Development and validation of a prediction model for lymph node metastasis in thyroid cancer: integrating deep learning and radiomics features from intra- and peri-tumoral regions

Introduction

Globally, the incidence of thyroid cancer is increasing (1), largely due to the rising prevalence of papillary thyroid carcinoma (PTC), which constitutes more than 80% of all cases (2). While PTC generally carries a favorable prognosis with a mortality rate of less than 10% (3,4), the occurrence of cervical lymph node metastasis in PTC is relatively high, affecting approximately 30–90% of patients. This metastasis is closely linked to recurrence and potential decreases in survival rates (3-7). However, there is ongoing international debate regarding the necessity of routine prophylactic central compartment lymph node dissection for PTC patients lacking clear evidence of preoperative or intraoperative central lymph node metastasis (CLNM) (8), as this procedure poses risks of complications (9). Thus, accurate preoperative assessment of cervical lymph node status is crucial for PTC patients.

Traditional imaging techniques are vital for evaluating lymph node metastasis preoperatively (3), but characterizing lymph node status remains a challenge in tumor imaging. While ultrasound is widely used for preoperative lymph node metastasis assessment (10), its reliance on operator experience and limitations in evaluating certain regions like the retropharyngeal, retrosternal, and mediastinal areas have been noted (11-13). Previous studies have demonstrated the low sensitivity of traditional ultrasound alone in detecting central compartment lymph node metastasis in PTC (3,7,14-17).

Radiomics, an emerging technology, extracts intricate information from medical images to identify high-throughput features imperceptible to the naked eye. These features are then combined into biomarkers (radiomic features) used for tumor detection, diagnosis, treatment strategy selection, prognosis inference, and tumor recurrence assessment, aiding clinical decision-making (18-23). While recent years have seen promising applications of radiomic analysis using ultrasound, magnetic resonance imaging (MRI), and computed tomography (CT) images for preoperative assessment of PTC lymph node metastasis (24-27), most studies have focused on intra-tumoral radiomic analysis. Notably, peri-tumoral features such as marginal infiltration, lymphatic vessel invasion, and peri-tumoral fibrosis have been linked to CLNM in thyroid cancer patients (28-30). However, current diagnosis of peri-tumoral information like marginal infiltration on ultrasound images relies solely on visual judgment and individual experience (31). Peri-tumoral radiomics and deep learning have been utilized in diagnosing and treating various cancers (32-37), yet reports on their application in PTC lymph node metastasis are limited (24).

Current prediction models for cervical lymph node metastasis in thyroid cancer mainly rely on intra-tumoral features, clinical factors (e.g., age, tumor size), or single-modality imaging data (24-27). Their main limitations include limited generalizability (based on single-center cohorts), lack of assessment of peri-tumoral biomarkers, and inconsistent predictive accuracy [area under the curve (AUC): 0.77–0.93]. Although existing studies have demonstrated associations between peri-tumoral characteristics and metastasis (28-30), no model to date has integrated both intra- and peri-tumoral radiomic features with deep learning. This highlights the need to develop a fused model to improve preoperative predictive accuracy.

This study aimed to investigate whether deep learning and radiomics analysis based on both intra- and peri-tumoral regions can enhance the prediction of lymph node metastasis in patients with PTC. We present this article in accordance with the TRIPOD reporting checklist (available at https://gs.amegroups.com/article/view/10.21037/gs-2025-50/rc).

Methods

The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the Ethics Committees of Shanghai Sixth People’s Hospital (approval No. 2023-YS-045) and Tongji Hospital Affiliated to Tongji University (approval No. 2021-KY-025). Patient informed consent was waived due to its retrospective design. The study was conducted from July 2020 to June 2022 at two centers (Center 1: Shanghai Sixth People’s Hospital; Center 2: Tongji Hospital Affiliated to Tongji University), including 405 patients with PTC who underwent initial thyroid surgery. Data from Center 1 were randomly split into a 70% training set (n=206) and a 30% internal validation set (n=88), while data from Center 2 (n=111) constituted the external test set. Sample size was estimated using the event-per-predictor (EPP) rule (≥10 events per predictor), ensuring robustness for selected features.

