Interictal spike localization for epilepsy surgery using magnetoencephalography beamforming

https://doi.org/10.1016/j.clinph.2020.12.019Get rights and content

Highlights

  • Clinical validation of beamforming for MEG spike localization was evaluated by surgical outcome in 13 drug-resistant focal epilepsy patients.

  • Beamforming can predict seizure-free surgical outcome at the lobar level.

  • Beamforming tends to localize the propagated cortical sources of MEG spikes modelled by sLORETA.

Abstract

Objective

Magnetoencephalography (MEG) kurtosis beamforming is an automated localization method for focal epilepsy. Visual examination of virtual sensors, which are source activities reconstructed by beamforming, can improve performance but can be time-consuming for neurophysiologists. We propose a framework to automate the method and evaluate its effectiveness against surgical resections and outcomes.

Methods

We retrospectively analyzed MEG recordings of 13 epilepsy surgery patients who had one-year minimum post-operative follow-up. Kurtosis beamforming was applied and manual inspection was confined to morphological clusters. The region with the Maximum Interictal Spike Frequency (MISF) was validated against prospectively modelled sLORETA solutions and surgical resections linked to outcome.

Results

Our approach localized spikes in 12 out of 13 patients. In eight patients with Engel I surgical outcomes, beamforming MISF regions were concordant with surgical resection at overlap level for five patients and at lobar level for three patients. The MISF regions localized to spike onset and propagation modelled by sLORETA in two and six patients, respectively.

Conclusions

Automated beamforming using MEG can predict postoperative seizure freedom at the lobar level but tends to localize propagated MEG spikes.

Significance

MEG beamforming may contribute to non-invasive procedures to predict surgical outcome for patients with drug-refractory focal epilepsy.

Introduction

Up to 85% of patients with drug-refractory epilepsy can achieve seizure freedom with surgical treatment (Spencer and Huh, 2008). Success depends on pre-surgical localization of the epileptogenic tissue and network, which indicate one or more confined cortical areas and the epileptogenic brain networks modulating the initiation and propagation of epileptic activity (Lüders et al., 2006, Zijlmans et al., 2019).

Because of limited recording duration, magnetoencephalography (MEG) cannot always capture ictal events. However, interictal spikes are more likely to be recorded, which are abnormal epileptic events between seizures. It is valuable to localize spikes or the ‘irritative zone’ in the pre-surgical evaluation of focal epilepsy patients (Rosenow and Lüders, 2001, Zijlmans et al., 2019). Studies have shown that localization of MEG seizures and interictal discharges can achieve similar accuracy (Mégevand and Seeck, 2018, Plummer et al., 2019, Ramanujam et al., 2017). MEG measures the magnetic field induced by electrical brain sources with millisecond-order temporal resolution (Meyer-Lindenberg, 2010). Compared to scalp electroencephalography, magnetic fields are less distorted by head tissue, particularly the skull.

Dipole modelling and current density reconstruction are the most commonly used methods of MEG source localization in clinical applications (Huang et al., 2004, Pascual-Marqui, 1999, Pascual-Marqui et al., 1994). Neurophysiologists visually annotate interictal spikes in continuous MEG recordings and inverse methods are modelled on single or averaged spikes to localize spike sources. However, annotation of spikes from MEG recordings can be time-consuming in the order of several hours per patient.

An alternative localization method is beamforming, which is a spatial filter that can reconstruct brain source activity from continuous MEG data by scanning the whole brain (Van Veen et al., 1997). Kurtosis beamforming has been proposed on the source time series on the scanned grid of beamforming (Kirsch et al., 2006, Robinson et al., 2004). Kurtosis detects the presence of outliers in data, which is an important characteristic of interictal discharges (Westfall, 2014). Kurtosis beamforming can be applied to continuous MEG data to localize spikes present in the data, which can reduce the requirement for neurophysiologists to perform time-consuming spike annotations.

Beamforming can reconstruct brain activity at the source level using ‘Virtual Sensors’ (VSs), which are analogous to implanted sensors at particular locations. We consider VSs located at the local maxima of kurtosis distributions as source candidates. It has been widely accepted in practice that visual inspection of VSs for genuine interictal spikes can improve localization performance, since artifacts and other brain waves, such as alpha activity, can also lead to high values of kurtosis. Agirre-Arrizubieta et al. (2014) and de de Gooijer-van de Groep et al. (2013) visualized all the samples on VSs over a threshold of the peak-to-root mean square ratio to detect spikes in VSs. Hall Michael et al., 2018, Rose et al., 2013, and Tenney et al. (2014) visualized VSs along with the corresponding MEG sensor data to include VSs that showed synchronized spikes with MEG/EEG sensors. Kirsch et al. (2006) visualized VSs to reject muscle artifacts. Oishi et al. (2006) performed visual inspection on VSs to reject non-epileptiform MEG epochs. However, the examination of VSs can be nearly as tedious as manual annotation of spikes at the MEG sensor level. In this study, we developed a framework to reduce manual inspection and localize interictal spikes using beamforming.

Prior source imaging work has shown that the irritative zone does not necessarily indicate regions or networks that need to be removed or altered to achieve seizure freedom (Bartolomei et al., 2016, Lüders et al., 2006, Plummer et al., 2019, Rosenow and Lüders, 2001, Zijlmans et al., 2019). Other biomarkers are needed to refine the extended irritative zone defined by the brain regions with genuine spikes present. Intracranial studies have shown that quantification of spikes could refine the irritative zone for pre-surgical evaluation. Goncharova et al. (2013) showed that the spatial distribution of spike rates is associated with the seizure onset zone, which is the area of the cortex that generates clinical seizures. Krendl et al. (2008) showed that the absolute spike rate can predict surgical outcome. Conrad et al. (2020) showed that the intracranial electrode with the highest spike frequency can localize the seizure onset zone. It has also been found that quantification of scalp EEG can aid interictal source localization. Coutin-Churchman et al. (2012) showed that dipole fitting localization from the spike cluster with the largest quantity of interictal discharges correlated with the outcome of surgical resection. We hypothesised that MEG can contribute to source localization for epilepsy surgery if spike frequency is factored into beamforming localization.

