A Novel Radiomics-Integrated Panel for Preoperative Stratification of Pancreatic Neuroendocrine Tumors (PNETs)

Area of Science:

• Radiology
• Oncology
• Medical Imaging

Background:

• Preoperative grading of pancreatic neuroendocrine tumors (PNETs) is challenging due to the lack of histologic grade information before resection.
• Current risk stratification methods for PNETs are limited, hindering effective preoperative planning.

Purpose of the Study:

• To investigate the potential of CT-based radiomic signatures, including Δ-radiomics, for preoperative discrimination of PNET progression and tumor grade.
• To develop and validate biologically informed radiomic signatures for improved PNET risk assessment.

Main Methods:

• Analysis of 44 patients with PNETs from two centers, using contrast-enhanced CT scans and surgical pathology data.
• Extraction of radiomic features using PyRadiomics, followed by batch correction and computation of Δ-radiomic features.
• Development of lesion-only and Δ-radiomic signature families, evaluated with logistic regression, random forest, and gradient boosting classifiers.

Main Results:

• Δ-radiomic models demonstrated strong discrimination for progression prediction (AUC up to 0.87) and higher-grade disease (AUC up to 0.93).
• A specific Δ-radiomic signature (ΔBusyness × Ki-67) was significantly associated with progression-free survival (HR 0.38).
• Cross-center validation confirmed the generalizability of the radiomic models.

Conclusions:

• CT-derived radiomics, particularly shape and texture features, can noninvasively identify aggressive PNET phenotypes.
• Integrating radiomic features with clinical markers enhances preoperative risk stratification for PNETs.
• Further prospective validation in larger multicenter cohorts is recommended.