Journal of Neuroscience Methods 184 (2009) 213–223

Contents lists available at ScienceDirect

Journal of Neuroscience Methods journal homepage: www.elsevier.com/locate/jneumeth

Real-time artifact filtering in continuous VEPs/fMRI recording Muhammad Nabeel Anwar a,b,∗ , Laura Bonzano c,d , Davide Rossi Sebastiano e , Luca Roccatagliata c,d , Giovanni Gualniera f , Paolo Vitali e , Carla Ogliastro c , Luciano Spadavecchia g , Guido Rodriguez e , Vittorio Sanguineti b , Pietro Morasso b , Fabio Bandini c a

Department of Computational Intelligence and Systems Science, Tokyo Institute of Technology, G3-50, 4259 Nagatsuta, Midori-ku, Yokohama 226-8502, Japan Department of Informatics, Systems and Telecommunications, University of Genoa, Genoa, Italy Department of Neurosciences, Ophthalmology and Genetics, University of Genoa, Genoa, Italy d Magnetic Resonance Research Centre on Nervous System Diseases, University of Genoa, Genoa, Italy e Clinical Neurophysiology, University of Genoa, Genoa, Italy f EBNeuro S.p.A., Florence, Italy g Institute of Cybernetics and Biophysics, National Research Council, Genoa, Italy b c

a r t i c l e

i n f o

Article history: Received 23 March 2009 Received in revised form 4 August 2009 Accepted 5 August 2009 Keywords: Continuous recording Real-time filtering Functional Magnetic Resonance Imaging (fMRI) Steady-state and transient Visual Evoked Potentials (VEPs) Optic neuritis

a b s t r a c t Continuous recording of Visual Evoked Potentials (VEPs) and functional Magnetic Resonance Imaging (fMRI) exploits the VEPs high temporal resolution and the fMRI high spatial resolution. In this work, we present a new method of continuous VEPs/fMRI recording to study visual function in seven normal subjects. Our real-time artifact filtering is characterized by a procedure based on an analytical study of echo-planar imaging (EPI) sequence parameters related electro-encephalogram (EEG)-artifact shapes. The magnetic field artifacts were minimized by using a dedicated amagnetic device and by a subtraction algorithm that takes into account the EPI sequence parameters. No significant decrease in signal-to-noise ratio was observed in case of EEG recording simultaneously with MR acquisition; similarly, transient and steady-state VEPs parameters were comparable during fMRI acquisition and in the off-phase of fMRI recording. We also applied this method to one patient with optic neuritis, and, compared with controls, found different results. We suggest that our technique can be reliably used to investigate the function of human visual cortex and properly correlate the electrophysiological and functional neuroimaging related changes. © 2009 Elsevier B.V. All rights reserved.

1. Introduction Visual Evoked Potentials (VEPs) allow to observe brain activity in response to visual stimuli at a millisecond time scale but with poor spatial resolution (Regan, 1989). Conversely, functional Magnetic Resonance Imaging (fMRI) exploits changes in the BOLD signal coupled with neuronal activity in order to localize brain activation, with a high spatial resolution but low temporal resolution (Ogawa et al., 1992). VEPs and fMRI have thus complementary features and their integration may provide more detailed information than either method alone (Di Russo et al., 2005; Henning et al., 2005). Given the inherent variability of evoked cortical responses and BOLD contrast, the integration of both methods would be more reliable using simultaneous recordings, which explore the subject’s brain activity in the same state and provide precise information derived from different biophysical origins on

∗ Corresponding author. Tel.: +81 45 924 5654; fax: +81 45 924 5684. E-mail address: [email protected] (M.N. Anwar). 0165-0270/$ – see front matter © 2009 Elsevier B.V. All rights reserved. doi:10.1016/j.jneumeth.2009.08.003

the processing of visual stimuli at the same time (Bonmassar et al., 2001). The feasibility of recording EEG signals inside an MR scanner was initially reported by Ives et al. (1993) and later by other researchers (Huang-Hellinger et al., 1995; Lemieux et al., 1997). These studies also investigated the possibility of recording a good quality EEG signal inside an MR scanner, as well as to obtain high quality MR images in the presence of EEG equipment, assuring patient safety. These findings opened the way to the acquisition of EEG and fMRI in the study of epilepsy (Bast et al., 2007; Hamandi et al., 2006), the neural mechanisms of sleep (Czisch et al., 2004; Horovitz et al., 2008; Schabus et al., 2007) and the VEPs (Becker et al., 2005; Bonmassar et al., 1999; Comi et al., 2005; Henning et al., 2005; Im et al., 2006; Lemieux et al., 1997; Sammer et al., 2005). Three methods are available in order to combine VEPs and fMRI data: (1) the recordings are made separately and data are fused subsequently (“conjoint VEPs–fMRI”); (2) the VEPs are recorded inside the MR scanner but the EEG data analysis is performed only during periods without MR acquisition (“interleaved VEPs–fMRI”) (Bonmassar et al., 1999, 2001); in this case EEG–fMRI recordings

