Development and validation of an intratumoral–peritumoral radiomics-clinical model for preoperative prediction of synchronous liver metastasis in pancreatic neuroendocrine tumors

Synchronous liver metastasis is an important prognostic factor in patients with pancreatic neuroendocrine tumors (PNETs). Imaging features of the primary tumor may provide information on metastatic propensity beyond direct assessment of the liver, but the contribution of the peritumoral region remains unclear. We therefore evaluated a CT-based radiomics model incorporating intratumoral and peritumoral features for preoperative prediction of synchronous liver metastasis in PNETs. This retrospective study included 109 patients with pathologically confirmed PNETs. Patients were randomly assigned to a training cohort ( n = 76) and a test cohort ( n = 33). Radiomics features were extracted from intratumoral, 3-mm peritumoral, and combined intratumoral–peritumoral regions. After model comparison, the best-performing radiomics model was combined with the independent clinical predictor identified by logistic regression. Model performance was evaluated using discrimination, calibration, decision curve analysis, and Shapley additive explanations. Among the radiomics models, the intratumoral–peritumoral model achieved the best performance, with areas under the receiver operating characteristic curve of 0.964 and 0.885 in the training and test cohorts, respectively. Tumor margin was the only independent clinical predictor. Incorporating the radiomics signature with tumor margin further improved model performance, yielding AUCs of 0.967 (95% CI, 0.933–1.000) in the training cohort and 0.907 (95% CI, 0.793–1.000) in the test cohort. The combined model also correctly identified six of eight patients with CT-occult synchronous liver metastasis. A CT-based radiomics model integrating intratumoral and peritumoral features showed good performance for preoperative risk stratification of synchronous liver metastasis in PNETs. Combining radiomics with CT semantic features may provide complementary information for identifying patients at increased risk of CT-occult synchronous liver metastasis.