Predicting persistent/recurrent cervical lymph node metastasis in papillary thyroid carcinoma with PET/CT-based...

Current evidence regarding radiomics-based assessment of persistent/recurrent cervical lymph node metastasis (CLNM) remains limited, particularly in the field of PET/CT in papillary thyroid cancer (PTC). Therefore, we aimed to construct a prediction model for persistent/recurrent CLNM of PTC by fluoro-18-deoxyglucose ( 18 F-FDG) PET/computed tomography (CT) multimodal radiomics.

We retrospectively analyzed postoperative patients with PTC who underwent 18 F-FDG PET/CT at our hospital between June 2021 and June 2024. A total of 425 CLNs (219 metastatic and 206 nonmetastatic) from 158 patients were included, and the patients were randomly assigned to a training set and a test set at a ratio of 7 : 3. Pearson correlation analysis and least absolute shrinkage and selection operator regression were utilized to select PET/CT radiomic features for model development. Five radiomics models were developed based on distinct feature sets: clinical features alone, CT radiomics, PET radiomics, PET/CT radiomics, and integrated PET/CT radiomics combined with clinical features. Subsequently, a nomogram model was built by combining radscore and clinical features.

The integrated model demonstrated superior diagnostic accuracy, with areas under the curve (AUCs) of 0.944 in the training set and 0.922 in the test set. PET/CT models demonstrated the following AUCs: PET/CT (0.887), PET (0.834), CT (0.775), and clinical features alone (0.783). A nomogram incorporating clinical features and radscores was then developed, achieving C-indices of 0.937 in the training set and 0.895 in the test set.

The integration of PET/CT radiomics and clinical features may provide a noninvasive tool for predicting persistent/recurrent CLNM in postoperative PTC patients, with potential to support clinical decision‑making.