Scalp ripples as prognostic biomarkers of epileptogenicity in pediatric surgery

Eleonora Tamilia, Matilde Dirodi, Michel Alhilani, P. Ellen Grant, Joseph R. Madsen, Steven M. Stufflebeam, Phillip L. Pearl, Christos PapadelisAnnals 2020 · DOI

TL;DR

This study demonstrates that non-invasive scalp electroencephalography (EEG) and magnetoencephalography (MEG) can accurately pinpoint the brain regions responsible for epileptic seizures in children. Specifically, "ripples" that co-occur with spikes (ripples-on-spikes) identified by these methods are strong indicators of the epileptogenic zone, and their successful surgical removal predicts a good outcome.

What Problem Does This Paper Address?

For children with medically refractory epilepsy, surgery to remove the epileptogenic zone (EZ) is the best hope for seizure freedom. However, precisely identifying this EZ is challenging. Traditionally, this involves invasive intracranial EEG (icEEG), which carries risks and has limited coverage. High-frequency oscillations (HFOs), particularly ripples (80-250 Hz), are known to be strong indicators of the EZ, but their detection and localization from scalp recordings have been difficult due to their low amplitude. This paper aims to bridge this gap by investigating whether scalp-recorded ripples, using advanced HD-EEG and MEG, can reliably localize the EZ and serve as prognostic biomarkers for surgical success, potentially reducing the need for invasive procedures.

Methods

The researchers employed a retrospective analysis of 20 children who underwent epilepsy surgery. They utilized high-density EEG (HD-EEG) and magnetoencephalography (MEG) to record brain activity.

  • Ripple Detection: A validated automated detector was used for ripple detection on both HD-EEG and MEG data, followed by human visual review to ensure accuracy and reject artifacts.
  • Source Localization: Both electric source imaging (ESI) for HD-EEG and magnetic source imaging (MSI) for MEG were used to pinpoint the cortical generators of detected ripples and spikes. Crucially, for co-occurring ripples and spikes, separate source localization was performed for each event type (ripples on 80-160 Hz data, spikes on 1-70 Hz data) to determine if they originated from distinct brain areas.
  • Categorization: Ripples were categorized as "ripples-on-spike" (co-occurring with spikes) or "ripples-alone" (independent of spikes). Similarly, spikes were categorized as "spikes-on-ripple" or "spikes-alone."
  • Validation: The localization of scalp-recorded ripples was compared against the gold standard of intracranial EEG (icEEG) ripples. Furthermore, the study assessed the prognostic value of resecting these localized ripple and spike sources by correlating it with surgical outcomes (Engel classification).

Datasets

The study used data from 20 pediatric patients with refractory epilepsy who underwent epilepsy surgery at Boston Children's Hospital. These patients had pre-operative HD-EEG and MEG recordings, as well as long-term icEEG monitoring. The data was collected between June 2011 and July 2016. No new benchmark datasets were introduced; rather, existing patient data was analyzed.

Key Insights and Ideas

  • Scalp Ripples are Localizable: The study provides strong evidence that non-invasive scalp HD-EEG and MEG can precisely localize the cortical generators of interictal ripples, showing high concordance with icEEG-defined ripple zones. This is a significant step towards using non-invasive methods for EZ localization.
  • Ripples and Spikes Often Have Distinct Sources: A key finding is that when ripples and spikes co-occur in time, their cortical generators are frequently spatially distinct (over 80% of cases). This suggests that these events, while temporally linked, may arise from different neuronal populations or mechanisms.
  • "Ripples-on-Spikes" are the Key Biomarkers: The study highlights that "ripples-on-spikes" are the most informative type of scalp ripple for predicting surgical success. Resecting the brain areas generating these specific ripples significantly improved the likelihood of good surgical outcomes.
  • "Ripples-Alone" are Likely Non-Epileptogenic: In contrast, "ripples-alone" were found to be less associated with epileptogenicity. Their generators were often located in areas that were spared during surgery and did not predict surgical outcome, suggesting they might reflect physiological events rather than seizure-generating tissue.

Key Results

  • Localization Accuracy: Scalp HD-EEG and MEG localized ripple sources with high precision, showing 79% and 83% concordance with icEEG-defined ripple zones, respectively.
  • Distinct Generators: In 83-84% of co-occurring ripple-spike events, the ripple and spike generators were spatially distinct.
  • Prognostic Value of Ripples-on-Spikes: Resection of ripple-on-spike sources predicted good surgical outcome with 90% accuracy for HD-EEG and 86% for MEG.
  • Prognostic Value of Spikes: Resection of spike sources (spikes-on-ripples) also predicted good outcome for HD-EEG (87% accuracy), but not for MEG.
  • Low Prognostic Value of Ripples-Alone: Missed resection of ripple-alone sources did not predict surgical outcome.
  • Detection Rates: Sixteen out of twenty patients showed scalp ripples. HD-EEG detected more ripples per minute than MEG.

Limitations and Future Work

  • Sample Size and Etiology: The study acknowledges the small sample size and the heterogeneity of etiologies, suggesting that larger prospective studies are needed to validate the predictive model and adjust for these factors.
  • Follow-up Duration: A minimum 2-year postoperative follow-up would have been ideal.
  • Patient Selection Bias: The study excluded patients who did not have pre-surgical HD-EEG/MEG or long-term icEEG, potentially limiting findings to neocortical epilepsy cases.
  • Non-Simultaneous icEEG: The icEEG was not simultaneous with HD-EEG/MEG, which prevented a direct investigation of exact correspondence.
  • Sleep vs. Wakefulness: The study did not differentiate between sleep and wakefulness during recordings, which can influence ripple activity. Future work should examine if spatial specificity of ripples is independent of vigilance state.
  • Low Detectability: The low detectability of scalp ripples remains a challenge, necessitating longer recordings, optimized protocols, and improved instrumentation for high-frequency recordings.

The authors suggest that their findings are promising for guiding icEEG implantation and facilitating prognosis in pediatric epilepsy surgery.

Summary updated

Abstract

HD-EEG and MEG localize interictal ripples with high precision in children with refractory epilepsy. Scalp ripples-on-spikes are prognostic, noninvasive biomarkers of epileptogenicity, since removing their cortical generators predicts good outcome. Conversely, scalp ripples-alone are most likely generated by non-epileptogenic areas.
BibTeX
@article{tamilia2020scalp,
  author = {Eleonora Tamilia and Matilde Dirodi and Michel Alhilani and P. Ellen Grant and Joseph R. Madsen and Steven M. Stufflebeam and Phillip L. Pearl and Christos Papadelis},
  title = {{Scalp ripples as prognostic biomarkers of epileptogenicity in pediatric surgery}},
  journal = {Annals},
  year = {2020},
  doi = {10.1002/acn3.50994},
}
Figure