Abstract
Preoperatively distinguishing follicular thyroid carcinoma (FTC) from follicular thyroid adenoma (FTA) remains a significant clinical challenge. Current ultrasound risk stratification systems show limited efficacy for follicular neoplasms, and existing artificial intelligence (AI) approaches lack sufficient validation. We developed and validated a deep learning model using ultrasound images to differentiate FTC from FTA and classify FTC into invasion subtypes. This multicenter retrospective study incorporated data from 31 hospitals, using 1531 patients for model development and 900 across three external test sets for validation. The model demonstrated high diagnostic performance, with AUCs of 0.816–0.847 for FTC vs FTA discrimination across external test sets and robust performance across subtypes (AUC range 0.754–0.910), and generalized well to varied clinical settings. Triple-classification macro-AUCs were 0.818–0.861. It consistently outperformed radiologists and improved diagnostic accuracy as an assistive tool. Our AI model provides a reliable, non-invasive tool for preoperative diagnosis and risk stratification of follicular thyroid neoplasms.
Similar content being viewed by others
Data availability
Some or all datasets generated during and/or analyzed during the current study are not publicly available but are available from the corresponding author on reasonable request.
Code availability
All the codes used in this manuscript are available at our GitHub repository (https://github.com/samadhi-fire/Thyroid-Follicular-Neoplasm.git).
References
Boucai, L., Zafereo, M. & Cabanillas, M. E. Thyroid cancer: a review. JAMA 331, 425–435 (2024).
Haddad, R. I. et al. Thyroid carcinoma, version 2.2022, NCCN clinical practice guidelines in oncology. J. Natl. Compr. Canc Netw. 20, 925–951 (2022).
Matsuura, D. et al. Follicular and hurthle cell carcinoma: comparison of clinicopathological features and clinical outcomes. Thyroid 32, 245–254 (2022).
Li, J. et al. US risk stratification system for follicular thyroid neoplasms. Radiology 309, e230949 (2023).
Peng, S. et al. Deep learning-based artificial intelligence model to assist thyroid nodule diagnosis and management: a multicentre diagnostic study. Lancet Digit Health 3, e250–e259 (2021).
Yu, J. et al. Lymph node metastasis prediction of papillary thyroid carcinoma based on transfer learning radiomics. Nat. Commun. 11, 4807 (2020).
Toro-Tobon, D. et al. Artificial intelligence in thyroidology: a narrative review of the current applications, associated challenges, and future directions. Thyroid 33, 903–917 (2023).
Wang, Y., Lu, W., Xu, L., Xu, H. & Kong, D. Deep learning-based ultrasound diagnostic model for follicular thyroid carcinoma. Eur. Radiol. 36, 357–366 (2025).
Chen, W. et al. The value of a neural network based on multi-scale feature fusion to ultrasound images for the differentiation in thyroid follicular neoplasms. BMC Med. Imaging 24, 74 (2024).
Shen, H. et al. Artificial intelligence-augmented ultrasound diagnosis of follicular-patterned thyroid neoplasms: a multicenter retrospective study. EClinicalMedicine 86, 103351 (2025).
Yang, Z. et al. Automated diagnosis and management of follicular thyroid nodules based on the devised small-dataset interpretable foreground optimization network deep learning: a multicenter diagnostic study. Int. J. Surg. 109, 2732–2741 (2023).
Pepe, M. S., Janes, H., Longton, G., Leisenring, W. & Newcomb, P. Limitations of the odds ratio in gauging the performance of a diagnostic, prognostic, or screening marker. Am. J. Epidemiol. 159, 882–890 (2004).
Loong, T.-W. Understanding sensitivity and specificity with the right side of the brain. BMJ 327, 716–719 (2003).
Rubens, D. J., Bhatt, S., Nedelka, S. & Cullinan, J. Doppler artifacts and pitfalls. Radio. Clin. North Am. 44, 805–835 (2006).
Gadgil, S. U., Endo, M., Wen, E., et al. CheXseg: combining expert annotations with DNN-generated saliency maps for X-ray segmentation. Medical Imaging with Deep Learning 190–204 (2021).