214

M.N. Anwar et al. / Journal of Neuroscience Methods 184 (2009) 213–223

are not completely simultaneous as the measurements are made in two separate time-windows; (3) the acquisition of VEPs and fMRI is entirely simultaneous and the whole EEG trace, including periods obtained during MR-acquisition, is used for analysis (“continuous VEPs–fMRI”). Regardless of whether long-lasting or rapidly alternating MR and EEG epochs were acquired, the first two methods share the disadvantages of being time-consuming (particularly for evoked potential studies) and hampering a real investigation of the subject’s brain activity in the same environmental and cognitive condition; the second method has also restrictive protocols, since only periods without MR acquisition are used in the EEG data analysis. Continuous VEPs–fMRI acquisition is very challenging since the fast varying gradient fields and the radio-frequency pulses induce a very intense electric noise that obscures the EEG signal (HuangHellinger et al., 1995; Ives et al., 1993), thus requiring a careful experimental setup and a sophisticated artifact removal technique in order to filter out the noise and retains the evoked responses in the signal. Such a challenging technique has been applied quite rarely (Becker et al., 2005; Comi et al., 2005; Henning et al., 2005; Im et al., 2006; Sammer et al., 2005), by means of the same filtering algorithm (Allen et al., 2000, 1998). One study investigated steadystate VEPs (SS-VEPs) (Sammer et al., 2005), while other authors recorded transient VEPs (T-VEPs) using a stimulation based on a sequence of letters (Comi et al., 2005), pattern reversal (Becker et al., 2005; Im et al., 2006), and motion patterns (Henning et al., 2005). Here we propose a new method of real-time artifact filtering in continuous VEPs (both T-VEPs and SS-VEPs) and fMRI acquisition, at low (1 cycle per degree) and medium-high (4 cycles per degree) spatial frequencies. In the present study, we analyze the electromagnetic interference in the shape and in space-temporal features. This estimation of echo-planar imaging (EPI) sequence parameters in EEG data gives the possibility to build a model that can be used as a reference (template) for filtering. Each raw sample is associated with the corresponding (in time) sample in the model; cleaned signal is the difference between raw sample and model. Recent commercially available devices are based on procedures to remove post-processing artifacts, thus hampering the appropriate control of the electrophysiological activity of the subject during fMRI scan. In this work we propose to use an online filtering technique that allows to monitor the EEG signal during the fMRI acquisition, which is particularly important for clinical investigations (e.g., optic neuritis, epilepsy) and for behavioral studies on the relationship between electrophysiological changes and brain activity during learning. Our aim was to evaluate whether the EEG signal could be restored from fMRI artifacts in order to obtain reliable VEPs. We chose to record both T-VEPs and SS-VEPs because they may tap independent pathways in the visual system (Bodis-Wollner et al., 1990). In addition, we applied our method in one patient with optic neuritis (ON), a clinical condition characterized by a conduction delay in the optic nerve, in order to investigate whether it is possible to detect different responses with respect to control subjects.

2. Materials and methods

left pupil to light stimulation as compared with the right pupil, on a swinging flashlight test). There was no swelling of the optic disk, and the remainder neurological examination was normal. The study protocol was approved by the Institutional Review Board, and written informed consent was obtained from all subjects in accordance with the declaration of Helsinki. 2.2. Experimental protocol The visual stimulation protocol consisted of full-field checkerboard patterns with a rectangular luminance profile, presented in a reversal mode with repetition frequencies of 1 Hz (T-VEPs) and 4 Hz (SS-VEPs). Two different fundamental spatial frequencies (SF) of 1 and 4 cycles per degree (cpd) were used. Visual stimulation was monocular: each eye was tested separately, while the contralateral eye was covered by a translucent patch to maintain light adaptation. A custom-made software, running under the MS-DOS operating system, was used to deliver the visual stimuli. A digital trigger signal was generated at each pattern reversal and sent to the EEG data acquisition hardware for synchronization. The projector was located outside the MR room. The black and white checkerboard pattern was projected on an acrylic screen inside the MRI room. The screen mean luminance was 60 cd/m2 , the contrast of the patterns was 70%. A mirror was placed on the head coil at 45◦ to the screen and the subject’s line of sight. The stimulation protocol consisted of alternating blocks of 20 s of rest (i.e., display of grey), 20 s of checkerboard pattern reversal at 1 cpd and 20 s of checkerboard pattern reversal at 4 cpd (Fig. 1). In the rest condition, the screen had the same luminance as the mean luminance of the pattern. Subjects were asked to gaze a fixation point at the centre of the screen with the uncovered eye. The sequence was repeated three times for each temporal frequency for each eye. We also implemented the procedure on an ON patient, who underwent VEPs recording outside the MRI room using the same stimulation paradigm. In addition, one healthy subject participated in two types of recording: one during fMRI acquisition (EPI condition) and one in the absence of fMRI (NO-EPI condition). Both recordings were made in a single continuous session. 2.3. Anatomical and functional MRI data acquisition MR images were acquired on a 1.5 T scanner (General Electric, Milwaukee, USA) with a phase-array eight-channel coil. Anatomical images were acquired using a 3D SPGR (Spoiled Gradient Recalled) sequence (TR/TE = 30/3 ms; slice thickness = 3 mm; FOV = 240 mm; Matrix: 256 × 256; Flip Angle = 35◦ ). Functional images were acquired using echo-planar imaging (EPI) sequences (TR/TE = 3000/50 ms; delay = 1000 ms; slice thickness = 6 mm; FOV = 260 mm; Matrix = 64 × 64). Functional MRI was acquired during the above-described stimulation according to a block design (Fig. 1). The left optic nerve involvement in the patient with ON was demonstrated using coronal acquisitions (short inversion time inversion recovery TR/TE = 2000/40 ms; TI = 148 ms; slice thickness = 3 mm; FOV = 220 mm; Matrix 128 × 256) and FSE T2 through the orbits (TR/TE = 6025/102 ms; slice thickness = 3 mm; FOV = 240 mm; fat saturation; Matrix 256 × 320).

