Related Experiment Video
Updated: Jun 18, 2026

A Swin Transformer-Based Model for Thyroid Nodule Detection in Ultrasound Images
Published on: April 21, 2023
A clinically anchored radiomics dictionary for explainable TI-RADS-based thyroid nodule classification in ultrasound;
Mohammad Salmanpour1, Shahram Taeb2, Ali Fathi Jouzdani3
1Department of Basic and Translational Research, BC Cancer Research Institute, Vancouver, BC, Canada; Department of Radiology, University of British Columbia, Vancouver, BC, Canada; Technological Virtual Collaboration Company (TECVICO CORP.), Vancouver, BC, Canada.
This study created an interpretable radiomics dictionary to link thyroid ultrasound features with TI-RADS categories, improving AI model trust for nodule classification. The framework enhances transparency in risk stratification using artificial intelligence.
Area of Science:
- Radiology and Medical Imaging
- Artificial Intelligence in Medicine
- Quantitative Ultrasound Imaging
Background:
- Artificial intelligence (AI)-based radiomics models for thyroid ultrasound (US) often lack interpretability, hindering clinical adoption and trust.
- Existing models struggle to connect quantitative US features to established semantic lexicons like the Thyroid-Imaging-Reporting-and-Data-System (TI-RADS).
Purpose of the Study:
- To develop and evaluate an interpretable radiomic feature (RF) framework for thyroid nodule classification.
- To link quantitative US features to the TI-RADS semantic lexicon using a clinically grounded radiomics dictionary.
- To enhance the transparency and clinical trust of AI-driven thyroid nodule risk stratification.
Main Methods:
- A radiomics dictionary was constructed, mapping TI-RADS categories to Image-Biomarker-Standardization-Initiative-compliant RFs from 2D US images.
- Expert consensus (physicians, physicists, radiology, biology experts) defined feature-to-lexicon relationships, analyzed with Shapley-Additive-Explanations (SHAP).
- A stability-aware composite scoring framework was used for robust model selection across 5,542 nodules from three multicenter datasets.
Main Results:
- The radiomics dictionary enabled direct interpretation of radiomic signatures within TI-RADS terminology.
- A Select-From-Model (logistic regression) plus Extra-Trees classifier achieved high testing performance (ROC-AUC: 0.941 ± 0.004).
- SHAP analysis identified texture heterogeneity (e.g., Gray Level Run Length Matrix non-uniformity) as a key malignancy indicator, aligning with high-risk TI-RADS descriptors.
Conclusions:
- An interpretable radiomics dictionary and a stability-aware model selection framework were successfully introduced.
- This approach addresses the interpretability limitations of AI radiomics models in thyroid ultrasound.
- The framework facilitates transparent and reliable thyroid nodule risk stratification using US data.
More Related Videos
Related Concept Videos
Computed Tomography
The technique was invented in the 1970s and is based on the principle that as X-rays pass through the body, they are absorbed or reflected at different levels. In the technique, a patient lies on a motorized platform while a computerized axial tomography (CAT) scanner rotates...
Anatomical Terminology
Radiological Investigation I: X-ray and CT
Goiter
Positron Emission Tomography
One of the main requirements of a PET scan is a positron-emitting radioisotope, which is produced in a cyclotron and then attached to a substance used by the part of the body being...
X-ray Imaging

