Unruptured brain arteriovenous malformations causing seizures localize to one common brain network

Abstract

Seizures are a frequent symptom of unruptured brain arteriovenous malformations (bAVMs). However, the brain regions responsible for these seizures remain unclear. To identify the brain regions causally involved in bAVM-related seizures, we retrospectively reviewed 220 patients with unruptured bAVMs. Using voxel-based lesion-symptom mapping (VLSM) analyses, we tested whether individual brain regions were associated with unruptured bAVM-related seizures. The result revealed that unruptured bAVMs causing seizures are anatomically heterogeneous at the voxel level. Subsequently, lesion network mapping (LNM) analyses was performed to determine whether bAVMs causing seizures belonged to a distributed brain network. LNM analyses indicated that these lesions were located in a functional network characterized by connectivity to the left caudate and precuneus. Moreover, the discrimination performance of the identified seizure network was evaluated in discovery set by calculating the individualized network damage score and was tested in validation set. Based on the calculated network damage scores, patients were divided into low-, medium-, and high-risk groups. The prevalence of seizures significantly differed among the three risk categories in both discovery (p = .003) and validation set (p = .004). Finally, we calculated the percentage of voxels in the canonical resting-state networks that overlapped with the seizure-susceptible brain regions to investigate the involvement of resting-state networks. With an involvement percentage over 50%, the frontoparietal control (82.9%), limbic function (76.7%), and default mode network (69.3%) were considered to be impacted in bAVM-related seizures. Our study identified the seizure-susceptible brain regions for unruptured bAVMs, which could be a plausible neuroimaging biomarker in predicting possible seizures.

Keywords: brain arteriovenous malformations; lesion network mapping; seizures; voxel-based lesion-symptom mapping.

FIGURE 1

The workflow for identifying the bAVM‐related seizure network. (a) Functional connectivity between each lesion location and the rest of the brain was computed using resting‐state functional magnetic resonance imaging data from 358 healthy control subjects. Individual lesion network maps were thresholded, binarized, and overlapped to identify common connections across the lesion locations. (b) Specificity analyses was used to identify connections specific to patients with lesions causing seizures versus those without. (c) Conjunction analyses identified the regions of interest (ROIs) whose connectivity was both sensitive and specific to lesions causing seizures. (d) The overlap between the positive and negative connectivity patterns to the ROIs defining the set of brain regions in the seizure network.

FIGURE 2

Lesion locations were not associated with seizures. (a) Lesion overlay map showing the percentage of patients among the three groups that overlapped in a particular voxel. Lighter colors indicate voxels where a larger proportion of patients had bAVMs. (b) Voxel lesion‐symptom mapping (VLSM) identified no voxels significantly associated with seizures (FDR‐adjusted p < .05).

FIGURE 3

BAVMs causing seizures mapped to a common brain network. (a) LNM analyses indicated that lesion locations causing seizures were positively (red) correlated with the superior frontal gyrus, middle frontal gyrus, inferior parietal lobule, inferior temporal gyrus, basal ganglia, etc., and negatively (blue) correlated with the medioventral occipital cortex, postcentral gyrus, lateral occipital cortex, precentral gyrus, superior frontal gyrus, etc. Lighter colors indicate voxels where a larger proportion of individual lesion networks overlapped. (b) Positive connectivity to the bilateral caudate and negative connectivity to the precuneus were specific to patients with bAVM‐related seizures compared to patients without seizures. (c) The conjunction analyses indicated that connectivity (positive in red and negative in blue) to the left caudate and the precuneus was both sensitive and specific for lesions causing seizures. (d) The overlap in brain regions negatively connected to the precuneus and positively connected to the left caudate defined the brain regions of the seizure network.

FIGURE 4

Validation of the seizure network. (a) In the discovery set, the percentages of patients with seizures in the low‐, medium‐, and high‐risk groups were 31.8% (7/22), 51.1% (46/90), and 72.5% (37/51), respectively. A significant difference was identified among the three groups based on the Chi‐square test (χ ²  = 11.680, p = .003). (b) In the validation set, the percentages of patients with seizures in the low‐, medium‐, and high‐risk groups were 10.0% (1/10), 48.7% (19/39), and 87.5% (7/8), respectively. A significant difference was validated among the three groups (χ ²  = 10.783, p = .004). (c) Overlap (yellow) between the regions in the seizure network (blue) and the lesions selected from the validation set (red). The patients with and without seizures were sorted by lesion size. The four patients with the smallest lesions and the four with the largest lesions among the patients with and without seizures were selected for illustrative purposes.