Editors' ChoiceEpilepsy

Epilepsy surgery: Think globally, act locally

See allHide authors and affiliations

Science Translational Medicine  05 Sep 2018:
Vol. 10, Issue 457, eaav0335
DOI: 10.1126/scitranslmed.aav0335

Abstract

A computational approach predicts the success of epilepsy surgery based upon presurgical structural connectivity.

For the 30% of epileptic individuals with medication-resistant seizures, surgical resection of the brain area responsible for seizure initiation (ictal focus) can be beneficial. Yet epilepsy surgery is not always curative: Up to one-third of patients continue to have seizures after surgery. Moreover, epilepsy surgery carries its own set of risks; many patients report post-surgical memory loss, visual field defects, and psychological sequelae, such as depression. Given these concerns, surgical candidates first undergo a multimodality workup that may involve anatomic and functional imaging, surface or intracranial electroencephalography (EEG), and neuropsychiatric testing, with the goal of mapping the ictal focus and surrounding region as accurately as possible. Epilepsy surgery is usually reserved for the subset of patients with a definable, easily accessible lesion.

For reasons not fully understood, even carefully selected surgical candidates who undergo technically optimal resections continue to have seizures. Gleichgerrcht et al. investigated whether analysis of pre-existing whole-brain networks could be used to predict surgical outcome. Working with existing imaging and clinical data from a cohort of surgically treated patients, the authors used a computational approach, training a deep network to classify surgery outcome based upon presurgical structural connectivity. The authors found that some of the connections associated with surgical outcome were remote from the seizure focus. Their computed predictions of seizure freedom based on structural connectivity were superior to those arising from a model based on standard clinical variables alone.

The predictive model built and evaluated in this study was derived from 50 individuals; therefore, whether its performance is reproducible in a larger cohort remains to be investigated. Furthermore, all study participants had temporal lobe epilepsy, a condition most common in adults and characterized by a specific site of seizure origin. Whether the networks implicated in surgical outcome in this cohort could be generalized to younger patients or to other forms of epilepsy also warrants further study. Despite the inherent limitation of its design, this work demonstrates that machine learning combined with imaging data could enable better risk stratification of prospective surgical candidates. Beyond that, it compels us to regard epilepsy as a condition involving both local and global brain dysfunction.

Highlighted Article

View Abstract

Navigate This Article