The same inclusion and exclusion criteria were applied at Center 1 and Center 2. The inclusion criteria were as follows: (I) postoperative pathology confirmed as the primary site of PTC; (II) neck lymph node dissection performed and confirmed by pathology; and (III) neck ultrasound performed for all patients within 1 month before surgery. The exclusion criteria included the following: (I) patients who underwent preoperative treatments such as radiofrequency ablation, radiotherapy, or chemotherapy in the head and neck region; (II) presence of other malignant tumors or distant metastases; (III) absence of preoperative ultrasound examination, loss of image data, or substandard image quality; and (IV) data omission or missing. All pathological specimens were retrospectively reviewed by two or more experienced pathologists. For patients with bilateral PTC or clinical evidence of contralateral cervical lymph node metastasis, bilateral central compartment neck dissection combined with total thyroidectomy was performed. The flowchart of our screening process is shown in Figure 1.

Figure 1 Flow chart of participants recruitment. Center 1: Shanghai Sixth People’s Hospital; Center 2: Tongji Hospital Affiliated to Tongji University. CLND, central lymph node dissection; PTC, papillary thyroid carcinoma.

Data collection and preprocessing

Clinical and pathological data of thyroid cancer patients, including gender, age, and postoperative pathological results, were meticulously gathered from the medical records systems of two healthcare institutions to ensure a comprehensive and diverse dataset. Ultrasound images were collected from the institutional databases using five different systems: GE LOGIQ E8, Siemens S2000, Philips EPIQ5, EPIQ7, and IU22. All devices were fitted with linear probes capable of operating in a frequency range of 4–12 MHz, allowing for the acquisition of high-resolution grayscale images. For each detected lesion, the largest available long-axis grayscale image without annotations was chosen, and its maximum longitudinal diameter was measured. All ultrasound images were stored in DICOM format to facilitate subsequent processing and analysis.

In the image preprocessing phase, two experienced radiologists independently outlined the nodule regions of interest (ROIs) using ITK-SNAP software. When discrepancies arose in their annotations, a third radiologist was consulted to review and reach a consensus on the final ROI delineation. Following this, image grayscale values were normalized to the range of 0 to 1 to minimize variations caused by different scanners and imaging settings. To meet the input requirements of deep learning models, all images were resized to 256×256 pixels using bilinear interpolation. Furthermore, the peri-tumoral ROI was generated by automatically expanding the intra-tumoral boundary outward by three voxels, as demonstrated in Figure 2.

Figure 2 Research workflow diagram. LASSO, least absolute shrinkage and selection operator; ROC, receiver operating characteristic.

Feature extraction and model development

Radiomic features were extracted from the delineated ROI regions using the Pyradiomics package, encompassing four categories of features: shape features, first-order histogram features, texture features, and wavelet features. To extract deep learning features, the study utilized a pre-trained ResNet-50 convolutional neural network (CNN) as the main feature extraction method. Specifically, the final fully connected layer of the ResNet-50 model was removed, retaining only the convolutional layers to extract high-level feature maps. The preprocessed ultrasound images were fed into the adapted ResNet-50 model, where global average pooling was utilized to generate fixed-dimension feature vectors. These features were then employed to train a support vector machine (SVM) classifier, with the goal of predicting the likelihood of CLNM in patients.

To select the most significant features for model development, the study applied both analysis of variance (ANOVA) and least absolute shrinkage and selection operator (LASSO) regression techniques. The machine learning prediction models primarily included: radiomic models (intra-tumoral radiomic model, peri-tumoral radiomic model, and combined intra- and peri-tumoral radiomic model), deep learning models (intra- deep learning model, peri-tumoral deep learning model, and combined intra- and peri-tumoral deep learning model), as well as a combined intra- and peri-tumoral radiomic and deep learning model.

Statistical analysis

The performance of the model was assessed through receiver operating characteristic (ROC) analysis, including metrics such as the AUC, accuracy, sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV). The performance differences between the integrated models and single-modality models were assessed using both internal validation sets and external test sets. Statistical analyses were conducted using SPSS software (version 25.0). Continuous variables were analyzed using independent sample t-tests or Mann-Whitney U tests, while categorical variables were evaluated using Chi-squared tests or Fisher’s exact tests. DeLong’s test was used to compare AUCs between models. All statistical tests were two-sided, and a P value of less than 0.05 was considered statistically significant.

