Development and validation of an integrated imaging-pathology model for recurrence risk stratification in...

Although curative resection yields favorable outcomes for non-functional pancreatic neuroendocrine tumors (pNETs), postoperative recurrence remains a major concern. Existing studies have largely relied on postoperative pathological factors, while imaging characteristics have seldom been comprehensively integrated into recurrence prediction models. This study aimed to develop and validate an integrated model combining imaging and clinicopathological features to improve recurrence risk stratification.

This multicenter retrospective cohort study included patients from three tertiary centers in China who underwent curative resection for non-functional, G1/G2 pNETs. Recurrence-related factors were identified using Fine-Gray competing risk regression. A weighted imaging invasiveness index was derived from six routine preoperative CT features, and an integrated prediction model combining imaging and clinicopathological variables was developed and validated.

The training and validation cohorts comprised 310 and 70 patients, respectively. Multivariable Fine-Gray model analysis identified four independent predictors of recurrence: tumor size (sHR = 1.111, 95% CI: 1.013-1.218, p = 0.026), tumor grade (G2 vs G1) (sHR = 2.809, 95% CI: 1.171-6.743, p = 0.021), perineural invasion (sHR = 2.715, 95% CI: 1.005-7.333, p = 0.049), and imaging invasiveness index (sHR = 1.996, 95% CI: 1.008-3.951, p = 0.047). Incorporating the imaging index improved predictive performance at 24 and 36 months in both cohorts, and a four-variable risk score stratified patients into distinct risk groups.

The integrated model combining the imaging invasiveness index with clinicopathological factors may improve recurrence prediction and support individualized risk stratification in patients with resected non-functional G1/G2 pNETs, although further validation in larger external cohorts is warranted.