Skip to main content
NCBI home page
Cognitive Neurodynamics logoLink to Cognitive Neurodynamics
. 2020 Jul 10;15(2):207–222. doi: 10.1007/s11571-020-09614-5

Localizing confined epileptic foci in patients with an unclear focus or presumed multifocality using a component-based EEG-fMRI method

Elias Ebrahimzadeh 1,✉, Mohammad Shams 2, Ali Rahimpour Jounghani 3, Farahnaz Fayaz 4, Mahya Mirbagheri 1, Naser Hakimi 5, Lila Rajabion 6, Hamid Soltanian-Zadeh 1,7
PMCID: PMC7969677  PMID: 33854640

Abstract

Precise localization of epileptic foci is an unavoidable prerequisite in epilepsy surgery. Simultaneous EEG-fMRI recording has recently created new horizons to locate foci in patients with epilepsy and, in comparison with single-modality methods, has yielded more promising results although it is still subject to limitations such as lack of access to information between interictal events. This study assesses its potential added value in the presurgical evaluation of patients with complex source localization. Adult candidates considered ineligible for surgery on account of an unclear focus and/or presumed multifocality on the basis of EEG underwent EEG-fMRI. Adopting a component-based approach, this study attempts to identify the neural behavior of the epileptic generators and detect the components-of-interest which will later be used as input in the GLM model, substituting the classical linear regressor. Twenty-eight sets interictal epileptiform discharges (IED) from nine patients were analyzed. In eight patients, at least one BOLD response was significant, positive and topographically related to the IEDs. These patients were rejected for surgery because of an unclear focus in four, presumed multifocality in three, and a combination of the two conditions in two. Component-based EEG-fMRI improved localization in five out of six patients with unclear foci. In patients with presumed multifocality, component-based EEG-fMRI advocated one of the foci in five patients and confirmed multifocality in one of the patients. In seven patients, component-based EEG-fMRI opened new prospects for surgery and in two of these patients, intracranial EEG supported the EEG-fMRI results. In these complex cases, component-based EEG-fMRI either improved source localization or corroborated a negative decision regarding surgical candidacy. As supported by the statistical findings, the developed EEG-fMRI method leads to a more realistic estimation of localization compared to the conventional EEG-fMRI approach, making it a tool of high value in pre-surgical evaluation of patients with refractory epilepsy. To ensure proper implementation, we have included guidelines for the application of component-based EEG-fMRI in clinical practice.

Keywords: Simultaneous EEG-fMRI, Epilepsy, Independent component analysis (ICA), Blood-oxygen-level dependent imaging (BOLD), Generalized linear model (GLM)

Introduction

An essential and vital step in the pre-surgery evaluation of patients with pharmaco-resistant focal epilepsy is the localization and delineation of epileptogenic areas in the brain. Although costly, challenging, and complex clinical tests and evaluations are performed, 15% of patients eligible for the surgery are not properly diagnosed and thus not considered for the surgery (Berg et al. 2003a, b).

Ictal electroencephalogram (EEG), single-photon emission computed tomography (SPECT), interictal EEG, magnetoencephalogram (MEG), positron emission tomography (PET), functional near-infrared spectroscopy (fNIRS) are among non-invasive localization methods each of which has its limitations. For example, EEG and MEG fail to identify the sources that are located in deep brain structures. On the other hand, model assumptions incorporated in EEG and MEG source localization often remain unverified. PET and SPECT techniques, while potent in detecting regional abnormalities, are not particularly useful in local observations. Moreover, fNIRS has limited spatial resolution in order to localize the sources (Jahani et al. 2015; Rahimpour et al. 2017, 2018; Mirbagheri et al. 2020a, b, 2019). Such restrictions cause a considerable number of patients, especially those with presumed extratemporal sources, to resort to invasive examinations performed by intracranial electrodes, which might inevitably expose them to fairly serious risks (Burneo et al. 2006; Spencer et al. 1998). Furthermore, due to the limited number of electrodes, which can only cover a small part of the brain, there is a challenging yet critical prerequisite to determine in advance which areas of the brain should be examined (Zijlmans et al. 2007; Schöller et al. 2018; Cukic et al. 2020). Therefore, over the past few years, a lot of attention has been paid to enhancing and strengthening non-invasive localization procedures in an ongoing effort to either minimize or optimize the use of invasive methods.

Among the various methods proposed for improving non-invasive source localization, EEG-fMRI analysis stands out distinctly as the most promising. Measuring the changes in the brain oxygenation rate concerning epileptic events, EEG-fMRI analysis provides comprehensive information from the two modalities in a manner that other techniques hardly do (Balasubramaniam et al. 2004; Bénar et al. 2002; Kobayashi et al. 2006a, b; Krakow et al. 1999). It is also well established that the blood-oxygen-level-dependent (BOLD) responses correspond with the source of spike generation, irrespective of structural MRI abnormalities (Al-Asmi et al. 2003; Bagshaw et al. 2004; Hamandi et al. 2004; Lemieux et al. 2001; Rahimpour et al. 2017; Zijlmans et al. 2007). Drawing a comparison between EEG-fMRI with EEG and intracranial stereo-EEG, the authors of Jatoi et al. (2014) and Gotman et al. (2006) demonstrated that intracranial EEG information complemented by EEG-fMRI information could form a powerful tool that contributes significantly to the preoperative evaluation of surgical candidates (Kobayashi et al. 2006a, b; Salek-Haddadi et al. 2006).

Despite the potential advantages, EEG-fMRI analyses generally suffer from inevitable limitations elaborated below.

  • They are premised on the identification of IEDs on the inside the scanner EEG. Therefore, if for any reason, the IEDs are not detectable or if the patient shows few or unclear IEDs, these methods fail to produce reliable results.

  • They are solely focused on the temporal information of the events, meaning that on one hand, all information that lies in the neural behavior of the epileptic generators in between the events is simply lost, and on the other hand, all the extracted events will be modeled regardless of their possible location (their spatial information) or their different amplitudes.

  • The fact that conventional EEG-fMRI analyses cannot provide information from between the events points to the blind approach of these methods since such information is undoubtedly as important as that of the specific time of the event, if not more.

  • In cases where the generator is located in deep brain structures, the scalp EEG commonly fails to represent the corresponding events. Consequently, the conventional methods using the temporal information of EEG will be of little use.

  • The linear regressor cannot adaptively model the different amplitudes and duration of IEDs which leaves a negative impact on the ultimate localization.

These shortcomings call for a new diagnostic tool, particularly in cases where other source localization techniques have failed to localize a single, circumscribed source—a prerequisite for surgery. Most recently, the component-based EEG-fMRI analysis has started to gain a lot of research interest in the literature (Ebrahimzadeh et al. 2019a, b, c, d, e, 2018a, b; Hejazi and Motie Nasrabadi 2019). We aim to apply this approach to the patients who deemed ineligible for surgical operation because neither EEG nor other non-invasive techniques delineated a single epileptic source. These patients had not undergone invasive EEG. We studied whether and to what extent the component-based EEG-fMRI analysis improved the result of epileptic source localization, and also how it could affect the presurgical decision-making in complex cases.

The current study attempts to identify the relevant brain sources as possible indicators of epileptic neural activity in the cortex domain through estimating the epilepsy generators in the sensor domain (scalp EEG). We then include physiological information and a few terms and conditions to come up with a more realistic estimation compared to the classical regressor. We initiate the process by performing the ICA algorithm on EEG signals to have the sources separated and choose those that are related to epileptic activity. Then, we compute the cross correlation between the time series of the extracted components of ICA and the spike template, and prioritize the components as the candidates of the epileptic sources accordingly. To distinguish the epilepsy generators, we pick out the components that meet the following conditions:

  • (i)

    The component is among those that are most likely related to epileptic activity.