2.1. Subjects We acquired fMRI data and VEPs in seven healthy volunteers (mean age 30 ± 2 years) with normal vision. We also studied one 28year-old patient with acute left ON. He presented with a one-week history of painful blurred vision in the left eye, accompanied by worsening of pain on eye movement. Examination revealed visual acuity of 1/10 in the left eye, impaired perception of color, and a left relative afferent pupillary defect (RAPD) (poor response of the

Fig. 1. Block diagram of the stimulation protocol for continuous recording of VEPs and fMRI. The protocol started with a 20-s grey display, followed by 20-s pattern reversal black and white checkerboard with spatial frequency of 1 and 4 cpd. This sequence was repeated three times for each eye (stimulation was monocular) for each temporal frequency (1 and 4 Hz).

M.N. Anwar et al. / Journal of Neuroscience Methods 184 (2009) 213–223

2.4. EEG data acquisition For EEG recording we used an MR-compatible amplifier (EBNeuro Spa, Florence, Italy). The EEG device was placed inside a purpose-built shielded box. The signal amplification and A/D conversion were performed and the digital signal was transferred via an optical fiber connection to a standard IBM laptop running EEG

215

data acquisition (Galileo NT) and data processing software (Garreffa et al., 2003). An MR-compatible standard 10–20 system EEG cap with Ag/AgCl electrodes filled with conductive gel was used (Bionen Sas, Florence, Italy); the interelectrode-skin impedance was lower than 5 k. The cap was connected to the amplifier with an amagnetic cable rested on foam pads attached to the ground and following a straight path in order to minimize the variability in

Fig. 2. (a) EEG signal with fMRI artifacts during EPI acquisition (TR = 3 s) and (b) EEG signal filtered from artifacts by gradient artifact reduction with time-based adaptive algorithm.

216

M.N. Anwar et al. / Journal of Neuroscience Methods 184 (2009) 213–223

the EEG signal due to the MR sequences and to avoid any possible cable movement caused by mechanical vibrations. The amplifier was set at 4 kHz sampling rate, which allows a suitable time resolution to pick up the switching effect of the readout gradient in the high slew rate condition, 20 bit resolution, and the EEG dynamic range of ±65.5 mV to prevent saturation of the EEG/ECG by the MRI artifact waveforms. EEG data were recorded from 19 electrodes positioned according to the International 10–20 systems (Fp1, Fp2, F3, F4, F7, F8, Fz, C3, C4, Cz, P3, P4, Pz, T3, T4, T5, T6, O1, O2), with right earlobe as reference (off-line EEG trace was recomputed to a common average). The ground electrode was placed on FCz (halfway position between Fz and Cz). The ECG signal was synchronously recorded, with the same amplitude resolution as EEG channels. The stimulation trigger signal was acquired at the same sampling rate as electrophysiological data.

2.5. EPI sequences evaluation and fMRI analysis First, in order to evaluate possible disturbances induced on MR images by the EEG acquisition system, the signal to noise ratio (SNR) was estimated in the presence and in the absence of EEG recording in 4 and 3 subjects, respectively. For each MR acquisition, the SNR was calculated as the ratio between the mean intensity value of 3 regions of interest (ROIs) placed in the occipital white matter and the standard deviation of 3 ROIs placed outside the brain in the background, avoiding ghosting artifacts. We calculated the SNR for different volumes during the course of each acquisition and considered the corresponding volumes (i.e., same acquisition phase) in the different acquisitions. fMRI volumes were then analyzed with SPM2 (Wellcome Department of Cognitive Neurology, London, UK) (Friston et al., 1995). All the images were realigned to the first image acquired for

Fig. 3. Shape of the artifacts in both temporal and frequency domain. (a) Single channel EEG recorded during EPI acquisition (TR = 3 s, delay = 1 s). (b) Main dominant frequencies during gradient switching events. A 1-min amplitude spectrum of one channel (Fz) recorded inside the MR scanner was computed by averaging 1-s epoch FFT signals. EPI sequence artifacts (gradients fields, static magnetic field and RF pulse) at low frequency range are magnified.

M.N. Anwar et al. / Journal of Neuroscience Methods 184 (2009) 213–223

217

Now consider that eeg(t) and n(t) have zero mean, then for M dynamics the averaged x(t) would be: xavg (t) = epiavg (t)

