al. Development and Validation of Clinical-Radiomics Nomogram for Preoperative Prediction of Central Lymph Node Metastasis in Papillary Thyroid Carcinoma. Acad Radiol 2024;31:2292-305. [Crossref] [PubMed]