Left-hemisphere glioma drives systematic patterns of contralesional functional connectivity.

Abstract

Gliomas can cause changes in functional networks both proximal and distal to the lesion. Understanding glioma-induced functional reorganization has implications for understanding variability across patients in cognitive outcomes, disease progression, and survival. Here, we leverage machine learning techniques to show that left-hemisphere gliomas are associated with systematic changes in right-hemisphere connectivity. We analyzed right-hemisphere functional connectivity patterns from resting-state functional MRI in 48 patients with left-hemisphere gliomas (mean age 50 years, 31 males) and 107 neurotypical controls (mean age 49 years, 44 males). We employed machine learning techniques, including support vector machines, to assess whether the pattern of right-hemispheric resting-state functional connectivity could distinguish left-hemisphere glioma patients from controls, and predict glioma characteristics, including isocitrate dehydrogenase mutation, World Health Organization grade, and relative size. A support vector machine binary classifier distinguished patients from controls based on right-hemisphere connectivity with 89% accuracy and 84% precision (both P = 0.001), indicating consistent contralesional connectivity differences as a function of glioma. The model also achieved 79% sensitivity for detecting patients (P = 0.028). Furthermore, patients with similar right-hemisphere connectivity profiles had lesions in similar locations within the left hemisphere, suggesting that the observed connectivity changes are influenced by glioma location. Additionally, the pattern of right-hemisphere connectivity could predict the presence of left-hemisphere gliomas specifically in regions of the parietal lobe. We also found that distinct contralesional connectivity patterns classified glioma molecular subtypes, achieving 78% accuracy in classifying patients by isocitrate dehydrogenase mutation (P = 0.004), with 82% precision (P = 0.003) and 73% sensitivity (P = 0.048) for mutant-tumors. However, right-hemisphere functional connectivity could not distinguish patients based on their tumor grade or relative size, with models performing no different from chance. These findings provide evidence for systematic changes in the contralesional connectome in glioma patients, consistent with theories of glioma-induced functional reorganization. This highlights the right hemisphere's role in adaptive responses to left-hemispheric gliomas and further underscores the importance of molecular profiling and tumor location in understanding reorganization potential.