Result

Basic characteristics of clinical information of patients

Between July 2020 and June 2022, a total of 405 eligible patients were enrolled from two centers: 294 from Center 1 and 111 from Center 2. Patients from Center 1 were split into a training cohort and an internal validation group. The mean ages were similar across all groups, with the training set averaging 46.39±12.38 years, the internal validation set 48.28±13.59 years, and the external test set 45.94±13.46 years. The overall male-to-female ratio was 1:2.2. Among the 405 patients diagnosed with PTC, 171 exhibited CLNM, distributed across the training set (n=91), internal validation set (n=36), and external test set (n=44) (Table 1).

Performance of radiomics models

For the radiomics analysis, 1,561 features were extracted from ultrasound images of both intra- and peri-tumoral regions. Feature selection using ANOVA and LASSO identified key features for constructing three SVM models: intra-tumoral radiomics, peri-tumoral radiomics, and combined intra- and peri-tumoral radiomics models.

In the internal validation cohort, the intra-tumoral radiomics model demonstrated an AUC of 0.760 (95% CI: 0.658–0.861), with an accuracy of 71.6%, sensitivity of 77.8%, and specificity of 67.3%. In the external test set, this model yielded an AUC of 0.748 (95% CI: 0.659–0.837) with an accuracy of 69.4%, a sensitivity of 29.5%, and a specificity of 95.5% (Table 2, Appendix 1).

The peri-tumoral radiomics model exhibited enhanced performance, achieving an AUC of 0.795 (95% CI: 0.700–0.890) in the internal validation, with 75.0% accuracy, 80.6% sensitivity, and 71.2% specificity. In the external test set, it reached an AUC of 0.804 (95% CI: 0.728–0.881), along with 73.0% accuracy, 63.6% sensitivity, and 79.1% specificity (Appendix 1).

The combined radiomics model showed further improvement in diagnostic performance, yielding an AUC of 0.828 (95% CI: 0.735–0.921) in the internal validation set, with 76.1% accuracy, 91.7% sensitivity, and 65.4% specificity. In the external test cohort, it achieved an AUC of 0.821 (95% CI: 0.742–0.900), along with 74.8% accuracy, 70.5% sensitivity, and 77.6% specificity (Appendix 1).

Performance of deep learning models

In the deep learning analysis, 2,048 features were extracted from each ultrasound image. Selected features were used to build intra-tumoral, peri-tumoral, and combined deep learning SVM models. The intra-tumoral deep learning model attained an AUC of 0.794 (95% CI: 0.689–0.898) in the internal validation, with 79.5% accuracy, 69.4% sensitivity, and 86.5% specificity. In the external test set, it demonstrated an AUC of 0.756 (95% CI: 0.659–0.853), along with 77.5% accuracy, 45.5% sensitivity, and 98.5% specificity (Appendix 2).

The peri-tumoral deep learning model showed improved performance with AUCs of 0.825 (95% CI: 0.733–0.918) in the internal validation set, achieving 75.0% accuracy, 80.6% sensitivity, and 71.2% specificity. In the external test cohort, it reached an AUC of 0.828 (95% CI: 0.747–0.909), with 80.2% accuracy, 70.5% sensitivity, and 86.6% specificity (Appendix 2).

The combined deep learning model attained AUCs of 0.846 (95% CI: 0.761–0.932) in the internal validation, with 80.7% accuracy, 63.9% sensitivity, and 92.3% specificity. In the external test set, it achieved an AUC of 0.853 (95% CI: 0.779–0.927), along with 81.1% accuracy, 75.0% sensitivity, and 85.1% specificity, as shown in Figure 3A,3B (Appendix 2).

Figure 3 Comparison among radiomics model, deep learning model and radiomics-deep learning model in the internal validation set and external test set. (A) AUC in the internal validation set; (B) AUC in the external test set; (C) DeLong’s test for the external test set; and (D) DCA curves for the external test set. AUC, area under the curve; CI, confidence interval; DCA, decision curve analysis; DL, deep learning; intra, intra-tumoral; peri, peri-tumoral; RAD, radiomics; SVM, support vector machine.

Performance of radiomics-deep learning fusion models

Fusion models integrating radiomics and deep learning features were constructed for intra-tumoral, peri-tumoral, and combined regions. The intra-tumoral fusion model achieved an AUC of 0.790 (95% CI: 0.694–0.886) in the internal validation set, with an accuracy of 71.6%, sensitivity of 83.3%, and specificity of 63.5%. In the external test set, it reached an AUC of 0.773 (95% CI: 0.682–0.865), along with 68.5% accuracy, 79.5% sensitivity, and 61.2% specificity (Appendix 3).

