RadGLO: an interactive platform for radiomic feature analysis and prognostic modeling in glioma

Abstract

Radiomic features, quantitative descriptors of tumor shape, texture, and intensity derived from MRI serve as powerful non-invasive biomarkers for glioma characterization and prognosis. We present Radiology of Glioma (RadGLO), an interactive platform that leverages these features across multi-institutional datasets (TCGA, UCSF, UPENN) to support grade-wise analysis, gene correlation, and survival prediction. RadGLO integrates two in-house developed modules, RaSPr for risk stratification and TumorVQ for region-specific tumor volume quantification, which also supports user-uploaded MRI data. By enabling personalized prognosis and aiding treatment planning, RadGLO offers a valuable resource that is openly accessible at https://project.iith.ac.in/cgntlab/radglo/ .

Fig. 1. Schematics of RadGLO.

RadGLO is an integrative platform designed to identify non-invasive prognostic markers for glioblastoma using multi-institutional MRI data (TCGA, UCSF, and UPENN). The workflow includes preprocessing of MRI data, tumor segmentation, radiomic feature extraction, visualization, and model development.

Fig. 2. Utility and applications of the RadGLO.

a Kaplan Meier survival analysis of the TCGA-GBM cohort demonstrates a significant survival difference between high and low RaSPr score groups, with higher scores indicating better prognosis. Similar trends were observed in validation cohorts from UCSF and UPENN. b Representative MRI visualizations of high and low RaSPr score samples highlight spatial and intensity differences captured through radiomic features. c Illustration of TumorVQ implementation within RadGLO, enabling cortical location-based tumor volume quantification and interpretation. d Radiomic feature-based comparison between low-grade glioma (LGG) and glioblastoma (GBM) reveals distinct patterns, emphasizing the platform’s ability to differentiate tumor grades non-invasively. e Survival analysis based on the radiomic feature HLL_GLRLM_LowGrayLevelRunEmphasis shows significant stratification, where higher RadScore samples are associated with poor prognosis. f Correlation analysis between radiomic features and gene expression identifies a significant association between radiomic features and the gene, suggesting potential radiogenomic linkage.