Abstracts

Rationale: The most common and often problematic cognitive comorbidity in patients with temporal lobe epilepsy (TLE) is memory impairment. Recent data suggest that individualized patterns of neural architecture based on brain imaging, i.e., the structural connectome (SC), can successfully predict memory impairment in a variety of neurological populations. However, no studies to date have used the SC to predict memory impairments in patients with TLE. In this study, we aimed to determine whether an individualized, white matter (WM) SC outperforms traditional tract-based measures, hippocampal volume (HCV), and clinical variables (CV) for discriminating TLE patients with memory-impairment from those with normal memory. Methods: Diffusion-weighted imaging (DWI), CVs, and neuropsychological data were available for 81 patients with TLE. Memory impairment was defined as >=1.5 standard deviations below the normative mean on at least 2 out of 3 tests of verbal memory. Fractional anisotropy values for five bilateral medial temporal lobe WM tracts implicated in memory were extracted from the DWI. SCs were derived by performing probabilistic tractography among cortical regions. Principal component (PC) analysis was performed on the SCs to reduce the dimensionality of the data to 10 PCs. XGBoost, a robust tree-based classifier, was used to test and compare the accuracy of four models derived using CV, tracts, HCV, and SC features. All models were trained on 48 patients from UCSD and tested on 33 patients from UCSF. Results: Both the SC (76% accuracy) and tracts (73% accuracy) were better classifiers of verbal memory performance than HCV (66% accuracy) or CV (61% accuracy) (all p < 0.05; see Table 1). Within the WM-based models, the SC performed similar to or better than the tract-based models, with slightly higher accuracy and positive predictive value (PPV; 0.81 for SC, 0.77 for tracts). Multivariate models were the most robust classifiers, providing the highest sensitivity (0.95 for Tracts+HCV) and specificity (0.67 for SC+HCV). However, the SC provided more specific anatomical information regarding the local patterns of abnormal connectivity that contributed the most to verbal memory performance (see Figure 1). These local connections included those from the left precuneus, left inferior temporal gyrus, and entorhinal cortex to other temporal and extratemporal regions, were most abundant in the left anteriomedial temporal lobe and may not have been captured in our tract-based analysis. Conclusions: SCs and tract-based measures appear to outperform HCVs and CVs for predicting memory impairment in TLE. However, network-based models combined with HCVs may provide the best classification accuracy and more anatomically precise information regarding the importance of local MTL connections to memory performance. These data could provide critical information for estimating individualized risk for memory decline following surgical interventions. Funding: Supported by NIH/NINDS R01 NS065838 (CRM) and R21 NS107739 (CRM; LB)