The peri-tumoral fusion model combining radiomics and deep learning features achieved an AUC of 0.832 (95% CI: 0.744–0.920) in the internal validation, with 79.5% accuracy, 72.2% sensitivity, and 84.6% specificity. In the external test set, it attained an AUC of 0.834 (95% CI: 0.756–0.912), along with 77.5% accuracy, 70.5% sensitivity, and 82.1% specificity (Appendix 3).

Notably, the fusion model integrating radiomics and deep learning features from both intra- and peri-tumoral regions showed the best diagnostic performance. It achieved an AUC of 0.897 (95% CI: 0.830–0.965) in the internal validation set, with 84.1% accuracy, 86.1% sensitivity, and 82.7% specificity. In the external test set, it reached an AUC of 0.881 (95% CI: 0.812–0.949), along with 80.2% accuracy, 75.0% sensitivity, and 83.6% specificity (Figure 3A,3B, Appendix 3).

Comparative performance and clinical utility

Comparative analysis revealed that the combined intra- and peri-tumoral radiomics-deep learning fusion model outperformed the other models in the external test set. According to DeLong’s test, this model showed statistically significant improvements compared to the intra-tumoral radiomics model (P=0.008), the intra-tumoral deep learning model (P=0.005), and the intra-tumoral fusion model (P=0.01). However, no significant differences were found when compared to the peri-tumoral radiomics model (P=0.051), combined intra- and peri-tumoral radiomics model (P=0.15), peri-tumoral deep learning model (P=0.07), peri-tumoral fusion model (P=0.07), and the combined intra- and peri-tumoral deep learning model (P=0.17) (Figure 3C).

Decision curve analysis further demonstrated that the combined intra- and peri-tumoral radiomics-deep learning fusion model offers greater clinical utility for predicting CLNM in patients with PTC (Figure 3D).

Discussion

This study effectively predicted CLNM in thyroid cancer patients by combining radiomic features from both intra- and peri-tumoral areas with deep learning characteristics. The combined radiomic and deep learning model showed superior performance in predicting CLNM compared to other diagnostic approaches involving radiomics or deep learning alone or in combination. The proposed model achieved better predictive stability, reducing both missed diagnoses and unnecessary central lymph node dissections (CLNDs), thereby improving preoperative CLND decision-making.

Ultrasound provides preoperative guidance for determining CLNM in patients with PTC; however, diagnostic accuracy is often unsatisfactory due to various factors (38). Previous studies have shown that imaging techniques such as radiomics and deep learning based on ultrasound images can effectively reveal information imperceptible to the human eye and yield promising results in diagnosing PTC lymph node metastasis. For instance, Feng et al. (39) constructed a clinical radiomics model integrating five features (age, tumor size, margin, lateral lymph node metastasis, and radiomics signature), achieving an internal validation set AUC of 0.925. Zhou et al. (40) developed a deep learning prediction model for thyroid cancer lymph node metastasis based on conventional ultrasound and color Doppler flow imaging (CDFI) images, with internal and external test AUCs of 0.86 and 0.77, respectively. Additionally, Yu et al. (1) established a deep transfer learning prediction model for thyroid cancer lymph node metastasis, achieving AUCs of 0.93 in two independent validation sets. Consistently, our study achieved promising outcomes, showing AUC values of 0.76 (internal) and 0.748 (external) for the intra-tumoral radiomics model, and 0.794 (internal) and 0.756 (external) for the intra-tumoral deep learning model across the test groups.

Previous research has indicated the association between peri-tumoral features and CLNM in thyroid cancer patients (28-30). However, to our knowledge, there has been no prior application of radiomics and deep learning studies focusing on peri-tumoral regions for predicting CLNM in PTC (24,41). The peri-tumoral radiomics model achieved AUCs of 0.795 (internal) and 0.804 (external), whereas the peri-tumoral deep learning model obtained AUCs of 0.825 and 0.828 in the internal and external test groups, respectively. There were no statistically significant differences in diagnostic performance between the intra- and peri-tumoral radiomics models or between the intra- and peri-tumoral deep learning models (P>0.05). However, combining intra- and peri-tumoral radiomics features significantly improved diagnostic performance compared with that of the intra-tumoral radiomics model (P<0.05). In contrast, the combination of intra- and peri-tumoral deep learning features did not lead to a significant improvement in diagnostic performance when compared to the intra-tumoral radiomics model (P>0.05). These findings indicate that incorporating peri-tumoral radiomics with deep learning features can enhance the prediction of lymph node metastasis in thyroid cancer.

