| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Letters to the Editor |
Department of Clinical and Experimental Epilepsy, Institute of Neurology, University College London, England*
MRI Unit, National Society for Epilepsy, Chesham Lane, Chalfont St Peter, Buckinghamshire SL9 0RJ, England
Afraim Salek-Haddadi, MD,*
Department of Clinical and Experimental Epilepsy, Institute of Neurology, University College London, England*
Karsten Krakow, MD
Klinik für Neurologie, Klinikum der Goethe-Universität, Frankfurt, Germany e-mail: l.lemieux@ion.ucl.ac.uk
Editor:
We read with interest the article by Dr Jäger and colleagues in the July 2002 issue of Radiology (1) and would like to raise a number of important issues in relation to their work.
The authors use electroencephalography (EEG)-triggered functional magnetic resonance (MR) imaging to assess signal intensity changes associated with interictal epileptiform discharges. It is well known that the MR imaging signal, when acquired in bursts, is (at least initially) dominated by T1 effects (2); however, a spike-related effect may be derived by subtracting the "resting" from "spike" time series (assuming constant delay between event and acquisition) (3). Instead, the authors describe an analysis method based on the correlation of a "paradigm timing" function with the images, after having discarded the first two volumes of each run. Although the paradigm timing function is not described in detail, we can infer that it is a step function with "on" periods corresponding to the postspike acquisitions. Furthermore, the acquisition parameters are unclear. The repetition time is given as 163,000 msec (presumably meaning 1.63 sec), but the actual repetition time varies between the resting and activation bursts of imaging (found by comparing figures 1 and 2). This could certainly lead to higher mean signal intensities on activation images compared with those on resting images (and perhaps even focal areas in conspiracy with head motion).
The most unsettling aspect of their findings concerns the magnitude of the spike-related signal intensity change (mean increases of up to 31%). This is in stark contrast to results in previous interictal EEG and/or functional MR imaging studies that have generally yielded blood oxygen level–dependent signal changes in the physiologic range of a few percent, indeed with most functional MR imaging findings at 1.5 T. By using a novel combined EEG and functional MR imaging technique with online pulse and image artifact suppression (4), we have recently studied the hemodynamic responses to interictal epileptiform discharges explicitly. Imaging continuously at a fixed repetition time reveals steady-state spike-related signal changes on the order of 1% (5,6). This has been corroborated in a similar report by Benar et al (7).
The authors cite a previously published postprocessing technique used to remove image artifact to yield "diagnostic quality EEG." However, a remnant artifact resembling focal delta waves is clearly visible on figure 1b during the intervals between image acquisitions.
The multiple comparisons problem is central in functional MR imaging, and an appropriate adjustment to statistical thresholds (eg, based on Bonferroni correction or Gaussian field theory) is an absolute requirement when searching across large volumes. The authors fail to address this explicitly, and the thresholding used remains unclear. The only anatomic figure fails to indicate the actual coverage (restricted to one 10 x 5-mm section). Moreover, where activations border areas of traditionally high susceptibility artifact, coregistration to echo-planar images is far more informative. We also note that the registration method used to project the activation maps onto the structural images is not described, making it impossible to judge the precision of the process.
It is unclear why rank tests were used to evaluate the correlations between EEG amplitude, signal intensity, and activation volume and also between mean signal intensities during two intervals. These appear to be continuous, rather than categoric, measurements. Where correlation is claimed between activated volume and mean spike amplitude, it would have been reassuring to see the data graphically. It is also unclear whether this relationship was observed for individual spikes or individual subjects. In general, interpretation is rendered difficult by an incomplete illustration of results. The lack of error bars in the plot of the mean spike and non–spike-related signal time courses are obvious. Given that the first 6 seconds of the response are discarded, it is impossible to judge whether the response really does peak at 6 seconds, as asserted by the authors.
The matrix of time series provided in figure 3c, which assumably represents a nonsequential arrangement of spike and resting images, is confusing (the lack of scales is frustrating). Why does there appear to be a slow and delayed increase in the activation level in relation to the reference function provided, especially given the data in figure 4?
Finally, in this series of five patients, the average number of spikes required to obtain significant activation was 17, with one case of 10 spikes. The authors indicate that this is in contrast to findings in two previously published studies in which spike-triggered functional MR imaging was used. They argue that the images obtained at rest could have been misclassified as a result of hidden spikes during acquisition and interruption of the EEG record (8). It should be stressed, however, that the proportion of any such misclassified images is likely to be small, since patients were selected with compatible spiking rates. Moreover, any unaccounted blood oxygen level–dependent changes would at worst be expected at only a fraction of the peak response. Furthermore, we have already shown that within this context, individual events can give rise to a detectable activation (9).
