Functional network correlates of aphasia and compensation in brain tumor patients: A graph theoretical analysis of resting-state fMRI
TL;DR
Graph metrics should be interpreted as exploratory system-level correlates of aphasia status and preserved language rather than stand-alone predictors or definitive markers of compensation, according to resting-state fMRI graph-theoretical analysis.
Abstract
Objective
To characterize functional network alterations associated with aphasia and preserved language function in patients with left-hemispheric brain tumors using resting-state fMRI graph-theoretical analysis, with secondary evaluation of whether whole-brain network metrics are associated with aphasia status within the tumor cohort.
Materials And Methods
This retrospective IRB-approved study included 120 participants: 40 aphasic patients, 40 non-aphasic patients with left-hemispheric intra-axial tumors, and 40 matched healthy controls. ROI-to-ROI rs-fMRI connectivity across seven canonical networks yielded graph metrics (global/local efficiency, clustering, path length, centrality) at whole-brain, hemispheric, and network levels with FDR-corrected comparisons. Multinomial logistic regression compared healthy controls, non-aphasic patients, and aphasic patients. A secondary exploratory patient-only binary logistic regression examined aphasia status using whole-brain graph metrics and demographic covariates, including age, sex, and handedness.
Results
The whole-brain multinomial model did not significantly distinguish healthy controls, non-aphasic patients, and aphasic patients (χ²=21.622, df=18, p=0.249; classification accuracy=57.5%). Therefore, individual graph-metric coefficients from this model were treated as exploratory rather than confirmatory. In a secondary exploratory patient-only regression, the model did not significantly distinguish aphasic from non-aphasic tumor patients after adjustment for age, sex, and handedness (Omnibus χ²=7.147, df=10, p=0.712; Nagelkerke R²=0.114; Hosmer-Lemeshow p=0.508; accuracy=62.5%). Whole-brain graph metrics were not independently associated with aphasia after demographic adjustment. ROI-level analyses showed focal differences in non-aphasic patients, including greater left IFG closeness, posterior parietal centrality, and anterior cerebellar connectivity, but these findings were interpreted as exploratory network correlates rather than definitive evidence of compensation.
Conclusion
Rs-fMRI graph-theoretical measures characterized network-level differences among aphasic patients, non-aphasic patients, and healthy controls, but the three-group multinomial model did not significantly distinguish the groups and the secondary patient-only regression did not independently distinguish aphasic from non-aphasic tumor patients after demographic adjustment. These findings suggest that graph metrics should be interpreted as exploratory system-level correlates of aphasia status and preserved language rather than stand-alone predictors or definitive markers of compensation.