By combining both intra- and peri-tumoral radiomics and deep learning features, we effectively predicted CLNM in thyroid cancer patients, achieving AUCs of 0.897 (internal) and 0.881 (external), which represented the highest diagnostic accuracy among all models tested. Gao et al. (38) predicted CLNM in thyroid cancer patients by integrating clinical, radiomics, and deep learning features, achieving an AUC of 0.841. Our combined intra- and peri-tumoral radiomics-deep learning model outperformed the model established by Gao et al. (38). As highlighted in recent oncology studies (42,43), integrating multi-source features significantly boosts predictive power, aligning with our findings.

This study has several limitations. Firstly, the sample size was relatively small, necessitating validation with larger samples to reduce the impact of overfitting. Secondly, delineation of intra- and peri-tumoral ROIs was conducted at the maximum two-dimensional level of the lesion, overlooking features from other levels of the lesion. Thirdly, selecting a three-voxel margin as the peri-tumoral ROI ignored features at other distances from the tumor, potentially leading to the loss of certain tumor-related features. Finally, this was a retrospective study, and prospective studies with long-term follow-up are needed to validate the model’s efficacy.

Conclusions

This study demonstrates that the radiomics-deep learning fusion SVM model integrating intra- and peri-tumoral features significantly improves preoperative prediction of CLNM in PTC. The model achieved superior diagnostic performance (AUC: 0.897 internal/0.881 external), outperforming standalone radiomics and deep learning models (P<0.05) and providing clinically actionable risk stratification.

In conclusion, our model may serve as a valuable tool to assist inexperienced radiologists in reducing the subjectivity of diagnoses and help clinicians formulate rational treatment plans.

Acknowledgments

None.

Footnote

Reporting Checklist: The authors have completed the TRIPOD reporting checklist. Available at https://gs.amegroups.com/article/view/10.21037/gs-2025-50/rc

Data Sharing Statement: Available at https://gs.amegroups.com/article/view/10.21037/gs-2025-50/dss

Peer Review File: Available at https://gs.amegroups.com/article/view/10.21037/gs-2025-50/prf

Funding: The study was funded by the Science and Technology Development of Pudong New Area, Shanghai, China (No. 2023-Y52).