In this way, the EPI synchronously averaged EEG represents an estimate of the average waveform of the whole EPI interference (RF pulses and gradient fields) in the EEG channels. This artifact model was then synchronized with the EEG data by zero-phase band pass, i.e., each raw sample (uncleaned data) was associated with the corresponding (in time) sample in the model. The synchronization of the artifact model with the raw data was done by finding peaks for identifying the slices and the dynamics in the raw data. In the calibration mode, the peaks were identified by a threshold method: a sample is virtually considered a potential peak if its amplitude exceeds a given threshold (estimated by using the mean signal amplitude); the peaks were calculated individually for each channel. The time and the position of the identified peaks were also calculated and used to match their right position in the raw data. This made it possible to set apart spurious peaks because they “do not occur at the right moment”. Finally, this artifact model was applied on the continuous EEG recording and the cleaned signal was given by the difference between the raw sample and the model. A zero-phase 10th order Butterworth low-pass filter of 70 Hz was applied to the raw signal before subtraction, to increase filter quality. Fig. 2 displays artifact-free data; Fig. 3 shows the shape of the artifacts and the dominant frequencies during the gradient switching events and the slice timing vs. TR. Differently from other studies with continuous VEPs–fMRI recording (Becker et al., 2005; Comi et al., 2005; Henning et al., 2005; Im et al., 2006; Sammer et al., 2005), the cardiac pulse artifact (originated by the slight micromovements of the head due to heartbeat and normal to the static magnetic field) was limited by: (i) the head-locking of the subject, (ii) the spatial properties of bipolar montage and (iii) the proper twisted lead settings (Goldman et al., 2000). In this way, we were able to prevent data corruption and to rebuild reliable VEPs from the EEG signal. We assumed that the nature of the magnetic interference during the EPI sequence is completely known and does not depend on the type of subjects studied; therefore, it is possible to obtain similar results for studying VEPs signal in healthy and pathological subjects. To ensure preservation of information in VEPs after filtering MR artifacts in a simultaneous recording, we trained the model of the filter for the optimal parameters with EPIs without visual stimulation (i.e., the first 20 s). In this way any response elicited by the checkerboard was not filtered out.

Fig. 4. Comparison between the signal to noise ratio (SNR) calculated on EPI images in the presence of EEG recording and in the absence of EEG electrodes: no significant decrease in SNR was evident in case of EEG recording during MR acquisition.

each stimulation condition, and a mean functional image was created. The mean functional images were then normalized to a brain template and the resulting transformation matrix was applied to the individual functional volumes. The images were smoothed with 8 mm full-width at half maximum Gaussian kernel. Statistical analyses were performed using the general linear model (Chatfield and Collins, 1980). t-Test contrasts were conducted separately for data acquired at 1 and 4 Hz for both SF (1 and 4 cpd), with respect to the rest condition. Clusters of activated voxels surviving an uncorrected threshold of P < 0.001 were considered significant in the calculation of the corresponding activation maps. 2.6. Artifact removal in EEG recordings fMRI artifacts were removed from the EEG data by the real-time filtering software. The filtering algorithm calibrates itself from the first EPI acquisitions; when the EPI sequences run, the filter enters calibration mode and only few scans are needed to build a robust dynamic model. The filtering method is based on estimating the effect of electromagnetic interference on the shape and the space-temporal features of EEG data recorded inside the MRI scanner during the EPI sequence. It estimates the average waveform of the whole EPI interference (RF pulses and gradient fields) in all EEG channels and includes the events related to each volume acquisition (i.e., number of slices/volume, TR, TE, RF pulses, Gradient selection and Readout Gradients). The estimation of EPI sequence parameters in the raw EEG data gives the possibility of identifying artifacts that represent a reference (template) for filtering. These artifacts were evaluated and then computed with a standard evoked potential averaging method. Consider x(t) as a set of uncorrelated components, the EEG basal activity eeg(t), the instrumental noise n(t), and the EPI-evoked response epi(t), then: x(t) = eeg(t) + n(t) + epi(t)

(2)

2.7. Preprocessing for EEG filtered data After gradient artifact reduction with time-based adaptive algorithm, the data were down-sampled to 512 Hz, filtered with a 0.3–35 Hz band-pass filter and stored for further analysis. 2.8. General evaluation of artifact removal algorithm For each healthy subject we used all the epochs of stimulus data. The artifact-free signal was divided into “inter-scan” and “scan” data (inter-scan = 1000 ms; scan = 3000 ms). We estimated the percentage differences in the spectral density of five bands (1–4, 4–8, 8–12, 12–16, and 16–24 Hz) by applying a previously reported for-

(1)

Table 1 Median of percentage absolute power difference between scan and inter-scan periods.

%Difference O1 %Difference O2

1–4 Hz

4–7 Hz

8–12 Hz

12–16 Hz

16–25 Hz

Total

10.1010 11.6824

9.8648 10.1215

7.7346 7.7315

10.5556 11.2623

10.6642 11.7858

9.7840 10.5167

Median of percentage absolute power difference between scan and inter-scan periods for O1 and O2.

218

M.N. Anwar et al. / Journal of Neuroscience Methods 184 (2009) 213–223

mula (Allen et al., 2000):

   PIS − PS  % Difference = 100 ×   P IS

where PIS is the power density of inter-scan periods and PS is the power spectral density of scan periods. The analysis was performed for O1 and O2 for each frequency, and the median values were calculated across all healthy subjects. We also analyzed the noise cancellation effect inherent in averaging, for frequency range 1–35 Hz, by calculating the spectral densities as a function of the number of averaged trials. 2.9. VEPs analysis EEG epochs with eye blinking and eye movement artifacts were removed by visual inspection (about 15% of data was rejected). To characterize the response to 1 Hz stimulation for both inter-scan and scan data, we computed T-VEPs by averaging the epochs. We then estimated the inter-scan and scan period difference in the peak latency of the P100 component. For SS-VEP analysis, a 1024 points Fourier transform was applied on inter-scan and scan epochs of all subjects; it yielded nearly independent Fourier estimates (Victor and Mast, 1991) at the stimulus fundamental frequency and its 2nd harmonic. These Fourier estimates were vector averaged to determine the steady-state response. Signal reliability was assessed by 2 statistics (Victor and Mast, 1991), which analyze whether the Tcirc Fourier vector at any particular frequency is significantly differ-

