How wrong can we be? The effect of inaccurate mark-up of EEG/fMRI studies in epilepsy

Abstract

Objective

The aim of this investigation was to determine the effect of inaccurate or inconsistent marking up of events in the EEG on statistical analysis of EEG/fMRI studies of patients with epilepsy.

Methods

EEGs obtained during EEG/fMRI studies conducted on 10 patients with epilepsy and six normal control subjects were reviewed. All clear epileptiform events were marked up in the patient EEGs, as were all small movement-related artefacts in the patient and control subject EEGs. We then considered the effect on the numbers of voxels above threshold in the resulting Statistical Parametric Mapping (SPM) analysis if events were omitted, mislabelled, or if event times were inconsistently marked up.

Results

Omitting true epileptiform events resulted in a decrease in the number of voxels that survive statistical threshold. Mixing epileptiform and non-epileptiform events in the SPM analysis generally (but not always) decreased the number of voxels that survived threshold. Inconsistent event mark-up had little effect if the inconsistency was small (<200 ms), but had more effect if it was large (>500 ms).

Conclusion

It is important to accurately mark-up EEGs acquired during EEG/fMRI studies in order to get the best results from subsequent analyses.

Significance

Our study reveals the consequences of inaccurate review of the EEG in EEG/fMRI studies and suggests guidelines for the review of EEG in these investigations which, if followed, should result in studies of acceptable quality.

Introduction

Statistical Parametric Maps (SPMs) based on Electroencephalography and functional Magnetic Resonance Imaging (EEG/fMRI) studies provide unique information about the spatial distribution of significant changes in cerebral Blood Oxygenation level Dependant (BOLD) signal correlated with the EEG event times. As these data become more widely available they are being introduced into studies of patients with epilepsy as an adjunct to the more commonly used localisation tools of scalp EEG, structural MRI, SPECT and PET studies (De Tiege et al., 2007a, De Tiege et al., 2007b, Liu et al., 2008, Manganotti et al., 2008, Zijlmans et al., 2007). The data produced by SPM analysis, however, are the culmination of several critical stages of processing and analysis where significant methodological errors can occur (Waites et al., 2005).

Unfortunately, recording EEG in the MRI setting presents substantial methodological problems and a number of novel artefacts appear that are different to anything encountered during routine EEG recordings. The prominent gradient artifact (caused by the fluctuating magnetic gradient field during fMRI acquisition sequences) and ballistocardiogram artefacts (caused by small, cardiac induced pulsatile movements) can render the raw EEG uninterpretable, although these can now be substantially reduced by a variety of methods (Allen et al., 1998, Allen et al., 2000, Anami et al., 2003, Benar et al., 2003, Bonmassar et al., 2002, Briselli et al., 2006, Ellingson et al., 2004, Goncalves et al., 2007, Masterton et al., 2007, Mullinger et al., 2008c, Nakamura et al., 2006, Siniatchkin et al., 2007, Wan et al., 2006a, Wan et al., 2006b). Nonetheless, even with these artifact reduction methods the EEG acquired during fMRI is still very sensitive to small movements of either the patient or of the electrode leads. These extrinsic sources of artifact in the EEG recording are very different to those encountered in any other clinical setting and they deserve special attention. Because of these technical issues it can be difficult for the Electroencephalographer to unequivocally identify (‘mark up’) the timing and duration of epileptiform events in the EEG recorded during the fMRI acquisition and this may be an important source of error in EEG/fMRI studies.

We have routinely conducted EEG/fMRI studies at 3T since 2003 (Archer et al., 2003a, Archer et al., 2003b, Archer et al., 2003c, Federico et al., 2005, Labate et al., 2005) and recently introduced head movement sensors into our EEG/fMRI acquisition protocol (Masterton et al., 2007). This has had a profound effect on our ability to interpret the EEG during our 3T EEG/fMRI studies. These sensors provide three channels of non-cephalic signal that detect only head movement. The data from these channels is displayed on the EEG trace and are also used in our artifact removal algorithm (Masterton et al., 2007). As a result of the introduction of these sensors, it became clear that small head and electrode cable movements could sometimes produce ambiguous EEG waveforms that may be misidentified as epileptiform by a reader inexperienced with EEG/fMRI studies (Fig. 1). Furthermore, these movement artefacts come in a variety of forms, with some appearing focal, while others appear to have a more generalized field, thus causing further confusion to an inexperienced reader. To date we have used the head movement sensors in 152 recordings, and we have found potentially ambiguous waveforms caused by movement in every recording, including recordings from normal control subjects with no history of neurological or EEG abnormalities.

This observation led us to consider what the effect would be if such events were incorrectly identified as epileptiform, and more generally, what the consequences are when the EEG is incorrectly marked up in EEG/fMRI studies. There are, of course, a large number of ways in which an EEG may be incorrectly marked up, but in the current study we examine the sorts of error that we consider most likely to occur.

In this study we consider four types of EEG mark-up error.

• What happens if the Electroencephalographer misses some true epileptiform events?

• What happens if the Electroencephalographer only marks up non-epileptiform, movement-related artefacts?

• What happens if some non-epileptiform movement events are included in the analysis of clear epileptiform events?

• What happens if the events are inconsistently marked in time? That is, what if the Electroencephalographer sometimes marks an epileptiform spike at the peak of the waveform, and other times marks similar events at some earlier or later part of the waveform?

Finally, we use the evidence obtained from our study to establish a set of practical guidelines that will help ensure the quality of EEG/fMRI studies.