Abstracts

This abstract is a recipient of the Young Investigator Award

Rationale: Imaging studies of temporal lobe epilepsy (TLE) often focus on canonical networks; while such studies often detect regional pathology congruent with semiology, TLE-specific networks remain ill-defined. Independent component analysis (ICA) can detect networks as multi-variate co-occurrence patterns across a volume. This is distinct from mass-univariate analytics typically applied to voxel-based morphometry (VBM), voxel-based physiology (VBP), and task activation (TA) data, though still applicable to primary data and coordinate based meta-analysis (CBMA). Activation likelihood estimation (ALE), the most common mass-univariate CBMA, has shown in TLE that hippocampal and medial dorsal nucleus (MDN) pathology mirror healthy functional connections (Network Degeneration Hypothesis). Support was also shown by meta-analytic ICA (meta-ICA) applied to TA/VBP data in trans-paradigm/diagnostic analyses (Smith 2009; Vanasse 2021). We applied meta-ICA to VBM/VBP reports of TLE-pathology, to infer TLE-specific network anomalies.

Methods: Meta-ICA was used to extract co-pathology networks as in TLE, computing co-occurrence of foci within and across studies. BrainMap tools were used to model 74 experiments (n=1599) as probability maps, converting left-rectified foci to spatial probability distributions, weighted by sample size. ICA was then applied (d=1&2) in compliance with published standards ( ≥ 20 experiments per result; Eickhoff 2016; Muller 2018). Regions of TLE pathology were also identified (ALE) as spatial convergence across studies, agnostic to within-study co-pathology. Results were compared via correlations (R).

Results: Two anatomically distinct TLE-networks were identified by meta-ICA d=2; excepting ALE peaks, no IC overlap was found. Network 1 (IC1) had cerebellar tonsil, precuneus, fusiform, supramarginal, precentral, insular, inf/mid temporal, contralesional MDN, precentral, bilateral lingual, occipital, frontal; similar (R=0.89) to IC1 at d=1 (Table 1). Network 2 (IC2) had mid frontal, sup temporal, contralesional parahippocampus, pulvinar, anterior nucleus, cuneus, bilateral paracentral, cingulate/frontal. Regions implicated in TLE by meta-ICA were a superset of those identified by ALE (IC1-tonsil; IC2-pulvinar, caudate, sup temporal; Both-hippocampus, MDN). Regions identified uniquely by meta-ICA were anterior nucleus, supramarginal, pre/paracentral, insula. Neither meta-ICA network matched canonical networks; IC1 loaded on language (speech-execution Z=6.95; cognition Z=4.10); IC2 loaded on emotion (positive reward Z=4.76) and cognition (attention Z=4.02; explicit memory Z=3.80). VBM/VBP distribution to ICs was homogenous (χ² p=0.07).

Conclusions: Both ICs fit TLE semiology. IC1 (verbal/visual) disruption can impair communication and cause visual hallucinosis; IC2 (limbic) aligns with pre/post-ictal social-emotional deficits; dyscognitive seizures disrupt IC2 cognition (attention/memory), impairing awareness/inducing postictal memory deficits. These findings (1) identify two novel TLE-networks; (2) highlight the first low-d meta-ICA detection of disease-network; and (3) have significant implications in biomarker development and pre-surgical seizure mapping.

Funding: R01MH074457, T32GM113896