ent than the null vector. The amplitude and phase of the response vector represent the distance from the origin and the direction respectively. In plots of amplitude and phase, the 95% error bars show the largest and smallest distance from this confidence circle to the origin, determined separately for each eye. VEPs were calculated for the ON patient with the same procedure as stated above. The latency of T-VEP P100 was measured. For the SS-VEPs, we calculated the coherence level at the 2nd harmonic (8 Hz) of the stimulation frequency; coherence was calculated as an indicator of evoked response on occipital brain areas, involved in detection and analysis of visual stimuli. The phase angle for SS-VEPs was calculated by Fourier analysis. The procedure was carried out for both eyes and the results of the ON patient were compared with the healthy subjects. For the subject tested in both EPI and NO-EPI conditions, dynamic field and static field artifact filtering models were applied on EEG data with EPI sequence whilst only the static field artifact filtering model was applied in NO-EPI condition. Results were compared in terms of P100 shape, amplitude and latency (T-VEPs) and in terms of coherence for the 2nd harmonic (SS-VEPs). 3. Results 3.1. EPI sequences evaluation and fMRI analysis The SNR in EPI images acquired during EEG recording was compared with the SNR of EPI images in absence of EEG electrodes

Fig. 5. Brain activation maps of a representative control subject (a and b) and an MS (ON) patient (c and d) at a temporal stimulation of 4 Hz and spatial frequency of 1 cpd (a and c) and 4 cpd (b and d). A reduction in the patient visual cortex activation can be noted at both 1 and 4 cpd.

M.N. Anwar et al. / Journal of Neuroscience Methods 184 (2009) 213–223

219

Fig. 6. Power spectrum density (1–35 Hz) on the number of averaged trials for scan and inter-scan data. Top: averaged power spectra for O1 and O2. Bottom: power spectrum dependency on number of averaged trials for O2.

(Fig. 4). No significant decrease in SNR was observed in case of EEG recording simultaneously with MR acquisition (Mann–Whitney UTest, p = 0.5). The activation maps of a representative subject are shown in Fig. 5. 3.2. Effect of filtering on EEG signal The inter-scan data show a frequency spectrum similar to the spectrum of the scan period. Table 1 presents the median percentage difference of spectral power density in 5 bands at O1 and O2. Fig. 6 shows the power spectral dependency as a function of number of trials and spectral power density. In small number of trials, the effect of noise cancellation for non-phase-lock components is similar in inter-scan and scan periods. This indicates that no residual of MRI artifacts are present in low frequency range.

3.3. T-VEPs The shape of the P100 component was identifiable after applying the filtering technique on EEG data acquired during fMRI acquisition. The data were first averaged within each subject, then aligned to the onset of P100 component and averaged over all subjects. The mean and the standard deviation of the latency of P100 component in inter-scan and scan period was measured as 101.3 ± 2.16 ms for 1c/d and 101.3 ± 2.8 ms for 4c/d (Fig. 7). 3.4. SS-VEP 2 analysis provided us with significant results. The The Tcirc null hypothesis that the signal does not contain the fundamental and the 2nd harmonic was rejected for both inter-scan and

Fig. 7. Averaged VEP response from all healthy subjects. The bold lines represent group averaged data and the dashed lines indicate ± SD (only data from O2 is shown).

220

M.N. Anwar et al. / Journal of Neuroscience Methods 184 (2009) 213–223

Table 2 2 , Amplitude in microvolt and phase values in radians of the SS-VEP for the fundamental frequency and 2nd harmonic obtained from healthy subjects. shows the Tcirc O1

O2

1st Harmonic

2 Tcirc Amplitude Phase

2nd Harmonic

1st Harmonic

2nd Harmonic

Scan

Inter-scan

Scan

Inter-scan

Scan

Inter-scan

Scan

Inter-scan

12.53 ± 0.4 6.22 ± 1.0 −0.12 ± 0.6

11.65 ± 0.4 5.39 ± 0.8 −0.3 ± 0.6

12.36 ± 0.4 5.34 ± 0.8 −0.4 ± 0.5

11.61 ± 0.5 4.89 ± 0.9 −0.4 ± 0.7

12.15 ± 0.4 6.15 ± 0.6 −0.3 ± 0.6

11.22 ± 0.5 4.95 ± 0.8 −0.1 ± 0.6

10.87 ± 0.5 4.72 ± 0.7 −0.2 ± 0.5

12.19 ± 0.4 4.24 ± 0.9 −0.3 ± 0.7

2 = scan data (p < 0.05). For scan data the mean ± SEM was Tcirc 2 12.5 ± 0.41 for the fundamental frequency; Tcirc = 12.3 ± 0.49 for 2 = 12.15 ± 0.47 for the fundathe 2nd harmonic at O1 and Tcirc 2 mental frequency; Tcirc = 10.87 ± 0.48 for the 2nd harmonic at 2 = 11.65 ± 0.43 O2. For inter-scan data the mean ± SEM was Tcirc 2 for the fundamental frequency, Tcirc = 11.61 ± 0.51 for the 2nd 2 = 11.22 ± 0.53 for the fundamental freharmonic at O1 and Tcirc 2 quency, Tcirc = 12.19 ± 0.43 for the 2nd harmonic at O2. The mean 2 , amplitudes and phases for the control group are summarized Tcirc in Table 2 and plotted in Fig. 8. For the subject tested in both EPI and NO-EPI conditions, at 1 Hz no significant changes were observed between the filtered signal acquired during fMRI scanning and the signal acquired in the same conditions but in absence of fMRI scanning (Fig. 9a). The 2nd harmonic response at O1 is shown in Fig. 9b: a defined peak at 8 Hz can