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://gs.amegroups.com/article/view/10.21037/gs-2025-50/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 Ethics Committee of Shanghai Sixth People’s Hospital (approval No. 2023-YS-045) and Tongji Hospital Affiliated to Tongji University (approval No. 2021-KY-025). Patient informed consent was waived due to its 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. 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]
2. Lee DH, Kim YK, Yu HW, et al. Computed Tomography for Detecting Cervical Lymph Node Metastasis in Patients Who Have Papillary Thyroid Microcarcinoma with Tumor Characteristics Appropriate for Active Surveillance. Thyroid 2019;29:1653-9. [Crossref] [PubMed]
3. Kim E, Park JS, Son KR, et al. Preoperative diagnosis of cervical metastatic lymph nodes in papillary thyroid carcinoma: comparison of ultrasound, computed tomography, and combined ultrasound with computed tomography. Thyroid 2008;18:411-8. [Crossref] [PubMed]
4. Yoo BJ, Song CM, Ji YB, et al. Efficacy of Central Neck Dissection for Clinically Node-Negative Papillary Thyroid Carcinoma: Propensity Scoring Matching. Front Endocrinol (Lausanne) 2019;10:172. [Crossref] [PubMed]
5. Liu C, Xiao C, Chen J, et al. Risk factor analysis for predicting cervical lymph node metastasis in papillary thyroid carcinoma: a study of 966 patients. BMC Cancer 2019;19:622. [Crossref] [PubMed]
6. Liu X, Ouyang D, Li H, et al. Papillary thyroid cancer: dual-energy spectral CT quantitative parameters for preoperative diagnosis of metastasis to the cervical lymph nodes. Radiology 2015;275:167-76. [Crossref] [PubMed]
7. Choi YJ, Yun JS, Kook SH, et al. Clinical and imaging assessment of cervical lymph node metastasis in papillary thyroid carcinomas. World J Surg 2010;34:1494-9. [Crossref] [PubMed]
8. Kutler DI, Crummey AD, Kuhel WI. Routine central compartment lymph node dissection for patients with papillary thyroid carcinoma. Head Neck 2012;34:260-3. [Crossref] [PubMed]
9. Hall CM, Snyder SK, Lairmore TC. Central Lymph Node Dissection Improves Lymph Node Clearance in Papillary Thyroid Cancer Patients with Lateral Neck Metastases, Even after Prior Total Thyroidectomy. Am Surg 2018;84:531-6.
10. Haugen BR, Alexander EK, Bible KC, et al. 2015 American Thyroid Association Management Guidelines for Adult Patients with Thyroid Nodules and Differentiated Thyroid Cancer: The American Thyroid Association Guidelines Task Force on Thyroid Nodules and Differentiated Thyroid Cancer. Thyroid 2016;26:1-133. [Crossref] [PubMed]
11. Lee Y, Kim JH, Baek JH, et al. Value of CT added to ultrasonography for the diagnosis of lymph node metastasis in patients with thyroid cancer. Head Neck 2018;40:2137-48. [Crossref] [PubMed]
12. Suh CH, Baek JH, Choi YJ, et al. Performance of CT in the Preoperative Diagnosis of Cervical Lymph Node Metastasis in Patients with Papillary Thyroid Cancer: A Systematic Review and Meta-Analysis. AJNR Am J Neuroradiol 2017;38:154-61. [Crossref] [PubMed]
13. Cho SJ, Suh CH, Baek JH, et al. Diagnostic performance of CT in detection of metastatic cervical lymph nodes in patients with thyroid cancer: a systematic review and meta-analysis. Eur Radiol 2019;29:4635-47. [Crossref] [PubMed]
14. Kim SK, Woo JW, Park I, et al. Computed Tomography-Detected Central Lymph Node Metastasis in Ultrasonography Node-Negative Papillary Thyroid Carcinoma: Is It Really Significant? Ann Surg Oncol 2017;24:442-9. [Crossref] [PubMed]
15. Yang SY, Shin JH, Hahn SY, et al. Comparison of ultrasonography and CT for preoperative nodal assessment of patients with papillary thyroid cancer: diagnostic performance according to primary tumor size. Acta Radiol 2020;61:21-7. [Crossref] [PubMed]
16. Khokhar MT, Day KM, Sangal RB, et al. Preoperative High-Resolution Ultrasound for the Assessment of Malignant Central Compartment Lymph Nodes in Papillary Thyroid Cancer. Thyroid 2015;25:1351-4. [Crossref] [PubMed]
17. Lee DW, Ji YB, Sung ES, et al. Roles of ultrasonography and computed tomography in the surgical management of cervical lymph node metastases in papillary thyroid carcinoma. Eur J Surg Oncol 2013;39:191-6. [Crossref] [PubMed]
18. 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]
19. Dong D, Tang L, Li ZY, et al. Development and validation of an individualized nomogram to identify occult peritoneal metastasis in patients with advanced gastric cancer. Ann Oncol 2019;30:431-8. [Crossref] [PubMed]
20. Gillies RJ, Kinahan PE, Hricak H. Radiomics: Images Are More than Pictures, They Are Data. Radiology 2016;278:563-77. [Crossref] [PubMed]