  • (ii)

    The component has a significant cross correlation with the spike template at the times of the event.

  • (iii)

    The source of interest is mapped in accordance with the spike field in the sensor domain.

The block diagram of our approach for localization of epileptic focus is shown in Fig. 1.

Fig. 1.

Fig. 1

Open in a new tab

Block diagram of the proposed approach for localizing the epileptic focus

In the Materials and Methods Section, we draw the focus on the most relevant components that are determined through the application of ICA and cross-correlation with the spike template. The time series of the identified components will be convolved with the canonical hemodynamic response function (HRF), resampled to the frequency of the fMRI recording and used as component-of-interest in the GLM analysis. In the Results Section, we look into the outcome of the proposed localization method and compare it with that of the conventional, spike-based approach.

Materials and methods

Patients

From January 2010 to February 2018, 68 people over the age of 16 suffering from focal epilepsy underwent surgery in the Pars Hospital, Tehran, Iran. 43 patients who had been rejected for surgery were asked to complete a questionnaire on whether they would be willing to go through further diagnostic examinations and if they would give permission to have their medical files re-studied. 32 patients (74%) responded, fifteen of whom were ultimately chosen to undergo sessions of EEG-fMRI at the National Brain Mapping Laboratory (NBML) in Tehran, Iran. The inclusion criteria were as follows: (a) there are no contraindications for MRI; (b) the patients had shown at least 10 IEDS in 20 min of previously recorded EEG; (c) the patient was excluded from surgery due to the inability to have a single focus localized.

All patients gave written informed consent and the EEG–fMRI procedure was approved by the local ethics committee of the Iran University of Medical Sciences, Tehran, Iran.

EEG-fMRI acquisition

EEG was recorded inside a 3T MRI scanner (Siemens Prisma), and patients underwent simultaneous EEG-fMRI recordings for 20 min at rest with eyes closed. A 64-channel magnetic resonance-compatible EEG cap was used according to the 10–20 system (reference Cz); ECG was recorded using a single lead (Pedreira et al. 2014; Vulliemoz et al. 2011). Moreover, EEG was recorded for 10 min with eyes closed outside the scanner immediately prior to fMRI scanning (Vulliemoz et al. 2011). Electrodes were equipped with an additional 5 k resistance, and impedances were kept as low as possible. Data were transmitted to an acquisition computer outside the MRI suite via an MR-compatible headbox through an optic fiber cable. EEG was acquired at 5 kHz using BrainAmp magnetic resonance-compatible amplifiers (Brain Products), and EEG was synchronized with the MRI clock.

A T1-MPRAGE anatomic acquisition was done (1 mm slices, 256 × 256 matrix, echo time [TE] = 3.74 ms, repetition time [TR] = 1810 ms, flip angle = 30°) and used to superimpose functional images. Functional data was obtained in 20 min runs with patients at rest, using a T2*-weighted gradient-echo (GRE) imaging sequence (234 × 234 matrix, 40 slices, 3 × 3 × 3 mm, TE = 26 ms, TR = 2500 s, flip angle = 60°). The patient’s head was immobilized with a pillow filled with foam microspheres to minimize movement and provide comfort.

EEG signal processing

The preprocessing of the EEG signals was accomplished using the EEGLAB toolbox (available at https://sccn.ucsd.edu/eeglab/). Generally, a low-pass filter should be applied before down-sampling the signal to avoid aliasing; nevertheless, the EEGLAB function employs the anti-aliasing filter automatically; With that in mind, we reduced the sampling rate from 5000 Hz to 250 Hz. We also suppressed the baseline drift, which contains low-frequency components, through a Butterworth high-pass filter with cut-off frequency at 1 Hz (Ebrahimzadeh et al. 2019a, b; Goshvarpour and Goshvarpour 2019; Daneshi et al. 2020). The channels are then reviewed to identify and remove the abnormal channels, i.e., those with a p value over 0.01 if there is any. In order to exclude noisy channels, the standard deviation was calculated for (over the entire recording), a mean Std. was calculated, and individual channels where Std. was +/− 3.1 std. from the mean were removed.

To eliminate the power line interference, containing the high-frequency components, we implement the Clean Line algorithm, which as is illustrated in Fig. 2, outweighs the classic notch filter in terms of retaining the main content of the signal (Ebrahimzadeh et al. 2019a). It adaptively estimates and removes sinusoidal artifacts, while unlike the notch filter, does not create band-holes in the EEG power spectrum.

Fig. 2.

Fig. 2

Open in a new tab

Activity power spectrum of channel 20 in the occipital region after applying the a high-pass filter; b notch filter; and c clean line filter

Lastly, the ICA method is applied on the EEG signal and the components of no relevance are removed. We then went back from the component domain to the sensor domain to have the spikes detected by a trained expert. Figure 3 demonstrates a samples of a removed component identified as artifact. As shown, IC1 represents the electrocardiogram (ECG) signal.

Fig. 3.

Fig. 3

Open in a new tab

The extracted components from the EEG signal. a Time series of extracted components. b IC1 on scalp EEG located in the left occipital lobe. c Power spectrum of the IC1 time series

Gradient and ballistocardiogram artifact removal

EEG-fMRI recordings are often highly contaminated by the gradient artifacts, associated with the time-varying magnetic field gradients, making the signals significantly hard to interpret. To resolve this issue, we use the fMRIb algorithm (available at https://fsl.fmrib.ox.ac.uk/eeglab/fmribplugin/) which first increases the sampling rate to 20 kHz and then applies a low-pass filter at 60 Hz (Ebrahimzadeh et al. 2019a, b, f, 2014, 2018c). The Ballistocardiogram (BCG) artifact, on the other hand, is caused by cardiac pulsations, which can move the electrodes and consequently compromise the quality of the EEG signal. Inside the magnet, the magnitude of the BCG artifact signal can be more than 3–4 times that of the EEG, spreading throughout the heart beat period; Despite its repetitiveness, the BCG artifact is far less stable than the GRA artifact and the magnitude, the time scale and the shape can vary substantially from beat to beat and from channel to channel.

For this reason, using the fMRIb method, in combination with, e.g., a QRS detector operating on the ECG channel to detect the heartbeats, often results in excessive residual artifacts in the EEG data and more sophisticated approaches and procedures are required to better evaluate the temporal and spatial layout of the BCG artifact. To remove the ballistocardiogram artifact, we used the Optimal Basis Set (OBS) method, as implemented in the FMRIB plug-in for the MATLAB toolbox EEGLAB (Ebrahimzadeh et al. 2019a, b, c, d, e).

Figure 4 illustrates the EEG signals inside the scanner before and after the removal of the artifacts.

Fig. 4.

Fig. 4

Open in a new tab

EEG signal recorded inside the MR scanner: a before and b after the elimination of gradient and BCG artifacts

Component-based EEG-fMRI analysis

The conventional spike-based methods create a linear regressor based on the temporal information obtained from EEG to show the times of the events and localize the epileptic focus on MR images accordingly. This means that all approaches of this kind solely take into account the exact time of the spikes. In other words, they convolve a binary regressor (made of 0 or 1) with the HRF function (Bast et al. 2004; Karoly et al. 2016a, b LeVan and Gotman 2009). On the contrary, the proposed method is postulated based on the fact that the neural behavior of the epileptic foci over a period of time, and not only at specified times, contains information of vital value which should not be left out of the localization process. The following elaborates at length how and based on what conditions a source is identified as a generator of epileptic activity.

