Abstract
Papillary thyroid carcinoma (PTC) generally has a favorable prognosis; however, overtreatment persists because of the lack of reliable noninvasive risk stratification tools. This study developed a radiomics-based approach to enhance the preoperative assessment of PTC. Imaging features from 255 patients were analyzed, and three tumor clusters were identified via unsupervised clustering, with one cluster (Cluster 2) displaying favorable clinical and molecular profiles. A radiomics score was constructed and validated internally and externally, achieving high diagnostic accuracy (area under the curve of 0.98) and independently predicting benign features such as a lower N stage and favorable treatment responses. Transcriptomic analysis revealed immune activation and survival-related gene expression in Cluster 2. The model demonstrated robust performance in stratifying patients for active surveillance and may complement current diagnostic frameworks, offering a precise, noninvasive tool to guide clinical decision-making.