21. Limkin EJ, Sun R, Dercle L, et al. Promises and challenges for the implementation of computational medical imaging (radiomics) in oncology. Ann Oncol 2017;28:1191-206. [Crossref] [PubMed]
22. Liu Z, Wang S, Dong D, et al. The Applications of Radiomics in Precision Diagnosis and Treatment of Oncology: Opportunities and Challenges. Theranostics 2019;9:1303-22. [Crossref] [PubMed]
23. Laghi A, Voena C. CT-based Radiomics for Biliary Tract Cancer: A Possible Solution for Predicting Lymph Node Metastases. Radiology 2019;290:99-100. [Crossref] [PubMed]
24. Gao X, Ran X, Ding W. The progress of radiomics in thyroid nodules. Front Oncol 2023;13:1109319. [Crossref] [PubMed]
25. Zhou Y, Su GY, Hu H, et al. Radiomics analysis of dual-energy CT-derived iodine maps for diagnosing metastatic cervical lymph nodes in patients with papillary thyroid cancer. Eur Radiol 2020;30:6251-62. [Crossref] [PubMed]
26. Hu W, Wang H, Wei R, et al. MRI-based radiomics analysis to predict preoperative lymph node metastasis in papillary thyroid carcinoma. Gland Surg 2020;9:1214-26. [Crossref] [PubMed]
27. Zhou SC, Liu TT, Zhou J, et al. An Ultrasound Radiomics Nomogram for Preoperative Prediction of Central Neck Lymph Node Metastasis in Papillary Thyroid Carcinoma. Front Oncol 2020;10:1591. [Crossref] [PubMed]
28. Jung CK, Kang YG, Bae JS, et al. Unique patterns of tumor growth related with the risk of lymph node metastasis in papillary thyroid carcinoma. Mod Pathol 2010;23:1201-8. [Crossref] [PubMed]
29. Jung YY, Lee CH, Park SY, et al. Characteristic tumor growth patterns as novel histomorphologic predictors for lymph node metastasis in papillary thyroid carcinoma. Hum Pathol 2013;44:2620-7. [Crossref] [PubMed]
30. Kim KJ, Hong SW, Lee YS, et al. Tumor margin histology predicts tumor aggressiveness in papillary thyroid carcinoma: a study of 514 consecutive patients. J Korean Med Sci 2011;26:346-51. [Crossref] [PubMed]
31. Kamaya A, Tahvildari AM, Patel BN, et al. Sonographic Detection of Extracapsular Extension in Papillary Thyroid Cancer. J Ultrasound Med 2015;34:2225-30. [Crossref] [PubMed]
32. Ding J, Chen S, Serrano Sosa M, et al. Optimizing the Peritumoral Region Size in Radiomics Analysis for Sentinel Lymph Node Status Prediction in Breast Cancer. Acad Radiol 2022;29:S223-8. [Crossref] [PubMed]
33. Kawazoe Y, Shiinoki T, Fujimoto K, et al. Investigation of the combination of intratumoral and peritumoral radiomic signatures for predicting epidermal growth factor receptor mutation in lung adenocarcinoma. J Appl Clin Med Phys 2023;24:e13980. [Crossref] [PubMed]
34. Vural Topuz Ö, Aksu A, Yılmaz Özgüven MB. A different perspective on 18F-FDG PET radiomics in colorectal cancer patients: The relationship between intra & peritumoral analysis and pathological findings. Rev Esp Med Nucl Imagen Mol (Engl Ed) 2023;42:359-66. [Crossref] [PubMed]
35. Zhang S, Shao H, Li W, et al. Intra- and peritumoral radiomics for predicting malignant BiRADS category 4 breast lesions on contrast-enhanced spectral mammography: a multicenter study. Eur Radiol 2023;33:5411-22. [Crossref] [PubMed]
36. Li N, Wan X, Zhang H, et al. Tumor and peritumor radiomics analysis based on contrast-enhanced CT for predicting early and late recurrence of hepatocellular carcinoma after liver resection. BMC Cancer 2022;22:664. [Crossref] [PubMed]
37. Qiu Y, Liu YF, Shu X, et al. Peritumoral Radiomics Strategy Based on Ensemble Learning for the Prediction of Gleason Grade Group of Prostate Cancer. Acad Radiol 2023;30:S1-S13. [Crossref] [PubMed]
38. Gao Y, Wang W, Yang Y, et al. An integrated model incorporating deep learning, hand-crafted radiomics and clinical and US features to diagnose central lymph node metastasis in patients with papillary thyroid cancer. BMC Cancer 2024;24:69. [Crossref] [PubMed]
39. 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]
40. Zhou LQ, Zeng SE, Xu JW, et al. Deep learning predicts cervical lymph node metastasis in clinically node-negative papillary thyroid carcinoma. Insights Imaging 2023;14:222. [Crossref] [PubMed]
41. Toro-Tobon D, Loor-Torres R, Duran M, et al. Artificial Intelligence in Thyroidology: A Narrative Review of the Current Applications, Associated Challenges, and Future Directions. Thyroid 2023;33:903-17. [Crossref] [PubMed]
42. Chen L, Chen B, Zhao Z, et al. Using artificial intelligence based imaging to predict lymph node metastasis in non-small cell lung cancer: a systematic review and meta-analysis. Quant Imaging Med Surg 2024;14:7496-512. [Crossref] [PubMed]
43. Zhu J, Tao J, Zhang F, et al. Machine learning algorithms for predicting malignancy grades of lung adenocarcinoma and guiding treatments: CT radiomics-based comparisons. J Thorac Dis 2025;17:2423-40. [Crossref] [PubMed]