The current study applies a template matching algorithm made based on the method presented by Karoly et al. (2016a, b). The spike template is set to a fixed length of 0.3 s and the template window is advanced through the EEG by 1 sample (2.5 ms). The initial spike template is set for each subject by hand-selecting and averaging 10–20 spikes. New spikes are then detected and added to the initial template (Karoly et al. 2016a, b). The match between the template, x, and each 300 ms of the scalp EEG, y, is defined as the magnitude of the sample correlation, |rxy|, where

rxy=∑i=1nxi-x¯yi-y¯∑i=1nxi-x¯2∑i=1nyi-y¯2

To add to the initial, hand-selected template, the first pass over the data is performed at a high threshold (rxy = 0.96 to 0.98 depending on the subject). The first pass is terminated when a final template containing 100 spikes is obtained. This process of creating a final template is repeated at semi-regular intervals of days to weeks (depending on the size of the stored data epochs) throughout the trial, to capture long-term variation in the spike shape (Karoly et al. 2016a, 2016b). The final template is used in the detection algorithm. A spike is detected when  > 0.85 and the spike amplitude exceeds the background mean by a sufficient amount. The background mean is defined as the channel average over the previous 1 s window. Candidate spikes are required to be within ± (0.5 × final template height) of the background mean. The background threshold helps minimize the false positive rate due to artifacts or other oscillations that contain smaller but similar shapes (Karoly et al. 2016a, b). A graphical representation of the detection algorithm is shown in Fig. 5. The figure shows that tuning the detection threshold gives the same output pattern but alters the balance of false positives and false negatives. There is no way to eliminate these errors; however, as this study is investigating long-term patterns of activity, the detection method is appropriate. Detection is carried out separately for each channel and spikes are time stamped (according to UTC/GMT) to the nearest 0.1 s. Ictal periods are excluded from spike detection (Karoly et al. 2016a, b).

Fig. 5.

Fig. 5

Open in a new tab

Spike Detection. A template is made by averaging similar spike patterns. A 1 s sliding window of 16-channel scalp EEG is used to determine background amplitude. The final 300 ms (shaded region) is matched with the template. The absolute correlation coefficient between EEG and template, |rxy| is thresholded for spike detection. Spikes are rejected if amplitude is not within an acceptable range of the background. The right-hand panel shows examples of the detection output (summed over all 16 channels) for 3 threshold values

Automated spike detection is validated by comparing the algorithm output to expert annotation in randomly selected, 1 h data segments. Data segments are selected to evenly cover inter-ictal and pre-ictal periods during night and day. The total spike count, true positive rate (TPR), and false positive rate (FPR) are evaluated. TPR is the number of marked spikes detected by the algorithm divided by the number of the marked spikes (Karoly et al. 2016a, b). False positive rate is the number of detected but unmarked spikes divided by the total detected spikes.

There is no clear definition that can enable different experts to reliably identify interictal spikes on EEG. A recent study to validate a spike detection algorithm found that two expert detections showed only 41% agreement (Karoly et al. 2016a, b). We designed our detection algorithm to err on the side of over-counting spikes, under the assumption that false positives would be evenly distributed, or at least would not strongly bias the distribution of true positives. The benefit of using a template matching algorithm is that false-positives are quantifiably ‘spike-like’ in morphology. Therefore, false detections may still reflect epileptic activity of interest (such as bursting).

Identification of epilepsy-related components

Once the ICA algorithm is applied on EEG signals, a large number of components emerge, each of which is associated with a different source ranging from physiological artifacts to cerebral source activity. Thus, the primary challenge is to correctly pick the components of interest, namely the ictal-related BOLD changes.

To identify and isolate the ICs generated by the epileptic activity, we implement the following procedure. First, IED-related spikes detected in the outside of scanner EEG are averaged to build a patient-specific spike template shown in Fig. 6. Being a time amplitude map, the template is then band-pass filtered between 1 and 30 Hz (Bast et al. 2004) and is ultimately outlined by dramatic spike deflection on the EEG channels, beginning from the onset at baseline all the way to the negative peak of the following slow wave.

Fig. 6.

Fig. 6

Open in a new tab

Schematic representation of the proposed method for identification of components

If a source is in fact an epilepsy generator, it is highly likely to appear in several ICA decompositions. That said, we repeat the ICA algorithm 10 times with random initializations in order to determine the number of extracted sources (LeVan and Gotman 2009). Then, we use a sliding time window with the same length as the template to investigate the cross-correlation between the spike template and the highly-ranked extracted components in each iteration. The components that show strong cross-correlation at the times of the events are chosen if and only if their respective sources in the cortex domain are concordant with the region of the observed spike in the sensor domain, that is, the electrode domain. We conclude that epilepsy generators are the components identified by an F-test (p < 0.05, corrected for the number of components), with the motion parameters used as confounds as in the GLM model. Finally, the time series of such components are convolved with HRF, resampled to the frequency of the fMRI recording and used as the component of interest in the GLM analysis.

In the event that there is a plurality in the number of IEDs a patient experienced in the scanner, there will be a separate component attributed for each, on the account that different epileptic foci can generate different types of IEDs (Worsley et al. 2002).

The identified components not only contain the information at the times of the spike occurrence, but also carry useful information that a traditional regressor simply cannot provide. This is underpinned by the fact that the epileptic generators are not active exclusively at the times of the spikes; they are actually active at all times and can provide vital information even when no spike is emitted.

It should be stressed that components selected through this method embody information from both the specific time of the events and the time when no spike is generated, which can well provide a wider perspective on the neural behavior of epilepsy generators, as they are in fact, active at all times.

Following this procedure, we are ensured to have drawn the focus on the components that signify the neural behavior responsible for the initiation of epileptic discharges, which can greatly contribute to a more realistic estimation of localization compared to the corresponding digital regressor.

Spike-based (conventional) analysis

An experienced neurophysiologist manually searched for each type of intra-MRI spikes in each patient. Spikes were modeled as zero-duration events, convolved with a standard HRF, and used as a regressor for the GLM model and fMRI analysis.

Functional MRI

We used the FSL software (FMRI Expert Analysis Tool, Version 5.0.9, FMRIB’s Software Library, http://www.fmrib.ox.ac.uk/fsl) to perform motion correction (realignment with 6-parameter rigid-body transformation) and smoothing (6-mm full width at half maximum). Temporal autocorrelations were corrected with an autoregressive model of order one (Worsley et al. 2002) and low-frequency drifts were modeled with a third-order polynomial. The traditional spike-based model uses the time and duration of each event to build an IED-specific regressor and convolves it with spike related HRFs, whereas, the current study proposes to convolve the independent component time series with 4 HRFs peaking at 3, 5, 7, and 9 s (Hao et al. 2018). All components were included in the same general linear model (GLM). A statistical t-map was obtained for each component using the other components as confounds.

EEG-fMRI analysis

To be prominent, a response is required to have 5 contiguous voxels with a t-value of 3.1 corresponding to p < 0.05, corrected for multiple comparisons based on the number of voxels and the 4 HRFs. We illustrated the t-map results using a red-yellow scale for positive BOLD changes, i.e., activation (Figs. 1, 7, 8). We ignored responses outside the brain parenchyma. Two experts reviewed the IED-related BOLD responses. For each patient, the analysis proceeded as follows.

Fig. 7.