Shen, H. et al. Noninvasive deep learning system for preoperative diagnosis of follicular-like thyroid neoplasms using ultrasound images: a multicenter, retrospective study. Ann. Surg. 10, 1097 (2025).
Nikiforov, Y. E. et al. Nomenclature revision for encapsulated follicular variant of papillary thyroid carcinoma: a paradigm shift to reduce overtreatment of indolent tumors. JAMA Oncol. 2, 1023–1029 (2016).
Chung, R. et al. Noninvasive follicular thyroid neoplasm with papillary-like nuclear features: epidemiology and long-term outcomes in a strictly defined cohort. Thyroid 31, 68–75 (2021).
Xu, W. et al. Deep learning model based on contrast-enhanced ultrasound for predicting vessels encapsulating tumor clusters in hepatocellular carcinoma. Eur. Radio. 35, 989–1000 (2025).
Liu, Z. et al. in 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) 11966–11976 (IEEE, 2022).
He, K., Zhang, X., Ren, S. & Sun, J. in 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) 770–778 (IEEE, 2016).
Xie, S., Girshick, R., Dollár, P., Tu, Z. & He, K. in 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) 5987–5995 (IEEE, 2017).
Sandler, M., Howard, A., Zhu, M., Zhmoginov, A. & Chen, L. C. MobileNetV2: Inverted Residuals and Linear Bottlenecks. (IEEE, 2018).
Vaswani, A. et al. Attention is all you need. Adv. Neural. Info. Process. Syst. 30 (2017).
Liu, Z. et al. Swin transformer: Hierarchical vision transformer using shifted windows. 10012-10022 (IEEE/CVF, 2021).
Pan, S. J. & Yang, Q. A survey on transfer learning. IEEE Trans. Knowl. Data Eng. 22, 1345–1359 (2010).
Paszke, A., Gross, S., Massa, F., Lerer, A. & Chintala, S. PyTorch: an imperative style, high-performance deep learning library. Adv. Neural. Info. Process. Syst. 32 (2019).
Deng, J. et al. in 2009 IEEE Conference on Computer Vision and Pattern Recognition. 248–255 (IEEE, 2009).
Xue, L.-Y. et al. Transfer learning radiomics based on multimodal ultrasound imaging for staging liver fibrosis. Eur. Radio. 30, 2973–2983 (2020).
Ding, K., Xiao, N. & Toh, K.-C. Adam-family methods with decoupled weight decay in deep learning. https://doi.org/10.48550/arXiv.2310.08858 (2023).
Acknowledgements
Study supported by the National Natural Science Foundation of China (Nos. 82471995, 82441010, and 82272029) and the Beijing Science Fund for Distinguished Young Scholars (No. JQ22013).
Author information
Authors and Affiliations
Contributions
J.L. and H.Z. were responsible for methodology development and formal analysis, with X.Y. contributing to formal analysis. J.L., H.Z., Y. C., L.Q., L.C., X.W, B.C., M.L., Y.X., R.M., D.S., L.L., P.L., C.X., L.M., G.H., S.H., Y.L., W.D., K.L., X.X., and L.Y. contributed to data investigation, collection, and resource provision. J.L. drafted the initial manuscript, and K.W. carried out the writing review and editing. F.M. and B.P. provided supervision throughout the study. P.L., K.W., and J.Y. conceived and designed the study. All authors had access to the raw data, participated in result interpretation, reviewed the manuscript, and approved the final version for publication.
Corresponding authors
Ethics declarations
Competing interests
The authors declare no competing interests.
Additional information
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Supplementary information
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/.
About this article
Cite this article
Li, J., Zhang, H., Zheng, H. et al. Artificial intelligence-enabled ultrasound diagnosis and stratification of follicular thyroid neoplasms: a multi-center study. npj Digit. Med. (2026). https://doi.org/10.1038/s41746-026-02489-6
Received:
Accepted:
Published:
DOI: https://doi.org/10.1038/s41746-026-02489-6
