Motivation: Glioma classification, particularly between IDH-mutant astrocytomas (AC) and oligodendrogliomas (OG), is challenging due to overlapping metabolic profiles. Improved differentiation is crucial for accurate tumor identification and treatment planning. Goal(s): To enhance the classification of IDH-mutant AC and OG gliomas by identifying distinct metabolic patterns through advanced imaging and machine learning. Approach: The study analyzed 3D MRSI data from nine IDH-mutant glioma cases using UMAP and clustering metrics, focusing on core and non-core tumor regions. Results: Distinct clustering emerged, with OG showing greater cohesion and clear boundaries, confirmed by Silhouette, Davies-Bouldin, Calinski-Harabasz scores, and entropy measures, enhancing classification capabilities for IDH-mutant gliomas. Impact: This study demonstrates that metabolic imaging combined with unsupervised machine learning effectively differentiates astrocytomas and oligodendrogliomas. Insights into tumor heterogeneity and spatial complexity advance glioma classification and could guide more personalized, subtype-specific treatment strategies in neuro-oncology.
Ungan et al. (Tue,) studied this question.