Fig. 7

Open in a new tab

Component-related BOLD response showed a focal activation in the left mid-dorsolateral prefrontal cortex (patient 1). a Raw EEG data acquired inside the MRI scanner. b Cleaned EEG after removing the gradient artifact. c Identified component time series. d The component identified on scalp EEG located in the left lateral frontal lobe. The active area is marked by yellow-red color. e Dipole localization (red) of the identified generator in deep brain structures. f Localization of the generator using simultaneous analysis of component-based EEG-fMRI. g Localization of the generator using simultaneous analysis of conventional EEG-fMRI

Fig. 8.

Fig. 8

Open in a new tab

Comparison of the conventional and proposed methods in localizing epileptic foci of an unclear and multifocal case using simultaneous EEG-fMRI recordings (Patient 3, IED 1)

Analysis of IED sets

Statistical parametric mapping was created using the F-test (Friston et al. 1998) with correction for multiple comparisons using Family Wiser Error as a method based on Gaussian field theory (GFT), where the results with a p-value of 0.05 or smaller were significant. We found that IED sets contained a significant BOLD response, therefore, the topographic and spatial adaptations were evaluated between the significant BOLD response and the EEG activity in: (1) the same area, ipsilateral; (2) the same area, contralateral; (3) ipsilateral, a different area; and (4) no concordance. Moreover, the number of IED-sets that could find a significant BOLD response (identified by each of the two observers) was evaluated.

Clinical evaluation of EEG-fMRI results

For each patient, the primary reason for rejection was categorized into: (1) inability to delineate a clear focus (including presumed frontal foci with unknown lateralization); and (2) multiple potential distinct foci or both. Only when the consensus-IEDs showed a significant (p < 0.05 based on GFT) BOLD response with a positive sign (Kobayashi et al. 2006a, b), the result was considered robust. Of the patients with at least one robust result, we re-evaluated the EEG-fMRI results of only those with a BOLD response in the region that could be expected on the basis of the IED (defined as topographically related) using the following two ways.

Firstly, the potential added value for the two categorized localization problems was assessed apart from clinical interpretation. To access the extent of the BOLD-response in the localization of a single focus and to access all possible (co-)activated areas in case of presumed multifocality, the uncorrected maps of voxels with p < 0.001 were considered together with the statistically corrected results. The outcomes included: (1) in terms of source localization (category 1), EEG-fMRI showed a circumscribed focus in the expected region, whereas EEG did not (Table 2: ‘modified’) or the focus localization remained unclear (‘not modified’); (2) in terms of multifocality (category 2), EEG-fMRI indicated a single focus (‘unifocal’), multifocality was confirmed consistent with the EEG (‘multifocal’), or a significant BOLD response was unifocal but the uncorrected BOLD responses (p < 0.001) were multifocal (‘likely multifocal’).

Table 2.

Patients re-evaluated for surgery by the component-based EEG-fMRI and the conventional EEG-fMRI analysis

No. Age/ Sex MRI Clinical problem Type of problem Proposed EEG-fMRI Analysis Conventional EEG-fMRI analysis
EEG-fMRI Outcome EEG-fMRI Outcome
1 28/♀ Tumour- right Temprofrontal Temporal right/left- multifocality Temporal left (n + c) Unifocal Bitemporal (n + c) likely multifocal
Frontal focus unclear Source localization Widespread left frontal (n + c) Not modified Widespread left-right frontal (n + c) Not modified
2 23/♀ – Frontal focus unclear Source localization Localized confined frontal focus (n + c) Modified Localized confined frontal focus (n + c) Modified
3 28/♂ Lesion-right parietotemporal Parietal left-right multifocality Right parietotemporal (n + c) Unifocal Right parietotemporal (n + c), Temporal left (n) likely multifocal
Parietal focus unclear Source localization Localized clear right parietal focus (n + c) Modified Localized clear right parietal focus (n + c) Modified
4 19/♀ – focus unclear Source localization deep confined focus (n + c) Modified Medial temporal (n) Not modified
5 34/♂ –

Temporal right-left,

Frontocentral left

 Multifocality confined temprofrontal left focus (n + c) Unifocal confined temprofrontal left focus (n + c) Unifocal
6 26/♀ – Frontotemporal focus unclear Source localization Localized clear focus (n + c) Modified Localized clear focus (n + c) Modified
7 24/♂ MTS Bilateral temporal Multifocality right temporal (n + c) Unifocal Right temporal (n + c) Unifocal
8 25/♀ – Biferantal Multifocality Biferantal (n + c) Multifocal Biferantal (n + c) Multifocal
9 21/♀ WML Temporal focus unclear Source localization Localized clear focus (n + c) Modified Wide-spread Temporal focus (n + c) Not modified
Open in a new tab

Abbreviation: WML White Matter Lesions, MTS Mesiotemporal Sclerosis, hemisph hemisphere. n: fMRI not corrected for statistical significance (p < 0.001 for each voxel), c: fMRI corrected according to GFT (p < 0.05)

Secondly, the proposed method was validated and compared with the conventional method as follows.

  1. For epileptic focus localized using intracranial EEG study or surgical resection, we used the same areas as the ground truth to evaluate the methods.

  2. For patients not surgically treated without an intracranial EEG study or an MRI visible tumor, we used the abnormal areas defined based on the consensus of all medical evidences to evaluate the methods.

  3. To evaluate the regions identified outside the ground truth area, we calculated the distance between the maximum BOLD response and the center of the ground truth area.

  4. When the conventional and proposed methods found the same area, we called them concordant.

We have performed an independent validation of component-based EEG-fMRI results for those patients who had underwent intracerebral EEG using electrodes which record the electrical activity of the regions that correspond with the BOLD response, or for those who were reported to have a focal lesion on MRI. It should be noted that a lesion is recognized as focal if it exists within one lobe. For example, we could not provide independent validation for polymicrogyria, which typically takes up more than one lobe. The BOLD data will be validated only if it in line with the independent validation. This would mean that SEEG demonstrates the spike in the same sublobe or MRI shows a focal lesion considered as an epileptic focus.

Results

Of 32 patients, 15 fulfilled the selection criteria for EEG-fMRI (mean age of 36.6 years, 13 women). Of these, twelve patients having IEDs were included for an EEG-fMRI study, three of whom were excluded, two patients due to significant movement artifacts during the recordings precluding further analysis, and one due to the lack of clear IEDs during the EEG-fMRI recordings. Therefore, nine patients were ultimately included whose descriptions are presented in Table 1. Of these, 5 patients (55%) had no structural lesions on the 3T MRI that included the coronal FLAIR images. The vacuum-drawn pillow could not be used for two patients because of their head size, which led to increased movement artefacts and in one patient discontinuation after 420 dynamic scans because of discomfort. In three patients, the measurement was done in two sessions each for practical reasons. The best measurements were used for analysis in these three patients.

Table 1.

