Cost-Efficient glioma classification using machine learning with optimized molecular biomarker selection

Determining whether a brain tumor is a high-grade or low-grade glioma has a profound impact on clinical decision-making, patient prognosis, and the overall treatment strategy. High-grade gliomas are aggressive, fast-growing, and typically require intensive intervention. By contrast, low-grade gliomas often follow a slower progression and may be managed more conservatively. Currently, accurate classification of a glioma relies heavily on tissue biopsies, which are invasive, carry procedural risks, and may involve sampling errors due to heterogeneity among tumors. As a result, there is tremendous need to develop reliable, non-invasive computational methods for glioma classification. Prior research has found that machine learning techniques, combined with molecular and genetic markers, can successfully differentiate glioma types with high accuracy. Such classification allows earlier and more personalized treatment options. Many of the most accurate approaches reported in the literature rely on ensemble-based methods that integrate multiple, complex models. Although effective, these ensemble approaches are often computationally expensive, difficult to interpret, and are less practical for real-time or resource-constrained clinical settings. In this study, we develop and evaluate a Random Forest Classification Model for glioma classification. The results of the study indicate that this approach achieves accuracy that is comparable to more complex ensemble methods, while offering improved computational efficiency, interpretability, and better potential for clinical utilization.