Multi-region ultrasound radiomics combining whole-tumor, habitat subregionand peritumoral features for preoperative prediction of lymph node metastasis in papillary thyroid carcinoma

This study aimed to develop a ultrasound radiomics model for preoperative prediction of lymph node metastasis (LNM) in papillary thyroid carcinoma (PTC) and to evaluate the predictive value of a clinic-radiomics combined model. A total of 390 patients with PTC were retrospectively included and divided into training and testing sets at a ratio of 7:3. Whole-tumor region, three intratumoral habitat subregions and 3-mm peritumoral region were constructed from preoperative ultrasound images for radiomics feature extraction and model development. In the testing set, the AUC of the whole-tumor, habitat and peritumoral models were 0.718, 0.755, and 0.758, respectively, while the radiomics fusion model achieved an AUC of 0.761. After integrating clinical variables, the clinic-radiomics combined model improved the AUC to 0.791. These findings suggest that habitat and peritumoral features provide complementary information beyond whole-tumor radiomics and that the combined model may improve preoperative LNM prediction in PTC.