Summary of IED studies which indicated a significant component-related BOLD response to consensus IEDs

Pt.-type Ictal EEG Interictal EEG IED Activation Deactivation
1-1 Temporal left Temporal right/left 14 Temporal right/left (++) –
1-2 Frontocentral bilateral Frontal left 9 Frontocentral bilateral (*)
2-1 Unclear Frontal right 11 – Frontal right (++)
3-1 Parietal left/right Parietal left 13 – Parietal right (+)
3-2 Temporal left Parietotemporal left 8 Parietotemporal left (++) Frontotemporal left-right (*)
4-1 Bilateral generalized Bilateral generalized 9 Thalamus (++) –
5-1 Unclear Temporal right-left 15 Temporal right (*) Temprofrontal (+)
5-2 Frontal right/left Frontal right 6 – –
5-3 Frontal left Frontocentral left 7 Frontal left (++) –
6-1 Left hemisphere Frontotemporal left 17 Frontotemporal left (++) –
7-1 Occipitotemporal right Occipitotemporal right 12 – Occipital right (++)
7-2 Bitemporal Bilateral temporal 7 Bilateral temporal (++) Bilateral temporal (++)
7-3 Left Parietal/Post Temporal Left Parietal/Post Temporal 11 Paritotemporal bilateral (*) Frontal right (-)
8-1 Frontopolar right Frontocentral right 14 Frontal right-left (+) Central right (-)
9-1 Unclear Temporal right 12 Temporoparietal right (*) Temporal right (++)
9-2 Parieto-occipital left Parietal left 9 Parietal left (++) Occipital right (-)
Open in a new tab

In the superscript, the topographical concordance between clinical localization and BOLD response is given as: (++): same area, ipsilateral; (+): same area, contralateral; (*): ipsilateral but a different area; (−): no concordance

EEG analysis results

The nine patients showed at least 1 IED type: 4 patients (1 type); 3 patients (2 types); and 2 patients (3 types). Thus, for each patient with each type, one source was identified, making a total of 16 EEG-fMRI recordings to be analyzed. Altogether, 428 IEDs were recorded during the fMRI acquisition. Seven studies were evaluated by a third observer because of low interobserver agreement. One EEG-fMRI recording demonstrated no BOLD response, but this occurred in 1 patient with multiple types of IEDs. Thus, at least one BOLD response was observed for each patient, leaving 15 recordings in all: 5 IED studies presented only activations, 3 IED studies presented only deactivations, and 7 IED studies showed both activations and deactivations.

Functional MRI results

A significant BOLD response was found in 16 consensus-IED sets from 9 patients (Table 1). Three patients (5 IED sets) had a robust outcome, i.e., a positive significant BOLD response to consensus IEDs. In two cases, at least one robust fMRI result was highly topographically related to the IED and these patients were included for further clinical consideration. Positive BOLD responses that were topographically incongruent with the EEG were found in the frontal region or in the mesial structures, together with a diffuse negative BOLD response. Negative BOLD responses were seen in the parietal region diffused with a positive response in the hypothalamus, in the right occipital and in the left temporal lobe.

Figure 7 shows the localization process used in this project step by step.

Clinical results

Component-related EEG-fMRI results were first interpreted in the light of the source localization problem (Table 2). In five out of six foci not clearly localized based on EEG, component-based EEG-fMRI indicated a circumscribed focus while the conventional analysis was only able to identify four of them. In one of the cases where the focus was not clearly localized based on EEG, the conventional spike-based EEG-fMRI method was also unable to create a bright spot. In one out of the five patients with presumed multifocality, component-based EEG-fMRI was also indicative of multifocality, but in the remaining four patients, clearly favored a single source. In the conventional analysis, these results were limited to only two unifocal and there were the confirmation of three multifocal out of the five cases. The details are as follows.

In patient 1, with presumed multifocality and a widespread EEG focus, one of the foci (located in the left temporal lobe) predominated in component-based EEG-fMRI and could also be localized to a more confined source than those of the EEG and the conventional EEG-fMRI analysis. Due to its dependence on the EEG events, the conventional EEG-fMRI analysis identified a bitemporal area and there was still the problem of non-recognition of a circumscribed focus. This patient underwent surgery during which acute electrocorticography showed fairly circumscribed IEDs as well as a prolonged, spontaneous electrographical discharge, confirming a seizure onset zone located in the right temporofrontal region.

In patient 3, a parietotemporal source was hypothesized but could not be lateralized. Component-related BOLD changes indicated single focus in the right parietal lobe, to right-sided as well as to left-sided parietal interictal discharges. He was offered further evaluation by means of implantation of a subdural electrode-grid, but this has been delayed because of a recent improvement in seizure control.

For Patient 4, given the presumed multifocality and interictals observed in most EEG channels, the conventional method was still unable to identify a single focus due to a deep cornea. Considering the number of independent component candidates and their individual examination, the proposed method finally succeeded in identifying a clear and confined focus in the brain depth.

Patient 5 indicated a middle temporal bilateral EEG-focus, but prevalent on the right side that could be associated with either a confined deep-lying source or a more widespread superficial focus. Component-related BOLD changes revealed a circumscribed BOLD response involving the left temprofrontal lobe. Because the focus was temporal and close to the auditory area, surgery was considered too risky.

In patient 7, interictal and ictal EEG activity was found in multiple areas, with bitemporal paroxysms with an unstable maximum over the right side. Component-based EEG-fMRI showed a predominant right temporal focus. This patient was considered not eligible for surgery, because there was overlap between the ictal onset zone and Wernicke’s area.

In patients 8, the results of the proposed and conventional EEG-fMRI analysis methods confirmed the existence of multiple epileptic sources, which corroborated the earlier decision to refrain from further investigations.

In the patient 9, the proposed EEG-fMRI method localized a clear and confined focus in the left temporal area while the conventional method identified a widespread focus in the temporal lobe.

Figure 8 illustrates the comparison of the conventional and proposed methods in localizing epileptic foci of an unclear and multifocal case using simultaneous EEG-fMRI recordings.

In patient 1, regarding the problem of lack of source localization, component-based EEG-fMRI results collateralized with previous investigations, but did not result in the identification of a more confined focus.

Overall, in seven out of the nine patients (78%), component-based EEG-fMRI results opened new prospects of surgery, which has been performed in one patient so far.

Discussion

The idea of combining the two modalities in epilepsy is self-explanatory and has been widely addressed in the literature, e.g., (Gotman and Pittau 2011; Kobayashi et al. 2006a, b). Much of the motivation to combine EEG and fMRI measurements originate from two complementary views separately provided by the two modalities, which together lead to a deeper grasp of the reality. However, there seem to be limits to how far this technique can be extrapolated. Studies that adopt the EEG-fMRI approach tend to depend to no small extent on the time of the spike-related BOLD response on the EEG signal. In other words, the BOLD response can be analyzed once the time of the event on EEG has been determined. While this may typically yield an acceptable representation of the event, there are several cases where the EEG fails to provide the required information, for instance when the generator is located in the deep brain structure, the signal-to-noise ratio is relatively low or when the patient does not produce inside-the-scanner IED detectable on EEG.

Consistent with the literature, the current study aims to offer a remedy for the shortcomings of the conventional EEG-fMRI analysis by putting forth a component-based approach that does not require clearly-visible inside-the-scanner IEDs and lends itself well to circumstances in which the classical analyses may be ineffective.

The component-based method brings to attention the variations in amplitude and duration of epileptic spikes, whereas the conventional methods simplistically assume that all events are equal. Moreover, the conventional approach overlooks the fact that IED activity is continuous and contains fluctuating sub-threshold epileptic activity that is not vividly observed on surface EEG recordings. Such valuable information is obtained by the ICA algorithm applied as part of the proposed method.

The component-based analysis is considered particularly potent in detecting the generators, which neither the single-modality based methods nor the conventional dual-modality methods have been able to delineate. This makes the current study the first to explore and highlight the value of component-related BOLD response in clinical practice as it improves the source localization in complex cases, such as those with unclear or multifocal seizure onsets, where the scales can be tipped in favor of surgery, and can, therefore, be applied as a vital part of the preoperative evaluation procedure.

It is also noteworthy that the component-based method may play a more prominent role in eliminating the need for invasive electrode implantations compared to the conventional EEG-fMRI analysis. However, when studying a cohort of patients rejected because source localization using a number of established methods failed, the a priori chance that any novel technique, especially one based on the interictal EEG, has a high impact, is relatively low.

