CT radiomics fusion model predicts pancreatic neuroendocrine tumor grade with high accuracy

CT-based radiomics fusion model integrating semantic features predicted pancreatic neuroendocrine neoplasm grading with AUC 0.891 in validation, outperforming clinical and radiomics models alone. SHAP identified zone entropy and sphericity as key predictors, reflecting lesion ag…

• Fusion model (radiomics + semantic features) achieved AUC 0.891 for preoperative grading of pancreatic neuroendocrine neoplasms in a validation cohort of 125 patients.
• SHAP analysis highlighted microscopic texture (zone entropy) and macro-geometric irregularity (sphericity) as dominant imaging biomarkers of high-grade lesions.
• Seven machine-learning algorithms were cross-validated to ensure methodological robustness.

Automated summary

RadPigeon summaries are original and for information only. They are not clinical advice.