Radiomics-based intratumoral heterogeneity subtypes in IDH wild-type glioblastoma with pathomic and transcriptomic association analyses: a multicenter study.

IDH wild-type glioblastoma (IDH-wt GBM) exhibits substantial intratumoral heterogeneity, which contributes to treatment resistance, disease recurrence, and variable clinical outcomes. This study aimed to identify radiomics-based intratumoral heterogeneity (RITH) subtypes in IDH-wt GBM and to assess their prognostic relevance, associated pathomic differences, and transcriptomic associations. We enrolled 333 patients with IDH-wt GBM across three cohorts. Radiomic features were extracted from CE-T1 and T2-FLAIR images, and partitioning around medoids (PAM) clustering was applied to identify RITH subtypes. Survival analyses included Kaplan-Meier analysis, log-rank tests, and univariable and multivariable Cox proportional hazards regression. Pathomic features extracted via CellProfiler were used to construct the PathScore to assess RITH subtype-associated pathomic differences. RNA-seq data from Cohort 3 were used for exploratory transcriptomic analysis, including candidate differentially expressed gene (DEG) identification, protein-protein interaction (PPI) network construction, and enrichment analysis. PAM clustering identified two RITH subtypes with significantly different overall survival (OS). The High RITH subtype was associated with shorter OS and remained independently associated with worse OS after adjustment for age, sex, hospital, MGMT promoter methylation status, and treatment category. The PathScore differed significantly between RITH subtypes, indicating associated histopathological differences. Exploratory transcriptomic analysis identified eleven candidate hub genes (CCL4, IL6, CSF2, HGF, IFNG, SERPINE1, CCR2, CXCR3, CXCL13, CCL20, and IL1B). Enrichment analyses suggested that these candidate hub genes were mainly associated with immune-inflammatory processes and pathways. The High RITH subtype was associated with poorer survival and showed corresponding pathomic and exploratory transcriptomic associations. These findings suggest that RITH subtypes may serve as non-invasive imaging phenotypes linked to prognosis and biological heterogeneity in IDH-wt GBM.