As mentioned above, a major source of limitation in conventional methods is that they are hinged on the inside-the-scanner IEDs. The typical inclusion criteria in such methods is that there is a minimum of 10 IEDs in the span of 20 min, yet in reality, patients tend to show fewer detectable IEDs than expected. This clearly lowers the functionality of the classical approaches, given the fact that they are fundamentally predicated on the inside-the-scanner IEDs. To break this dependence, the component-based method identifies the time series of the component at all times and then feeds them into the GLM model; therefore the behavior of the corresponding voxel in the fMRI analysis will be evaluated against the overall activity of the focus. On the other hand, computing the cross-correlation between the spike-template from outside the scanner and the extracted time series of the inside-the-scanner EEG successfully eliminates the need for identifying clear IEDs on inevitably contaminated EEG, as well as the visual inspection by an expert leading to an increased yield of EEG-fMRI, which would have been even higher had we used less strict selection criteria.

In similar past efforts, Jann et al. (2008) tried to apply the ICA algorithm to separate the sources and observe the BOLD changes based on the predictor of the ICA component. They first implemented the ICA algorithm on the inside-the-scanner EEG prior to noise and gradient artifact removal. To identify and eliminate the noise sources, they once again applied the ICA, but this time on the outside-the-scanner EEG, and studied the differences between the two. The problems with this method are that for one thing, not all the identified sources on the outside-the-scanner signal are indicative of authentic sources, and secondly, there is no guarantee that the same noise sources appear both on the inside and outside of scanner EEGs. Equally importantly, the noise sources are not typically recognized as independent sources; instead, they are combined with the informative sources and contaminate them. Finally, the standard components between the signals cannot necessarily be taken to indicate the correct sources; neither can the differences between the two be suggestive of the noise sources.

Other classical methods mostly pivot on the location of the source, which means that the deeper the source, the less effective the combination of the two modalities. Therefore, the inclusion of the cases whose epileptic foci were previously shown to be in deep brain structure can increase the yield of EEG-fMRI and better highlight the distinction between the proposed method and the conventional approaches. Also, as mentioned earlier, one of the limitations of the conventional methods is that they heavily depend on the number and precise representation of the events such that if there are rare events, the signal-to-noise ratio will be low in fMRI, leading to false-negative results.

In contrast, the proposed method shifts the focus onto the component domain and highlights all the information concerning the focus activity, which leads to an increase in signal-to-noise ratio and a decrease in false-negative responses. Accordingly, this method can be applied to eliminate the false sources and set a reliable ground in case of false or presumed multifocality (patient no.9). When the source is located near the surface, the conventional and the proposed methods seem to produce relatively similar results. For example, in patient 2, both methods detected a single clear focus despite an unclear focus in the frontal lobe. However, the proposed method also detected irrelevant sources, which increases the false negatives. It, therefore, appears that if a source is close to the surface, the conventional methods often yield a lower false-negative error and thus offer a more favorable performance.

We defined the clinical value as neutrally as possible, in recognition of the fact that practices differ among epilepsy centers, especially in the availability and use of intracranial EEG (Ridley et al., 2017). We believe that EEG-fMRI will prove particularly useful in the first steps of clinical interpretation, in delineating an unclear EEG source and defining multifocality similar to Zijlmans et al. (2007) finding. The clinical yield of component-based EEG-fMRI is expected to increase with the ongoing improvement of insight and knowledge of registration and analysis techniques. Nevertheless, the final decision regarding surgical candidacy or the need for additional investigations will reflect the philosophy of individual centers. Although the threshold for invasive EEG recordings is relatively high in Iran, the surgery-rejection rate is similar to that reported in other countries (Berg et al. 2003a, b; Zijlmans et al. 2007) and non-invasive source localization techniques like specific MRI sequences and SPECT as well as intracranial EEG monitoring with subdural electrodes are available.

Conclusion

EEG and fMRI are known to complement one another in the spatiotemporal representation of the brain activity as each method has its strengths and limitations. Simultaneous recording and analysis of EEG and fMRI brings about a higher prospect of understanding the reality of neural behaviors. While the EEG-fMRI studies have provided a precise perspective for epileptic focus localization and the use of two simultaneous modalities have had a remarkable benefit for identifying the foci, they suffer from limitations in clinical applications. In this study, we succeeded in diminishing the limitations through applying a method in the component domain for localizing epileptic foci, taking into account the clinical application, so that more satisfactory results than the conventional EEG-fMRI methods could be obtained. In this respect, Zijlmans et al. (2007) indicated that the EEG-fMRI methods have significantly improved the identification of epileptic focal in complex patients and resolved many medical teams’ problems. Our study demonstrated that in patients with multifocality or unclear focus, the proposed method generated clear results for diagnosis, i.e., either the generators were identified using the component-based method or their multifocality were confirmed. The results of our study were confirmed by the medical team of the Pars Hospital in Tehran after they did surgery on the patients who were not initially surgical candidates.

Acknowledgments

The EEG-fMRI data used in this study were acquired in the National Brain Mapping Laboratory (NBML), Tehran, Iran. The authors would like to show their gratitude to Dr. Paolo Federico (Department of Clinical Neurosciences, University of Calgary, Calgary, Alberta, Canada) and his group for sharing their pearls of wisdom during the course of this research. They are also immensely grateful to Prof. Ali Moti Nasrabadi and Dr. Negar Mohammadi for their valuable comments, although any errors are of our own and should not tarnish the reputation of these esteemed individuals. The first author also expresses his gratitude to the Cognitive Science and Technologies Council (COGC), Tehran, Iran for their tremendous support.

Authors’ contributions

EE and HS conceived of the presented idea. EE developed the theory and performed the computations. Material preparation, data collection and analysis were performed by EE. The first draft of the manuscript was written by EE, ARJ, FF and all authors commented on previous versions of the manuscript. MSH, LR and HS verified the analytical methods. The visualization and validation were done by MM and NH. All authors provided critical feedback and helped shape the research, analysis and manuscript. All authors read and approved the final manuscript. HS supervised the project.

Compliance with ethical standards

Conflict of interest

The authors declare that they have no conflict of interest.

