Abstract
BACKGROUND
Lesion network mapping (LNM) is a method used to identify potential networks that can be ascribed to particular neurological functions/ deficits. LNM has yet to be implemented for large brain lesions such as tumors. OBJECTIVES: To apply LNM for potential identification of vulnerable epileptogenic networks in tumors causing medically-refractory epilepsy (MRE), compared with non-epileptogenic tumors.
METHODS
MRE and non-epileptogenic lesions were normalized to standard space for group analysis. These were used as a seed in high-resolution normative resting state fMRI, which was then transformed to t-maps and thresholded by t = 5.1; this corrected for multiple comparisons (Bonferroni corrections) across the whole brain at pcor < 0.05. The statistically-significant thresholded maps were binarized and summed connectivity maps were generated for both groups. This allowed computation of voxel-wise odds ratios (VORs) in order to identify voxels that were more likely associated with tumors that either did or did not result in MRE.
RESULTS
Twenty-seven patients were included. Eleven brain metastases with no history of seizures, M/F: 5/6, mean age 68.4+/-8.4 years, and 16 had MRE (10 low-grade glioma, 2 cavernoma, 3 “other”), M/F: 7:9, mean age 33.7 +/-12.2 years. Lesions causing MRE were preferentially located in the cingulate gyrus, calcarine fissure, parahippocampal gyrus and lateral temporal neocortex. The resting-state networks that were >1.5x likely to be connected with MRE lesions were the salience, executive control, and dorsal default mode networks.
CONCLUSION
In this proof of concept study, we have demonstrated that (1) in addition to stroke, tumors may also be amenable to LNM and (2) the underlying normative neural circuitry may in part explain the propensity of particular lesions toward development of MRE. This has ramifications in patient counseling and surgical management planning, as earlier surgery could be applied for lesions thought to be more prone to development of MRE.
Group Analysis of Resting-State fMRI by Hierarchical Markov Random Fields
