Back to all papers

A Machine Learning Model Based on Ultrasound Radiomics and Multi-modal Clinical Features for Predicting Lymph Node Metastasis of Papillary Thyroid Carcinoma in the Setting of Hashimoto's Thyroiditis.

November 16, 2026pubmed logopapers

Authors

DuanYang S,Tang Y,Wang Y,Li Q,Song H,Hou T,He B,Nie F

Affiliations (5)

  • The Second Clinical Medical School, Lanzhou University, Lanzhou, China; Department of Ultrasound Medical Center, The Second Hospital of Lanzhou University, Gansu Province Medical Engineering Research Center for Intelligence Ultrasound, Gansu Province Clinical Research Center for Ultrasonography, Lanzhou, China.
  • School of Information Science and Engineering, Lanzhou University, Lanzhou, China.
  • Department of Ultrasound Medical Center, The Second Hospital of Lanzhou University, Gansu Province Medical Engineering Research Center for Intelligence Ultrasound, Gansu Province Clinical Research Center for Ultrasonography, Lanzhou, China.
  • Department of Rehabilitation Medicine, Lanzhou University First Hospital, Lanzhou, China.
  • Department of Ultrasound Medical Center, The Second Hospital of Lanzhou University, Gansu Province Medical Engineering Research Center for Intelligence Ultrasound, Gansu Province Clinical Research Center for Ultrasonography, Lanzhou, China. Electronic address: [email protected].

Abstract

To investigate the value of multi-modal features (ultrasound characteristics, BRAF V600E mutation, and TPO-Ab) in predicting lymph node metastasis (LNM) of papillary thyroid carcinoma (PTC) in the context of Hashimoto's thyroiditis (HT), and to construct machine learning models for individualized risk assessment. A retrospective cohort of 377 patients with pathologically confirmed PTC and concurrent HT was enrolled and divided into LNM n=186 and non-LNM n=191 groups. Preoperative ultrasound features (e.g., Minimum Tumor-to-Capsule Distance (DTC_min), calcification type and ratio, lesion location, nodule morphology), serological and pathological indicators were collected. Random forest (RF), support vector machine (SVM), extreme gradient boosting (XGBoost and XGB), logistic regression (LR), and k-nearest neighbors (KNN) were used to build prediction models. Feature importance was assessed, and model performance was evaluated using receiver operating characteristic (ROC) curves and calibration curves. Univariate analysis showed that calcification type, nodule diameter, DTC_min, and aspect ratio were significantly associated with LNM, with nodule diameter contributing the most. Overall, the integration of multi‑omics features significantly improved the accuracy of lymph node metastasis prediction. Radiomics features contributed the largest share (81.5%), followed by morphological features (M1: DTC_min/Size/Calcification, 12.7%), while molecular markers and genetic biomarkers provided supplementary information (M2: TPO‑Ab, 3.8%; M3: BRAF V600E, 2.0%). Notably, the optimal XGBoost model achieved a stable accuracy of 0.879 in an independent external cohort (n = 107), confirming its robust generalizability for clinical risk assessment. In the HT setting, machine learning models based on multi-omics feature fusion enable effective individualized risk stratification for PTC LNM, offering precise support for clinical decision-making.

Topics

Journal Article

Ready to Sharpen Your Edge?

Subscribe to join 11k+ peers who rely on RadAISlice. Get the essential weekly briefing that empowers you to navigate the future of radiology.

We respect your privacy. Unsubscribe at any time.