be observed in both cases. The coherence level remains the same in both conditions. 3.5. VEPs analysis: ON patient The SS-VEP response of the normal eye and the ON eye is shown in Fig. 10a (1 cpd) and in Fig. 10b (4 cpd). The amplitude of SSVEP decreased sharply in the affected eye at high spatial frequency while at low spatial frequency the responses were similar. We used coherence as an indicator for the magnitude of the stimulus-evoked response on a particular brain area, e.g., O1 and O2 (Fig. 11a). As expected, the coherence between stimulus and affected eye of the patient was very low as compared to the same eye of a normal subject, which indicated the absence of response in the affected eye. When we analyzed the phase lag of the 2nd harmonic response, it proved to be significantly longer in the ON eye compared to the

2 values at fundamental frequency (black) and 2nd harmonic (grey). The values were derived at Fig. 8. SS-VEP responses of subjects for the left eye stimulation: (a) the Tcirc O1 and O2 for scan and inter-scan period, the error bars represent the SEM. (b) The Fourier components at O1 and O2 in terms of amplitude and phase was calculated for each subject at fundamental stimulus frequency and its 2nd harmonic. Amplitude and phase represent the distance and the direction from the origin, respectively, with 95% confidence level.

M.N. Anwar et al. / Journal of Neuroscience Methods 184 (2009) 213–223

221

Fig. 9. Effect of filtering on VEPs: filtered signals acquired during fMRI scanning (left panel) and signals acquired in absence of fMRI scanning (right panel). (a) T-VEPs: no significant signal changes could be observed. (b) SS-VEPs: a defined peak at 8 Hz could be observed in the 2nd harmonic response at O1 for the right eye stimulation; the coherence level remained unchanged in both conditions.

fellow eye, while no difference was found in the control subjects (Fig. 11b). 3.6. Integration of SS-VEPs and fMRI The continuous recording of VEPs and fMRI allowed us to integrate the data obtained from the two techniques. Fig. 11c shows the combined results of SS-VEPs and fMRI for both SF. The responses of the normal and the affected eye of the patient were plotted as phase lag and coherence of VEPs against the activated brain volume in fMRI. The affected eye had a longer latency and a decreased coherence with respect to the normal eye. The activated volume in visual areas (Brodmann areas 17, 18 and 19) of the patient was significantly smaller than the controls. The activation maps of a representative control subject and the patient for 4 Hz stimulation are shown in Fig. 5. A reduction of the activated brain areas can be noticed in the visual cortex activation maps of the ON patient at both 1 and 4 cpd. 4. Discussion Continuous EEG/fMRI recording was first reported for the analysis of epileptic events (Lemieux et al., 2001). Notwithstanding the reported usefulness of combining these two techniques (Di

Russo et al., 2005; Menon et al., 1997), only a few studies have been conducted so far in order to record VEPs (Becker et al., 2005; Comi et al., 2005; Henning et al., 2005; Im et al., 2006; Sammer et al., 2005). The interference with the MR system, which results in distortion of the EEG signal recordings (i.e., static magnetic field interference, dynamic magnetic field interference and cardiac pulse interference), could account for the lack of this kind of study. The disadvantage of a continuous recording approach could be the possible loss of EEG information due to artifacts (Becker et al., 2005). In addition, a pulse artifact is induced by blood flow inside the head, normal to the static magnetic field. To address these problems an adaptive subtraction filtering algorithm (Allen et al., 2000, 1998) was used in previous studies (Becker et al., 2005; Comi et al., 2005; Henning et al., 2005; Im et al., 2006; Sammer et al., 2005). Such an algorithm is mainly tailored to remove post-processing artifacts. Consequently, it has neither control on the electrophysiological changes occurring during fMRI scan nor a strong ‘prior knowledge’ of the artifact, which is a mandatory condition to avoid loss of information on the physiological signals recovered after filtering. In a SS-VEPs study, it was reported that gradient and/or ballistocardiographic artifacts could cause small distortions of the periodical behavior of the SS-VEP in the scanner, removing some of the magnitude of the targeted peak and shifting its frequency to a slightly lower value than the stimulation frequency (Sammer et

222

M.N. Anwar et al. / Journal of Neuroscience Methods 184 (2009) 213–223

Fig. 10. (a) The SS-VEP response of the fellow eye and the ON eye for 4 Hz–1 cpd. (b) The SS-VEP response of the fellow eye and the ON eye for 4 Hz–4 cpd. The amplitude of SS-VEP decreased sharply in the ON eye at high spatial frequency while at low spatial frequency the response amplitudes were similar.

