Dataset for MRI radiomics in glioma

Background: We hypothesized that an integrated Radiomics–Immune Score (R/I-Score), combining MRI-derived tumor heterogeneity with peripheral immune profiling, would predict progression-free survival (PFS) benefit from CDK6 inhibitor–containing therapy (CDK6i). Methods: We conducted a two-center, retrospective cohort study. The final study population comprised 116 adult patients who had undergone baseline MRI within 21 days prior to treatment initiation. To ensure methodological rigor and internal consistency, we adopted a two-stage study design. In the first stage (Biological Modeling), the total cohort (N=116) was partitioned into a training set (n=81) and an independent testing set (n=35) to develop and validate the R/I-Score. In the second stage (Clinical Validation), we addressed the selection bias inherent in observational data by performing 1:2 Propensity Score Matching (PSM). This yielded a balanced analytical cohort of 108 patients (36 treated with CDK6i and 72 concurrent controls). The primary endpoint was the treatment-by-biomarker interaction for progression-free survival (PFS), specifically assessing the R/I-Score's ability to stratify therapeutic responders. Results: In the overall matched cohort (N=108), patients treated with CDK6i exhibited numerical but statistically non-significant trends toward improved efficacy compared to concurrent controls in terms of median PFS (4.8 vs. 3.5 months; HR 0.72, P=0.091), objective response rate (19.4% vs. 9.7%; P=0.180), and disease control rate (44.4% vs. 30.6%; P=0.165). The Combined R/I-Clinical Model demonstrated exceptional diagnostic performance in the independent testing cohort (N=108), achieving a Treatment-Effect AUC of 0.97 and excellent calibration. In the PSM efficacy analysis, the R/I-Score successfully identified a distinct responder subpopulation: patients with a High R/I-Score derived significant survival benefit from CDK6i compared to concurrent controls (HR 0.48; 95% CI 0.25–0.91; P=0.008), whereas those with a Low R/I-Score did not (HR 1.05; 95% CI 0.58–1.92). The interaction between the R/I-Score and treatment was statistically significant (Pinteraction=0.012). Furthermore, High R/I-Score patients achieved superior objective response rates (33.3% vs. 5.6%) and sustained treatment duration (median 6 vs. 3 cycles) without increased toxicity (Grade =3 AEs: P>0.05). Biological interpretability analysis revealed that the score was driven by tumor morphological complexity and NK/CD8+ T-cell infiltration. Conclusions: The multi-modal R/I-Score serves as a robust, non-invasive predictive biomarker for CDK6i therapy in GBM.