Node abnormality predicts seizure outcome and relates to long-term relapse after epilepsy surgery

Abstract

Objective:We assessed pre-operative structural brain networks and clinical characteristics of patients with drug resistant temporal lobe epilepsy (TLE) to identify correlates of post-surgical seizure outcome at 1 year and seizure relapses up to 5 years. Methods:We retrospectively examined data from 51 TLE patients who underwent anterior temporal lobe resection (ATLR) and 29 healthy controls. For each patient, using the pre-operative structural, diffusion, and post-operative structural MRI, we generated two networks: 'presurgery' network and 'surgically-spared' network. The pre-surgery network is the whole-brain network before surgery and the surgically-spared network is a subnetwork of the pre-surgery network which is expected to remain unaffected by surgery and hence present post-operatively.Standardising these networks with respect to controls, we determined the number of abnormal nodes before surgery and expected to remain after surgery. We incorporated these 2 abnormality measures and 13 commonly acquired clinical data from each patient in a robust machine learning framework to estimate patient-specific chances of seizures persisting after surgery. Results:Patients with more abnormal nodes had lower chance of seizure freedom at 1 year and even if seizure free at 1 year, were more likely to relapse within five years. In the surgically-spared networks of poor outcome patients, the number of abnormal nodes was greater and their locations more widespread than in good outcome patients. We achieved 0.84 ± 0.06 AUC and 0.89 ± 0.09 specificity in detecting unsuccessful seizure outcomes at 1-year. Moreover, the model-predicted likelihood of seizure relapse was significantly correlated with the grade of surgical outcome at year-one and associated with relapses up-to five years post-surgery. Conclusion:Node abnormality offers a personalised non-invasive marker, that can be combined with clinical data, to better estimate the chances of seizure freedom at 1 year, and subsequent relapse up to 5 years after ATLR.

Multivariate white matter alterations are associated with epilepsy duration

“…Our finding that patients with poorer surgical outcomes were significantly further from controls than patients with seizure-free outcomes suggests a predisposing factor to surgical treatment success. This agrees with a large number of recent studies suggesting that pre-operative diffusion metrics may be predictive of post-surgical outcomes (Bonilha et al, 2006;Bonilha et al, 2006;Munsell et al 2015;Keller et al 2017;Sinha et al 2019;Taylor et al, 2018).…”

“…All subjects underwent diffusion weighted MRI acquisition on the same scanner, 3T GE Signa Excite HDx, as described previously (Sinha et al, 2020;Taylor et al, 2018;Winston et al, 2013). Diffusion MRI data were acquired using a cardiac-triggered single-shot spin-echo planar imaging sequence (Wheeler-Kingshott et al, 2002) with echo time =73 ms. Sets of 60 contiguous 2.4 mm-thick axial slices were obtained covering the whole brain, with diffusion sensitising gradients applied in each of 52 noncollinear directions (b value of 1,200 mm 2 s − 1 [δ = 21 ms, Δ = 29 ms, using full gradient strength of 40 mT m − 1]) along with 6 nondiffusion weighted scans.…”

Removal of Interictal MEG-Derived Network Hubs Is Associated With Postoperative Seizure Freedom

“…Given that all brain networks (epileptogenic or otherwise) have a mixture of high and low strength nodes, we interpret that seizures are facilitated by high strength nodes, but that not all high strength nodes are necessarily pathological. The normalization of patient networks against those from controls allows for the identification of pathological "abnormal" nodes (27,45). Future studies should investigate these relationships between node hubness and node abnormality.…”

The sensitivity of network statistics to incomplete electrode sampling on intracranial EEG

“…It is conceivable that a network metric would demonstrate low reliability to subsampling as defined in this paper, but still generate consistent predictions about the optimal site of surgical intervention depending on how these predictions are formed. Different groups have proposed different methods to direct surgical targeting using network statistics (Kini et al, 2019;Proix et al, 2018;Shah, Bernabei, et al, 2018;Sinha et al, 2017Sinha et al, , 2019, and our proposed subsampling method can be used to test the sensitivity of any targeting predictions to incomplete spatial sampling, similar to the jackknife procedure described in the section above.…”

“…In this study we seek to rigorously assess the extent to which different network metrics are sensitive to intracranial electrode sampling. Our goal is not to determine which, if any, network statistic correctly localizes the seizure onset zone or predicts surgical outcome, as this important work is currently under way by several groups (Kini et al, 2019;Proix, Jirsa, Bartolomei, Guye, & Truccolo, 2018;Shah, Bernabei, et al, 2018;Sinha et al, 2017Sinha et al, , 2019. Rather, our goal is to determine (a) whether and how incomplete spatial sampling affects the practical utility of network statistics, and (b) how sensitivity to spatial sampling can estimate patient-specific uncertainty in network model predictions.…”