Author: Nobuki Imano, Yuzuha Kadooka, Daisuke Kawahara, Misato Kishi, Yuji Murakami, Shumpei Onishi 👨🔬
Affiliation: Department of Radiation Oncology, Graduate School of Biomedical and Health Sciences, Hiroshima University, Graduate School of Biomedical and Health Sciences, Hiroshima University, Department of Neurosurgery, Hiroshima University Hospital 🌍
Purpose: Radiomics has proven useful in predicting overall survival in glioblastoma (GBM) patients, but consistent molecular correlations remain unidentified, leaving its biological basis unclear. This study aims to develop a prognostic model using radiomic analysis, link prognostic imaging features to their underlying biological significance, and leverage these insights to explore potential therapeutic agents.
Methods: A retrospective analysis of 98 glioblastoma (GBM) patients was conducted using TCIA/TCGA databases. Four MRI sequences—T1-weighted, contrast-enhanced T1-weighted, T2-weighted, and FLAIR—were used to extract 81,216 radiomic features. Univariate and multivariate Cox regression analyses identified clinical risk factors, while LASSO-Cox regression selected radiomic features to calculate a radiomics score (Rad-score). A nomogram-based prognostic model, evaluated via C-index (0.81), Kaplan-Meier analysis, and log-rank test, stratified patients into high- and low-risk groups. Gene set enrichment analysis of a validation cohort with CT and RNA sequencing data revealed the biological and immune significance of radiomics. Using a novel enrichment-based AI-drug discovery approach, drugs linked to favorable and poor prognostic groups were identified, providing potential therapeutic targets.
Results: Fourteen radiomic features were selected via LASSO-Cox regression to compute the Rad-score, and multivariate Cox regression identified the Rad-score and age as independent prognostic factors. The nomogram demonstrated strong predictive performance (C-index: 0.81) and stratified patients into high- and low-risk groups (P < 0.01). Enrichment-based analysis linked the Rad-score to pathways such as protein synthesis, immune signaling, and cell adhesion. Notably, isotretinoin and dasatinib, both related to these pathways, ranked highest and were identified through the developed drug discovery approach, achieving Combined Scores of 766.95 and 283.60, derived from the odds ratio and p-value, respectively.
Conclusion: The proposed framework accurately predicts OS in GBM patients, uncovers the biological significance of radiomic features, and drives drug discovery by integrating imaging data with therapeutic insights.