AI-based pathomics model predicts regulatory T cell infiltration and radiotherapy response in IDH-wild-type glioblastoma.

Regulatory T cells (Tregs) contribute significantly to immune suppression and therapy resistance in isocitrate dehydrogenase (IDH)-wild-type glioblastoma (GBM), a highly aggressive brain tumor with poor prognosis. In this study, we developed an artificial intelligence (AI)-powered pathomics model to predict Treg infiltration and stratify prognosis in GBM patients undergoing radiotherapy. Using high-dimensional features extracted from hematoxylin and eosin-stained biopsies, we constructed a pathomics score (PS) via gradient boosting after feature selection with Minimum Redundancy Maximum Relevance (mRMR) and Relief algorithms. The model demonstrated strong predictive performance across multi-center cohorts (n > 300), where high PS was significantly associated with elevated Treg levels and reduced overall survival (TCGA: HR = 2.16; validation cohort: HR = 1.706). Gene set enrichment analysis linked high PS to immune-evasive pathways, including Notch and IL-6/JAK/STAT3 signaling, along with increased expression of DNA repair gene RAD50, suggesting a potential association with radiotherapy response. This AI-based pathomics framework offers a robust and interpretable tool for immunoprofiling and outcome prediction, paving the way for precision radiotherapy and Treg-targeted therapeutic strategies in glioblastoma.