al., 2005). The differences in latency of VEPs in inter-scan periods and outside the scanner have been already addressed (Bonmassar et al., 1999). Other authors investigated the differences in the VEPs recorded during inter-scan period and scan period (Becker et al., 2005). In this study we emphasized the feasibility of our real-time fMRI artifact filtering and tested its validity for both T-VEP and SS-VEP in time and frequency domain at low and medium-high spatial frequencies. The main feature of our real-time artifact filtering is the preliminary analytical study of EPI sequence parameters related EEG-artifact shapes, which also allows a further post-processing procedure. The basic idea of the filtering method adopted in this experiment is an identification of the electromagnetic interference in its shape and space-temporal features. The exact estimation of the effects of EPI sequence parameters on the raw EEG data gives the possibility of identifying an analytical expression of “artifacts function”. This represents a reference (template) for filtering and also permits to find “endogenous” trigger events. By utilizing our algorithm we were able to filter the data with fMRI artifacts and to reconstruct a cleaned EEG signal. Our results show that artifactcorrected scan data had similar spectrum to inter-scan data with an average difference of about 10% in five different bands, similar to a previous study with offline filtering of fMRI artifacts from EEG data (Becker et al., 2005). No difference was found between interscan and scan VEPs after real-time artifact removal both in P100 latency and shape for transient evoked potentials and in the amplitude and phase response for steady-state evoked potentials. The latter are characterized by the amplitude and phase of the Fourier components at one or more frequency of interest. In addition, EEG acquisition in two different conditions (i.e., acquisition inside the MR scanner during EPI acquisition and acquisition in the MR scanner without EPI gradients applied) was compared in one subject in terms of shape, amplitude and latency of

P100 component in T-VEP and coherence and phase lag of the 2nd harmonic in SS-VEP. The shape and the latency of P100 component were preserved after applying filtering technique. The amplitude of the 2nd harmonic response of the stimulation frequency was unchanged and a clear peak was observed at 8 Hz. In the affected eye of the ON patient, T-VEPs recording showed the typical latency delay of the P100 component (Halliday et al., 1972) and SS-VEPs demonstrated a significantly longer phase lag of the 2nd harmonic, which has been associated with elongation of neuronal processing time. Our fMRI results were consistent with previous findings in patients with acute ON, with a dramatic reduction of volume of activation in the primary visual cortex in response to stimulation of the affected eye (Gareau et al., 1999; Rombouts et al., 1998; Russ et al., 2002). At a variance with either conjoint or interleaved recording of VEPs and fMRI, the continuous acquisition of electrophysiological and BOLD signals allowed us to reliably and accurately integrate VEPs and fMRI data. We were therefore able to analyze the correlation of the phase lag or the coherence of the 2nd harmonic response at 4 Hz for both 1 and 4 cpd with the activated volumes. We found a negative correlation between the phase lag of the 2nd harmonic response of the affected eye and the activated volumes. We also found a positive correlation between the coherence at the 2nd harmonic response and the activated volumes. Caution is needed to draw conclusions from data obtained in a single patient; however our findings might suggest that the electrophysiological and functional imaging changes in the visual brain during ON can be directly correlated. Continuous recording of VEPs and fMRI provides noninvasive, precise and reliable information on visual processing by studying the subjects during exactly the same states of visual brain activity. Furthermore, we showed that real-time fMRI artifact filtering, combined with an optimal EPI protocol, does not modify the mor-

M.N. Anwar et al. / Journal of Neuroscience Methods 184 (2009) 213–223

Fig. 11. (a) Distribution of coherence: in a control subject the coherence of 2nd harmonic response at the occipital sites had a higher value than the patient’s affected eye. (b) Polar plot of the averaged phase response of all the subjects and for the patient (affected eye). The response was measured at O1 and O2. A decrease in the amplitude and phase lag of the 2nd harmonic is evident for the affected eye. (c) Integration of phase and coherence with the activated volumes at occipital cortex: the 2nd harmonic response of the affected eye (♦) had a longer latency than the fellow eye (+). The value of coherence decreased in the affected eye at 4 cpd. The activated volume in visual areas (Brodmann areas 17, 18 and 19) by stimulation of the affected eye was smaller than the fellow eye.

phology and the amplitude of evoked potentials recorded from the visual cortex. It also allows to obtain reliable VEPs, proving to be a reliable tool for studying human visual function. Only a true concurrent approach enables proper correlation of data derived from different biophysical sources on the processing of visual stimuli at the same time. This is crucial for a better understanding of the relationship between neurophysiological and hemodynamic responses of the activated visual cortex. Even if recorded in just one patient, continuous VEP–fMRI recording seems to be also promising for clinical application in the study of visual function in optic neuropathies. References Allen PJ, Josephs O, Turner R. A method for removing imaging artifact from continuous EEG recorded during functional MRI. Neuroimage 2000;12:230–9.