We propose a paradigm that only requires limited manual review and we investigated spike frequency as a potential biomarker to help refine the irritative zone. Beamforming localization results were validated against prospectively determined standardized low-resolution brain electromagnetic tomography (sLORETA) solutions from our earlier publication (Plummer et al., 2019) and with surgical resection linked to outcome.

Section snippets

Patients and MEG recordings

We retrospectively studied MEG data of 13 patients from a cohort of over 80 refractory focal epilepsy patients, referred for recordings from August 2013 to October 2015. The 13 patients studied here are the patients who underwent subsequent resection surgeries and had a minimum 1 year follow-up (range 14–39 months, median 21 months) (Plummer et al., 2019). The selection criteria for MEG recordings in the pre-surgical evaluation were: no lesions or complex lesions presented on MRI; discordant

Results

Patient demographics and standard clinical work-ups can be found in Table 1 of Plummer et al. (2019). Of the 13 patients, six were female and seven were male. Their median duration with epilepsy until surgery was 20 years (range 4–33 years). Ten patients showed normal MRI images; the other three had complex lesions visible on MRI. The standard workups did not reach agreement on the localized sources in all patients. The seizure reduction rates were evaluated with a minimum follow-up period of

The effectiveness of the method

This study proposed a method to reduce the amount of visual inspection required for post-processing procedures of kurtosis beamforming. The proposed approach only requires manual inspection of the representative discharges extracted from the virtual sensor with the maximum kurtosis in each beamforming segment. In previous studies, visual inspection was performed on continuous virtual sensor data to ensure the presence of authentic interictal spikes rather than artefacts (Agirre-Arrizubieta et

Limitations

The method proposed in this study has limitations. The method failed to detect spikes on VSs in one patient (Patient 8). The probable reason is that we only included one VS with the maximal kurtosis in each beamforming 3-minute segment in the subsequent analysis. The limited inclusion of VSs may contribute to failure to detect spikes, and it may only reveal incomplete spatial distributions of spike frequencies. In future work, it will be important to include more VSs rather than only the first

Conclusion

The framework proposed in this study can reveal authentic interictal spikes localized by beamforming effectively with limited visual inspection. In non-lesional and complex lesional cases, beamforming can localize putative epileptogenic regions at the lobar level and predict postoperative seizure outcome. Beamforming can help guide and validate the localization of sLORETA-based modelling as part of the much-needed improvement in the non-invasive work-up of patients with drug-refractory focal

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Acknowledgements

The authors acknowledge Associate Professor Michael A. Murphy, who performed resection surgeries. The authors acknowledge the facilities and the scientific and technical assistance of the National Imaging Facility at the Swinburne Node, Swinburne University of Technology, with particular thanks to Dr Rachel Batty, Ms Mahla Cameron-Bradley, and Ms Johanna Stephens for their technical support. We acknowledge the Australian National Imaging Facility for the support of W. Woods and the MEG system

References (34)

  • T. Caliński et al.

    A dendrite method for cluster analysis

    Commun. Stat. Theory Methods

    (1974)
  • E.C. Conrad et al.

    Spatial distribution of interictal spikes fluctuates over time and localizes seizure onset

    Brain

    (2020)
  • de Gooijer‐van de Groep KL, Leijten FSS, Ferrier CH, Huiskamp GJM 2013. 'Inverse modeling in magnetic source imaging:...
  • Engel Jr J, Van Ness PC, Rasmussen TB, Ojemann LM. 1993. 'Outcome with Respect to Epileptic Seizures'. In: Engel J, Jr...
  • B.H. Hall Michael et al.

    An evaluation of kurtosis beamforming in magnetoencephalography to localize the epileptogenic zone in drug resistant epilepsy patients

    Clin. Neurophysiol.

    (2018)
  • M.X. Huang et al.

    Commonalities and differences among vectorized beamformers in electromagnetic source imaging

    Brain Topogr.

    (2004)
  • R. Krendl et al.

    Absolute spike frequency predicts surgical outcome in TLE with unilateral hippocampal atrophy

    Neurology

    (2008)
  • Cited by (4)

    • Comparison of beamformer and ICA for dynamic connectivity analysis: A simultaneous MEG-SEEG study

      2023, NeuroImage
      Citation Excerpt :

      Indeed, since epilepsy is mainly a disease of brain networks (Bartolomei et al., 2017), MEG should be able to define, in a non-invasively way, the spatio-temporal organization of such networks. In order to improve the non-invasive analysis of interictal activities, different techniques of brain source separation and localization have been explored to extract the spatio-temporal dynamics of the underlying network in M/EEG (Hall et al., 2018; Hassan et al., 2017; Li et al., 2021; Malinowska et al., 2014; Wilenius et al., 2020). In 1999, Kobayashi (Kobayashi et al., 1999) proposed for the first time the use of the independent component analysis (ICA) to separate the interictal components from the background EEG activity and applied the dipole source modeling approach without the need for averaging.

    • Magnetoencephalography for Epilepsy Presurgical Evaluation

      2024, Current Neurology and Neuroscience Reports
    View full text