The use of whole-brain network topology and weighted gene coexpression analysis to identify potential imaging and molecular markers for glioma grading

BACKGROUND: Glioblastoma (GBM), which is characterized by high invasiveness and spatial heterogeneity, poses significant challenges in terms of prognosis and treatment. In this study, whole-brain network topology was combined with gene coexpression analysis to identify potential imaging and molecular markers for glioma grading, shifting the research focus from localized tumor lesions to the interaction mechanisms between the nervous system and gliomas. METHODS: In this study, data from open-access datasets, including glioma imaging data from the Brain Tumor Segmentation Challenge (BraTS'2019), data on glioma-related genes from BrainBase, and whole-brain transcriptomic data from the Allen Human Brain Atlas, were used. Wavelet-based features were extracted separately from T1- and T2-weighted magnetic resonance (MR) images and integrated to construct individualized structural brain networks with the aim of characterizing glioma heterogeneity. LASSO regression was applied to perform feature selection and extract the representative structural network metrics. We evaluated the effectiveness of the derived network features in glioma grading from this novel perspective via four classification models: support vector machine (SVM), random forest, logistic regression, and XGBoost. Furthermore, weighted gene coexpression network analysis (WGCNA) was employed to link disruptions in structural connectivity between high- and low-grade gliomas with the expression patterns of glioma-related genes, thereby identifying potential molecular markers associated with tumor progression. RESULTS: The results demonstrated that the whole-brain network-based grading approach achieved test-set AUC values above 0.85 across all four models, with the SVM and logistic regression models showing the best performance (AUC = 0.87). Additionally, the high invasiveness of high-grade gliomas may lead to the decoupling and reorganization of structural network modules. Protein kinase C beta (PRKCB) has been identified as a potential molecular biomarker associated with glioma grade. CONCLUSIONS: Wavelet-based individualized brain structural networks have been proposed for glioma grading models, resulting in novel imaging and molecular markers for cancer neuroscience.