Abstracts

Rationale: Predicting seizure outcome (SO) after anterior temporal lobectomy (ATL) is a major clinical goal. With clear evidence that even focal epilepsies disrupt large scale brain networks, resting-state functional connectivity (FC) methods have been increasingly used on a pre-surgical basis to characterize the impact of seizures on brain activity. In this project, we sought to determine whether the functional integrity of resting-state networks (RSNs) prior to surgery can discriminate between patients who obtain good versus poor seizure control after ATL.Methods: We collected 5-minute resting state fMRI data on 46 refractory adult unilateral TLE patients prior to their brain surgery (ATL). SO at least six months post-surgery were identified (“good” outcome, GO= no seizures since surgery, Engel Class I, n=35; “poor”=at least one seizure after surgery, Engel Class≥II, n=21) (Engel et al., 1993). Twenty-three RSNs obtained from a large sample of normal healthy participants (Doucet et al., 2011) were used as normative templates. For each network and each participant, an “intra-connectivity coefficient” (ICC) was computed as the average correlation between the time-series of all the voxels within the RSN and the averaged time-series within the RSN. A high value indicates strong temporal coherence between the voxels of the network while a low value indicates that the network is not functionally stable. Lastly, a logistic regression was computed to predict seizure outcome, using the 23 RSN ICCs as the continuous independent variables.Results: Using the ICCs of the 23 RSNs, the regression model was able to accurately classify 82.1% of the population (62.5% against chance classification). This model explains 48% of the variance (Cox & Snell R2). Among the networks, three were significant (RSN7, 11 and 23), all involved lateral frontal-parietal regions (Fig. 1). For RSN7 and 11, higher ICC was associated with GO, while lower ICC of the RSN23 reflected GO. When restricting the regression model to these 3 RSNs, the model accurately classified 71.4% of the patients and explained 22% of the variance. Separately for each of these 3 networks, we then tested whether the FCs between the regions or ICC for each region of the RSN were better predictors of SO. The FC measures between regions were, in no case, predictive of SO (p>0.05). In contrast, the ICCs within the right inferior temporal cluster of RSN7 (p=0.028, higher ICC associated with GO) and the left middle frontal cluster of the RSN23 (p=0.014, lower ICC associated with GO) were significant predictors of SO.Conclusions: The results suggest that prior to surgery the functional network organization of the brain may differ for patients who experience a good vs poor outcome following ATL. Importantly, networks covering mesial temporal lobe regions were not predictors of SO. The networks most reliably predicting SO involved lateral frontal and temporal regions. These findings suggest that both weaker ICC in the frontal and stronger ICC in the lateral temporal cortex predict worst SO, potentially reflecting new epileptogenic foci.