A clinically anchored radiomics dictionary for explainable TI-RADS-based thyroid nodule classification in ultrasound;...

Artificial Intelligence-based radiomics models for thyroid ultrasound (US) often lack interpretability, limiting clinical trust. This study develops and evaluates an interpretable radiomic feature (RF) framework for thyroid-nodule classification by linking quantitative-US features to the Thyroid-Imaging-Reporting-and-Data-System (TI-RADS) semantic lexicon through a clinically grounded radiomics dictionary.

A radiomics dictionary was constructed to map TI-RADS categories, including composition, echogenicity, shape, margin, and echogenic foci, to Image-Biomarker-Standardization-Initiative-compliant RFs extracted from two-dimensional-US images. Relationships were defined through expert consensus (four physicians, three physicists, one radiology expert, one biologist) and further examined using Shapley-Additive-Explanations (SHAP) as a model-based interpretability analysis. Three multicenter datasets were combined, yielding 5,542 nodules, from which 107-RFs were extracted using PyRadiomics and normalized with min-max scaling. Twenty-seven feature-selection methods were paired with twenty-five classifiers and evaluated using stratified five-fold cross-validation on 70 % of the data, followed by evaluation on a held-out multicenter testing set comprising the remaining 30 % for benign-versus-malignant nodule classification. Robust model selection employed a stability-aware-composite-scoring framework combining mean performance and variability across accuracy, precision, recall, F1-score, and Receiver-Operating-Characteristic-Area-Under-the-Curve (ROC-AUC).

The dictionary enabled direct interpretation of radiomic signatures in TI-RADS terms. The Select-From-Model (logistic regression) plus Extra-Trees classifier achieved strong testing performance (ROC-AUC:0.941 ± 0.004). SHAP identified texture heterogeneity as the dominant malignancy signal, with Gray Level Run Length Matrix non-uniformity, intensity dispersion, and kurtosis aligning predictions with high-risk TI-RADS descriptors.

This study introduces an interpretable radiomics dictionary and stability-aware model selection framework, addressing interpretability limitations and enabling transparent thyroid nodule risk stratification from US.