Footnotes

Publisher's Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

References

  1. Al-Asmi A, Bénar C-G, Gross DW, Khani YA, Andermann F, Pike B, Gotman J, et al. fMRI activation in continuous and spike-triggered EEG–fMRI studies of epileptic spikes. Epilepsia. 2003;44(10):1328–1339. doi: 10.1046/j.1528-1157.2003.01003.x. [DOI] [PubMed] [Google Scholar]
  2. Bagshaw AP, Aghakhani Y, Bénar C-G, Kobayashi E, Hawco C, Dubeau F, Gotman J, et al. EEG-fMRI of focal epileptic spikes: analysis with multiple haemodynamic functions and comparison with gadolinium-enhanced MR angiograms. Hum Brain Mapp. 2004;22(3):179–192. doi: 10.1002/hbm.20024. [DOI] [PMC free article] [PubMed] [Google Scholar]
  3. Balasubramaniam R, Wing AM, Daffertshofer A. Keeping with the beat: movement trajectories contribute to movement timing. Exp Brain Res. 2004;159(1):129–134. doi: 10.1007/s00221-004-2066-z. [DOI] [PubMed] [Google Scholar]
  4. Bast T, Oezkan O, Rona S, Stippich C, Seitz A, Rupp A, Scherg M, et al. EEG and MEG source analysis of single and averaged interictal spikes reveals intrinsic epileptogenicity in focal cortical dysplasia. Epilepsia. 2004;45(6):621–631. doi: 10.1111/j.0013-9580.2004.56503.x. [DOI] [PubMed] [Google Scholar]
  5. Bénar C-G, Gross DW, Wang Y, Petre V, Pike B, Dubeau F, Gotman J. The BOLD response to interictal epileptiform discharges. Neuroimage. 2002;17(3):1182–1192. doi: 10.1006/nimg.2002.1164. [DOI] [PubMed] [Google Scholar]
  6. Berg AT, Langfitt J, Shinnar S, Vickrey BG, Sperling MR, Walczak T, et al. How long does it take for partial epilepsy to become intractable? Neurology. 2003;60(2):186–190. doi: 10.1212/01.WNL.0000031792.89992.EC. [DOI] [PubMed] [Google Scholar]
  7. Berg AT, Langfitt J, Shinnar S, Vickrey BG, Sperling MR, Walczak T, Spencer SS et al (2003) How long does it take for partial epilepsy to become intractable? [DOI] [PubMed]
  8. Burneo JG, Steven DA, McLachlan RS, Parrent AG. Morbidity associated with the use of intracranial electrodes for epilepsy surgery. Can J Neurol Sci. 2006;33(2):223–227. doi: 10.1017/S0317167100005023. [DOI] [PubMed] [Google Scholar]
  9. Cukić M, Stokić M, Simić S, et al. The successful discrimination of depression from EEG could be attributed to proper feature extraction and not to a particular classification method. Cogn Neurodyn. 2020 doi: 10.1007/s11571-020-09581-x. [DOI] [PMC free article] [PubMed] [Google Scholar]
  10. Daneshi A, Azarnoush H, Towhidkhah F, et al. Brain activity during time to contact estimation: an EEG study. Cogn Neurodyn. 2020;14:155–168. doi: 10.1007/s11571-019-09563-8. [DOI] [PMC free article] [PubMed] [Google Scholar]
  11. Ebrahimzadeh E, Pooyan M, Bijar A. A Novel approach to predict sudden cardiac death (SCD) using nonlinear and time-frequency analyses from HRV signals. Plos One. 2014;9(2):e81896. doi: 10.1371/journal.pone.0081896. [DOI] [PMC free article] [PubMed] [Google Scholar]
  12. Ebrahimzadeh E, Manuchehri MS, Amoozegar S, Araabi BN, Soltanian-Zadeh H. A time local subset feature selection for prediction of sudden cardiac death from ECG signal. Med Biol Eng Compu. 2018;56(7):1253–1270. doi: 10.1007/s11517-017-1764-1. [DOI] [PubMed] [Google Scholar]
  13. Ebrahimzadeh Elias, Soltanian-Zadeh H, Araabi BN. Localization of epileptic focus using simultaneously acquired EEG-FMRI data. Comput Intell Electr Eng (ISEE) 2018;9(2):15–28. doi: 10.22108/ISEE.2018.111024.1123. [DOI] [Google Scholar]
  14. Ebrahimzadeh E, Kalantari M, Joulani M, et al. Prediction of paroxysmal atrial fibrillation: a machine learning based approach using combined feature vector and mixture of expert classification on HRV signal. Comput Methods Progr Biomedi. 2018;165:53–67. doi: 10.1016/j.cmpb.2018.07.014. [DOI] [PubMed] [Google Scholar]
  15. Ebrahimzadeh E, Shams M, Fayaz F, Rajabion L, Mirbagheri M, Araabi BN, Soltanian-Zadeh H. Quantitative determination of concordance in localizing epileptic focus by component-based EEG-fMRI. Comput Methods Progr Biomed. 2019;177:231–241. doi: 10.1016/j.cmpb.2019.06.003. [DOI] [PubMed] [Google Scholar]
  16. Ebrahimzadeh E, Soltanian-Zadeh H, Araabi BN, Fesharaki SSH, Habibabadi JM. Component-related BOLD response to localize epileptic focus using simultaneous EEG-fMRI recordings at 3T. J Neurosci Methods. 2019;322:34–49. doi: 10.1016/j.jneumeth.2019.04.010. [DOI] [PubMed] [Google Scholar]
  17. Ebrahimzadeh E, Nikravan M, Nikravan M, Manuchehri MS, Amoozegar S, Dolatabad MR, Bagheri M, Soroush MZ. Simultaneous EEG-fMRI: a multimodality approach to localize the seizure onset zone in patients with epilepsy. Int J Biol Med. 2019;1:130–139. doi: 10.36811/ijbm.2019.110017. [DOI] [Google Scholar]
  18. Ebrahimzadeh E, Shams M, Rahimpour Jounghani A, Fayaz F, Mirbagheri M, et al. Epilepsy presurgical evaluation of patients with complex source localization by a novel component-based EEG-fMRI approach. Iran J Radiol. 2019;16:e99134. doi: 10.5812/iranjradiol.99134. [DOI] [Google Scholar]
  19. Ebrahimzadeh E, Soltanian-Zadeh H, Araabi BN, Fesharaki SSH, Habibabadi JM (2019e) Localizing epileptic focus through simultaneous EEG-fMRI recording and automated detection of IED from inside-scanner EEG. Iranian J BioMed Eng (IJBME) 13(2):135–145. http://www.ijbme.org/article_35722.html
  20. Ebrahimzadeh E, Foroutan A, Shams M, Baradaran R, Rajabion L, et al. An optimal strategy for prediction of sudden cardiac death through a pioneering feature-selection approach from HRV signal. Comput Methods Progr Biomed. 2019;169:19–36. doi: 10.1016/j.cmpb.2018.12.001. [DOI] [PubMed] [Google Scholar]
  21. Friston KJ, Fletcher P, Josephs O, Holmes A, Rugg MD, Turner R. Event-related fMRI: characterizing differential responses. Neuroimage. 1998;7(1):30–40. doi: 10.1006/nimg.1997.0306. [DOI] [PubMed] [Google Scholar]
  22. Goshvarpour A, Goshvarpour A. EEG spectral powers and source localization in depressing, sad, and fun music videos focusing on gender differences. Cogn Neurodyn. 2019;13:161–173. doi: 10.1007/s11571-018-9516-y. [DOI] [PMC free article] [PubMed] [Google Scholar]
  23. Gotman J, Pittau F. Combining EEG and fMRI in the study of epileptic discharges. Epilepsia. 2011;52:38–42. doi: 10.1111/j.1528-1167.2011.03151.x. [DOI] [PMC free article] [PubMed] [Google Scholar]
  24. Gotman J, Kobayashi E, Bagshaw AP, Bénar C-G, Dubeau F. Combining EEG and fMRI: a multimodal tool for epilepsy research. J Magn Reson Imaging. 2006;23(6):906–920. doi: 10.1002/jmri.20577. [DOI] [PubMed] [Google Scholar]
  25. Hamandi K, Salek-Haddadi A, Fish DR, Lemieux L. EEG/functional MRI in epilepsy: the Queen Square experience. J Clin Neurophysiol. 2004;21(4):241–248. doi: 10.1097/00004691-200407000-00002. [DOI] [PubMed] [Google Scholar]