223

Allen PJ, Polizzi G, Krakow K, Fish DR, Lemieux L. Identification of EEG events in the MR scanner: the problem of pulse artifact and a method for its subtraction. Neuroimage 1998;8:229–39. Bast T, Wright T, Boor R, Harting I, Feneberg R, Rupp A, et al. Combined EEG and MEG analysis of early somatosensory evoked activity in children and adolescents with focal epilepsies. Clin Neurophysiol 2007;118:1721–35. Becker R, Ritter P, Moosmann M, Villringer A. Visual evoked potentials recovered from fMRI scan periods. Hum Brain Mapp 2005;26:221–30. Bodis-Wollner I, Brannan J, Ghilardi M, Mylin L. The importance of physiology to visual evoked potentials. In: Desmedt J, editor. Visual evoked potentials. Amsterdam: Elsevier; 1990. p. 1–24. Bonmassar G, Anami K, Ives J, Belliveau JW. Visual evoked potential (VEP) measured by simultaneous 64-channel EEG and 3T fMRI. Neuroreport 1999;10:1893–7. Bonmassar G, Schwartz DP, Liu AK, Kwong KK, Dale AM, Belliveau JW. Spatiotemporal brain imaging of visual-evoked activity using interleaved EEG and fMRI recordings. Neuroimage 2001;13:1035–43. Chatfield C, Collins AJ. Introduction to multivariate analysis. London: Chapman and Hall; 1980. Comi E, Annovazzi P, Silva AM, Cursi M, Blasi V, Cadioli M, et al. Visual evoked potentials may be recorded simultaneously with fMRI scanning: a validation study. Hum Brain Mapp 2005;24:291–8. Czisch M, Wehrle R, Kaufmann C, Wetter TC, Holsboer F, Pollmacher T, et al. Functional MRI during sleep: BOLD signal decreases and their electrophysiological correlates. Eur J Neurosci 2004;20:566–74. Di Russo F, Pitzalis S, Spitoni G, Aprile T, Patria F, Spinelli D, et al. Identification of the neural sources of the pattern-reversal VEP. Neuroimage 2005;24:874– 86. Friston KJ, Holmes AP, Poline JB, Grasby PJ, Williams SC, Frackowiak RS, et al. Analysis of fMRI time-series revisited. Neuroimage 1995;2:45–53. Gareau PJ, Gati JS, Menon RS, Lee D, Rice G, Mitchell JR, et al. Reduced visual evoked responses in multiple sclerosis patients with optic neuritis: comparison of functional magnetic resonance imaging and visual evoked potentials. Mult Scler 1999;5:161–4. Garreffa G, Carni M, Gualniera G, Ricci GB, Bozzao L, De Carli D, et al. Real-time MR artifacts filtering during continuous EEG/fMRI acquisition. Magn Reson Imaging 2003;21:1175–89. Goldman RI, Stern JM, Engel Jr J, Cohen MS. Acquiring simultaneous EEG and functional MRI. Clin Neurophysiol 2000;111:1974–80. Halliday AM, McDonald WI, Mushin J. Delayed visual evoked response in optic neuritis. Lancet 1972;1:982–5. Hamandi K, Salek-Haddadi A, Laufs H, Liston A, Friston K, Fish DR, et al. EEG–fMRI of idiopathic and secondarily generalized epilepsies. Neuroimage 2006;31:1700–10. Henning S, Merboldt KD, Frahm J. Simultaneous recordings of visual evoked potentials and BOLD MRI activations in response to visual motion processing. NMR Biomed 2005;18:543–52. Horovitz SG, Fukunaga M, de Zwart JA, van Gelderen P, Fulton SC, Balkin TJ, et al. Low frequency BOLD fluctuations during resting wakefulness and light sleep: a simultaneous EEG–fMRI study. Hum Brain Mapp 2008;29:671–82. Huang-Hellinger FR, Breiter HC, McCormack G, Cohen MS, Kwong KK, Sutton JP, et al. Simultaneous functional magnetic resonance imaging and electrophysiological recording. Hum Brain Mapp 1995;3:13–23. Im CH, Liu Z, Zhang N, Chen W, He B. Functional cortical source imaging from simultaneously recorded ERP and fMRI. J Neurosci Methods 2006;157:118–23. Ives JR, Warach S, Schmitt F, Edelman RR, Schomer DL. Monitoring the patient’s EEG during echo planar MRI. Electroencephalogr Clin Neurophysiol 1993;87:417– 20. Lemieux L, Allen PJ, Franconi F, Symms MR, Fish DR. Recording of EEG during fMRI experiments: patient safety. Magn Reson Med 1997;38:943–52. Lemieux L, Salek-Haddadi A, Josephs O, Allen P, Toms N, Scott C, et al. Event-related fMRI with simultaneous and continuous EEG: description of the method and initial case report. Neuroimage 2001;14:780–7. Menon V, Ford JM, Lim KO, Glover GH, Pfefferbaum A. Combined event-related fMRI and EEG evidence for temporal-parietal cortex activation during target detection. Neuroreport 1997;8:3029–37. Ogawa S, Tank DW, Menon R, Ellermann JM, Kim SG, Merkle H, et al. Intrinsic signal changes accompanying sensory stimulation: functional brain mapping with magnetic resonance imaging. Proc Natl Acad Sci USA 1992;89:5951–5. Regan D. Evoked potentials and evoked magnetic fields in science and medicine. Human brain electrophysiology. New York: Elsevier; 1989. Rombouts SA, Lazeron RH, Scheltens P, Uitdehaag BM, Sprenger M, Valk J, et al. Visual activation patterns in patients with optic neuritis: an fMRI pilot study. Neurology 1998;50:1896–9. Russ MO, Cleff U, Lanfermann H, Schalnus R, Enzensberger W, Kleinschmidt A. Functional magnetic resonance imaging in acute unilateral optic neuritis. J Neuroimaging 2002;12:339–50. Sammer G, Blecker C, Gebhardt H, Kirsch P, Stark R, Vaitl D. Acquisition of typical EEG waveforms during fMRI: SSVEP, LRP, and frontal theta. Neuroimage 2005;24:1012–24. Schabus M, Dang-Vu TT, Albouy G, Balteau E, Boly M, Carrier J, et al. Hemodynamic cerebral correlates of sleep spindles during human non-rapid eye movement sleep. Proc Natl Acad Sci USA 2007;104:13164–9. Victor JD, Mast J. A new statistic for steady-state evoked potentials. Electroencephalogr Clin Neurophysiol 1991;78:378–88.