Quantifying intra-tumoral genetic heterogeneity of glioblastoma toward precision medicine using MRI and a data-inclusive machine learning algorithm

• DOI: 10.1371/journal.pone.0299267
• Publication Date: 2024-04-03T17:34:55Z
• Authors: Lujia Wang; Hairong Wang; Fulvio D’Angelo; Lee Curtin; Christopher P. Se...; Gustavo De Leon; Kyle W. Singleton; Javier Urcuyo; Andrea Hawkins-Da...; Pamela R. Jackson; Chandan Krishna; Richard S. Zimmerman; Devi P. Patra; Bernard R. Bendok; Kris A. Smith; Peter Nakaji; Kliment Donev; Leslie C. Baxter; Maciej M. Mrugała; Michele Ceccarelli; Antonio Iavarone; Kristin R. Swanson; Nhan L. Tran; Leland S. Hu; Jing Li
• Keywords: Tumour heterogeneity

Background and objective Glioblastoma (GBM) is one of the most aggressive lethal human cancers. Intra-tumoral genetic heterogeneity poses a significant challenge for treatment. Biopsy invasive, which motivates development non-invasive, MRI-based machine learning (ML) models to quantify intra-tumoral each patient. This capability holds great promise enabling better therapeutic selection improve patient outcome. Methods We proposed novel Weakly Supervised Ordinal Support Vector Machine (WSO-SVM) predict regional alteration status within GBM tumor using MRI. WSO-SVM was applied unique dataset 318 image-localized biopsies with spatially matched multiparametric MRI from 74 patients. The model trained three driver genes (EGFR, PDGFRA PTEN) based on features extracted corresponding region five contrast images. For comparison, variety existing ML algorithms were also applied. Classification accuracy gene compared between different algorithms. SHapley Additive exPlanations (SHAP) method further compute contribution scores Finally, used generate prediction maps tumoral area help visualize heterogeneity. Results achieved 0.80 accuracy, 0.79 sensitivity, 0.81 specificity classifying EGFR; 0.71 0.70 0.72 PDGFRA; 0.78 0.83 PTEN; these results significantly outperformed Using SHAP, we found that relative contributions images differ genes, are consistent findings in literature. revealed extensive region-to-region individual terms genes. Conclusions study demonstrated feasibility enable non-invasive patient, can inform future adaptive therapies individualized oncology.