  26. Hao Y, Khoo HM, von Ellenrieder N, Zazubovits N, Gotman J. DeepIED: an epileptic discharge detector for EEG-fMRI based on deep learning. NeuroImage: Clinical. 2018;17:962–975. doi: 10.1016/j.nicl.2017.12.005. [DOI] [PMC free article] [PubMed] [Google Scholar]
  27. Hejazi M, Motie Nasrabadi A. Prediction of epilepsy seizure from multi-channel electroencephalogram by effective connectivity analysis using Granger causality and directed transfer function methods. Cogn Neurodyn. 2019;13:461–473. doi: 10.1007/s11571-019-09534-z. [DOI] [PMC free article] [PubMed] [Google Scholar]
  28. Jahani S, Berivanlou NH, Rahimpour A, Setarehdan SK (2015) Attention level quantification during a modified stroop color word experiment: an fNIRS based study. In 2015 22nd Iranian conference on biomedical engineering (ICBME), pp 99–103
  29. Jann K, Wiest R, Hauf M, Meyer K, Boesch C, Mathis J, Koenig T, et al. BOLD correlates of continuously fluctuating epileptic activity isolated by independent component analysis. NeuroImage. 2008;42(2):635–648. doi: 10.1016/j.neuroimage.2008.05.001. [DOI] [PubMed] [Google Scholar]
  30. Jatoi MA, Kamel N, Malik AS, Faye I. EEG based brain source localization comparison of sLORETA and eLORETA. Austral Phys Eng Sci Med. 2014;37(4):713–721. doi: 10.1007/s13246-014-0308-3. [DOI] [PubMed] [Google Scholar]
  31. Karoly PJ, Freestone ÃDR, Boston ÃR, Grayden DB, Himes D, Leyde K, Cook MJ et al (2016) Interictal spikes and epileptic seizures: their relationship and underlying rhythmicity. 10.1093/brain/aww019 [DOI] [PubMed]
  32. Karoly PJ, Freestone DR, Boston R, Grayden DB, Himes D, Leyde K, Cook MJ, et al. Interictal spikes and epileptic seizures: their relationship and underlying rhythmicity. Brain. 2016;139(4):1066–1078. doi: 10.1093/brain/aww019. [DOI] [PubMed] [Google Scholar]
  33. Kobayashi E, Bagshaw AP, Bénar CG, Aghakhani Y, Andermann F, Dubeau F, Gotman J. Temporal and extratemporal BOLD responses to temporal lobe interictal spikes. Epilepsia. 2006;47(2):343–354. doi: 10.1111/j.1528-1167.2006.00427.x. [DOI] [PubMed] [Google Scholar]
  34. Kobayashi E, Bagshaw AP, Grova C, Dubeau F, Gotman J. Negative BOLD responses to epileptic spikes. Hum Brain Mapp. 2006;27(6):488–497. doi: 10.1002/hbm.20193. [DOI] [PMC free article] [PubMed] [Google Scholar]
  35. Krakow K, Woermann FG, Symms MR, Allen PJ, Lemieux L, Barker GJ, Fish DR, et al. EEG-triggered functional MRI of interictal epileptiform activity in patients with partial seizures. Brain. 1999;122(9):1679–1688. doi: 10.1093/brain/122.9.1679. [DOI] [PubMed] [Google Scholar]
  36. Lemieux L, Salek-Haddadi A, Josephs O, Allen P, Toms N, Scott C, Fish DR, et al. Event-related fMRI with simultaneous and continuous EEG: description of the method and initial case report. Neuroimage. 2001;14(3):780–787. doi: 10.1006/nimg.2001.0853. [DOI] [PubMed] [Google Scholar]
  37. LeVan P, Gotman J. Independent component analysis as a model-free approach for the detection of BOLD changes related to epileptic spikes: a simulation study. Hum Brain Mapp. 2009;30(7):2021–2031. doi: 10.1002/hbm.20647. [DOI] [PMC free article] [PubMed] [Google Scholar]
  38. Mirbagheri M, Hakimi N, Ebrahimzadeh E, Pourrezaei K, Setarehdan SK (2019) Enhancement of optical penetration depth of LED-based NIRS systems by comparing different beam profiles. Biomed Phy Eng Express 5(6):065004. https://iopscience.iop.org/article/10.1088/2057-1976/ab42d9/meta
  39. Mirbagheri M, Hakimi N, Ebrahimzadeh E, Setarehdan SK. Quality analysis of heart rate derived from functional near-infrared spectroscopy in stress assessment. Inform Medicine Unlocked. 2020;18:100286. doi: 10.1016/j.imu.2019.100286. [DOI] [Google Scholar]
  40. Mirbagheri M, Hakimi N, Ebrahimzadeh E, Setarehdan SK. Simulation and in vivo investigation of LED-NIR Gaussian beam profile. J Near Infrared Spectrosc. 2020;28(1):37–50. doi: 10.1177/0967033519884209. [DOI] [Google Scholar]
  41. Pedreira C, Vaudano AE, Thornton RC, Chaudhary UJ, Vulliemoz S, Laufs H, et al. Classification of EEG abnormalities in partial epilepsy with simultaneous EEG–fMRI recordings. Neuroimage. 2014;99:461–476. doi: 10.1016/j.neuroimage.2014.05.009. [DOI] [PubMed] [Google Scholar]
  42. Rahimpour A, Dadashi A, Soltanian-Zadeh H, Setarehdan SK (2017) Classification of fNIRS based brain hemodynamic response to mental arithmetic tasks. In: 2017 3rd International conference on pattern recognition and image analysis (IPRIA), pp 113–117
  43. Rahimpour A, Noubari HA, Kazemian M. A case-study of NIRS application for infant cerebral hemodynamic monitoring: a report of data analysis for feature extraction and infant classification into healthy and unhealthy. Inform Med Unlock. 2018;11:44–50. doi: 10.1016/j.imu.2018.04.001. [DOI] [Google Scholar]
  44. Ridley B, Wirsich J, Bettus G, Rodionov R, Murta T, Chaudhary U, et al. Simultaneous intracranial EEG-fMRI shows inter-modality correlation in time-resolved connectivity within normal areas but not within epileptic regions. Brain Topogr. 2017;30(5):639–655. doi: 10.1007/s10548-017-0551-5. [DOI] [PubMed] [Google Scholar]
  45. Salek-Haddadi A, Diehl B, Hamandi K, Merschhemke M, Liston A, Friston K, Lemieux L, et al. Hemodynamic correlates of epileptiform discharges: an EEG-fMRI study of 63 patients with focal epilepsy. Brain Res. 2006;1088(1):148–166. doi: 10.1016/j.brainres.2006.02.098. [DOI] [PubMed] [Google Scholar]
  46. Schöller H, Viol K, Aichhorn W, et al. Personality development in psychotherapy: a synergetic model of state-trait dynamics. Cogn Neurodyn. 2018;12:441–459. doi: 10.1007/s11571-018-9488-y. [DOI] [PMC free article] [PubMed] [Google Scholar]
  47. Spencer NJ, Bywater RAR, Holman ME, Taylor GS. Inhibitory neurotransmission in the circular muscle layer of mouse colon. J Auton Nerv Syst. 1998;70(1–2):10–14. doi: 10.1016/S0165-1838(98)00045-9. [DOI] [PubMed] [Google Scholar]
  48. Vulliemoz S, Carmichael DW, Rosenkranz K, Diehl B, Rodionov R, Walker MC, Lemieux L, et al. NeuroImage Simultaneous intracranial EEG and fMRI of interictal epileptic discharges in humans. NeuroImage. 2011;54(1):182–190. doi: 10.1016/j.neuroimage.2010.08.004. [DOI] [PubMed] [Google Scholar]
  49. Worsley KJ, Liao CH, Aston J, Worsley K (2002) 10.1.1.93.3431, 1–27
  50. Zijlmans M, Huiskamp G, Hersevoort M, Seppenwoolde J-H, van Huffelen AC, Leijten FSS. EEG-fMRI in the preoperative work-up for epilepsy surgery. Brain. 2007;130(9):2343–2353. doi: 10.1093/brain/awm141. [DOI] [PubMed] [Google Scholar]

Articles from Cognitive Neurodynamics are provided here courtesy of Springer Science+Business Media B.V.

RESOURCES