INTRODUCTION: Gliomas affect the whole brain through structural and functional disconnection, causing widespread network modifications¹. Resting-state fMRI (rs-fMRI) is an established technique to display and characterize different brain networks in the healthy subject, as well as to study their modification in response to brain disorders². In brain tumors, resting-state connectivity shows widespread changes including loss of network efficiency and communication^1,3. However, the characterization of these effects in individual networks is still a topic of investigation. This study aims at assessing the impact of gliomas on the cortical synchronization of multiple brain networks generated from resting-state fMRI (rs-fMRI). We hypothesized a different vulnerability of brain networks depending on underlying function (characteristic role of the network), tumor location and grade.
METHODS: The study received IRB approval. We recruited 147 glioma patients (89M, 50.95±16.13y, 92 high-grade, 55 low-grade) and 200 healthy controls (HCs) with rs-fMRI. The dataset included FLAIR, T1-weighted (T1w), and resting-state sequences acquired with TR = 2500 ms and 160 volumes. The subject data was pre-processed to eliminate motion artifacts and aligned to a common standard space (MNI152) using SPM12 on MATLAB⁴. Fully automated glioma segmentation was performed using the DeepSeg tool in 3Dslicer⁵. Tumor boundaries were segmented using a pre-trained deep learning model validated with the MICCAI Brain Tumor Segmentation Challenge 2020 (BraTS) dataset⁶. The Group ICA of fMRI Toolbox (GIFT) was used to extract the independent components (ICs), and a reference template was created using data from 200 healthy controls through Group Independent Component Analysis (G-ICA)^7,8. This template served as a guide during Group information-guided ICA (GIG-ICA) to extract the corresponding independent components for each subject and label them with their respective resting-state networks (RSNs) using NeuroMark fMRI 1.0 atlas^9-11. To determine alterations of specific ICs in single patients, we computed the cosine similarity (CS) between the two vectors as in equation (Fig 1), and ran a permutation test to compare, for each IC, the CS of each patient against the distribution of CS values in HCs¹². The alterations were identified using a thresholding technique based on the standardization in Fig 2. The information regarding the alteration of ICs was subsequently used to conduct the contingency analysis with a chi-squared test to verify the significance of these alterations, as well as tumor location and grade. The statistical threshold was set at p<0.05.
RESULTS: Out of the 20 network components in HCs, 10.38±1.43 resulted altered in patients, including cognitive control network (CCN); default mode network (DMN); sensorimotor network; visual network (VN). Auditory network and subcortical network did not show any significant difference. The most pronounced effect was noted in the CCN network, with significant alteration of every component (3/3). Additionally, CCN showed significant alterations with tumors in the temporal lobe (p=0.005), Broca's (p=0.01), and Wernicke's area (p=0.041). Tumors in Wernicke's area also altered the DMN (p=0.04). Networks alterations persisted with increased distance from the tumor and were more pronounced with higher WHO-grade (p<0.001).
DISCUSSION: In this study, we developed a methodology to study the effects of gliomas on brain networks, utilizing G-IGA and GIG-ICA for analysis. This methodology can quantify the alteration induced by the tumor at the individual network level, allowing to image the functional damage through rs-fMRI in each subject. Our results indicate specific vulnerability of cognitive networks to tumor growth. Functional alterations extend beyond tumor boundaries, and increase with WHO-grade. Tumor location in known eloquent areas exerts widespread effects on brain networks. Limitations of the study include retrospective design and lack of complete neuropsychological testing.
CONCLUSION: Understanding specific network abnormalities in patients with brain tumors is crucial for managing cognitive disabilities before and after surgery.

Acknowledgements

No acknowledgement found.