Abstracts

Rationale: Literature has examined pre-surgical functional and structural neuroimaging features predictive of either neurocognitive or seizure control outcomes post-surgery. These two distinct outcomes are always considered independently. Here, we test for pre-surgical functional or structural biomarkers that are useful for predicting both outcomes.

Methods: 58 focal left temporal lobe epilepsy patients (age, m=42.1±17.6) who underwent ant. temporal lobectomy received pre-operative resting state (RS) fMRI and pre/post-surgical cognitive testing. Seizure outcome (SO) was based upon a modified Engel scale ( > 6 mo. post-surgery). Cognitive/psychiatric outcomes were based upon reliable change indices of episodic memory (CVLTII, delayed free recall), language (Phonemic Fluency), and depression (Beck Depression Inv.). Patients were classified as obtaining either “good” or “poor” outcomes. Random forest (RF) models identified features most important for distinguishing two outcome combinations (good/good=good cognitive [psychiatric] and SO; poor/poor=poor cognitive [psychiatric] and SO). T1 and RS scans were used to derive structural and functional measures in RF models (independent variables). Subject-specific independent component analysis of the RS data quantified: (1) spatial match of each patient’s individual component with 11 key intrinsic connectivity networks (ICNs, Fig 1), (2) average strength of the functional connectivity (FC) within each patient’s individual rendering of the ICNs, (3) segmented gray matter volume [FSL software] underlying the regions making up each patient-defined version of the ICNs, (4) “variance not explained” by the ICNs, capturing idiosyncratic, yet coherent network components systematically related to outcomes. Features with high RF variable importance ( >.60) are depicted in Fig 2.

Results: GM volume in networks outside the ictal temporal lobe were important to good combined outcomes. SO/psychiatric outcomes depended on GM integrity, with involvement of the cerebellum and non-ictal hemisphere. Deviation from the normative ICNs only mattered to combined SO/cognitive, not SO/psychiatric outcomes. SO/memory and SO/language outcomes relied on similar ICN measures. SO/language outcomes were distinguished by features not always reflective of high ICN integrity. SO/memory outcomes were best predicted by ictal-sided features, and a good match to the network aligned with episodic memory (Executive).

Conclusions: Combined good SO/memory and SO/language outcomes relied on pre-surgical networks not specifically dedicated the cognitive function measured post-operatively. This indicated compensatory cognitive reorganization was present before surgery (atypical cognitive architecture), with such architecture also supportive of good SO. These data revealed pre-surgical biomarkers predictive of both cognitive/psychiatric and SO. The integrity of both functional and structural ICN features mattered to combined outcomes, though good integrity did not always relate to good combined outcomes.

Funding: Please list any funding that was received in support of this abstract.: JIT, NIH/NINDS, R01 NS112816-01.