NIMG-16 - The prognostic value and biological significance of MRI CE-T1-based radiomics models in CNS5-identified GBM

Background: Following the publication of the 2021 WHO Classification of CNS Tumors (Fifth Edition) (CNS5), prognostic markers to guide the individualized treatment of patients with glioblastoma (GBM) (CNS5) need to be explored. Radiomics is a non-invasive, reproducible, and cost-effective method that has been used in the prognostic assessment of multiple solid tumors. The prognostic value and biological significance of MRI T1 - weighted enhancement (CE-T1) - based radiomics models in GBM (CNS5) has not been reported.

Methods: The tumor region of MRI CE-T1 phase was segmented, and radiomics features were extracted. The consistency analysis, univariate Cox regression and two machine learning algorithms were used for feature reduction. Then, the radiomics prognostic model was established to calculated the radiomics score (RS). The independent external validation was independently verified using the bicenter radiotherapy cohorts. The enrichment analyses of the differential genes between RS-high and -low groups were implemented to explore the biological significance.

Results: A prognostic model composed of six radiomics features were built. A high RS (HR=3.718, 95%CI: 2.222−6.220, P< 0.001) was an independent risk factor for overall survival (OS). The result was externally validated by two cohorts. Through biological significance exploration, RS was found to be significantly correlated with DNA repair (P=0.009) and glycolysis (P=0.001) pathway enrichment scores. RS was associated with γδT cell infiltration and the expression of LAG3.

Conclusion: The MRI CE-T1 based radiomics prognostic models constructed can predict GBM (CNS5) prognosis noninvasively and effectively, which is relevant to DNA repair, and may guide the screening of radiosensitive populations.