Given the lack of methodologic detail and other issues raised here, we remain unsure of the exact nature of the effects reported in this article.
REFERENCES
Institute of Clinical Radiology, Klinikum Grosshadern, University of Munich, Marchioninistrasse 15, 81366 Munich, Germany. e-mail: jaeger@ikra.med.uni-muenchen.de
We read with interest the letter to the editor by Dr Lemieux and colleagues in response to our article (1), and we would like to address the issues raised.
It is well known that the functional MR imaging signal in blood oxygen level–dependent sequences is dominated by T1 effects when acquired in bursts. To reduce this T1 effect on the evaluation of the functional MR imaging data, the first two volumes were discarded, as is common practice in functional MR imaging data evaluation (2,3) and as was documented in our study. Furthermore, the influence of T1 effects on the evaluation of the functional MR imaging data is reduced by choosing identical functional MR imaging acquisition techniques for spike-related and baseline data.
Both data sets were acquired nonperiodically and were interleaved as single events, not continuously over a longer time period. This method is described in the Materials and Methods section of our article. The functional MR imaging data evaluation was performed according to a well-established technique, the block design, which was also used for spike-triggered functional MR imaging by Krakow et al (4,5) and Dr Lemieux and colleagues (6). The paradigm timing of the block design is shown clearly in figure 3c (arrowhead). The repetition time was constantly 1.63 seconds throughout acquisition of all functional MR imaging data concerning patient evaluation. It should be noted that the given value of "163,000 msec" was a printing error and was not present in the submitted manuscript. To demonstrate the efficiency of our EEG recordings and postprocessing techniques, the repetition time of 2 seconds (figure 2) was chosen for one acquisition to facilitate the identification of a normal unaffected EEG between two volume acquisitions and to demonstrate that there was no saturation of the EEG amplifier, which was the case in articles published by Krakow et al (4,5).
As we stated in our article, a wide range of signal intensity changes due to spike activity at functional MR imaging is reported in the literature. We also found this in our subjects. We did not use any spatial Gaussian smoothing of our data, as was applied by Salek-Haddadi et al (7) at a full width at half maximum of 12 mm. This smoothing would generate an average signal intensity change in approximately 6 neighboring voxels, if a voxel edge of approximately 2.2 mm was used, as in our study. Because of this, strong signal intensity changes within a voxel may be hidden by the lack of signal intensity change of neighboring voxels. As was shown in our study (figure 3c), strong signal intensity changes with a high level of significance were present only in a discrete group of voxels; therefore, signal intensity changes within an activated voxel would be reduced by applying Gaussian smoothing of our data.
Furthermore, in all subjects in our study, a standard finger-tapping paradigm was performed to control our evaluation technique. Results of these control studies showed a signal intensity increase of approximately 3% in activated areas, which is consistent with the results published by others (8). We also showed the reproducibility of location and signal intensity changes within one subject. Since EEG recordings obtained with scalp electrodes have only limited depth, activity deep in the brain is not mapped with this technique. These spike activities are not detected with scalp EEG; however, they may influence perfusion changes, and thus, signal intensity increases in the activated voxels may differ between the subjects.
The remnant minor artifacts in the postprocessed EEG of figure 1b in our study are found in three lines of the applied 21 EEG channels (O2–G19, P8–G19, and PZ–G19). These are edge artifacts that are visible at the end of functional MR imaging data acquisition after postprocessing. Since they always have the same shape, however, they can be identified easily and do not hinder the EEG reading. To our knowledge, Dr Lemieux and colleagues have not published a complete EEG montage according to the international standard for electrode positions 10–20 (9,10), as it was recorded originally during functional MR imaging data acquisition and after postprocessing. However, they published EEG recordings during functional MR imaging data acquisition with amplifier saturation during imaging, which resulted in a total loss of EEG information (reference 4, figure 1a, 1b: positions Fp2–F8, F8–T4, T6–01, T3–T5, and T5–01; reference 5, figure 1: positions Fp2–F8, T4–T6, and T5–01). This hinders any further EEG processing. We described precautions to prevent this (11). According to standard practice, full montage of the 10–20 system is mandatory to enable exact spike location (9,10). So far, Dr Lemieux and colleagues have not followed this practice in any simultaneous spike-related functional MR imaging studies, to our knowledge. In addition, on the basis of the 10–20 system, 10 of the required 21 electrodes are missing in the EEG presented by Salek-Haddadi et al (positions FP1, F3, C3, P3, FP2, F4, C4, P4, Fz, and Pz) (7). Interestingly, in that case, all parietal and some frontal electrodes were missing, but the functional MR imaging activation maps showed activity in these areas, which raises the question of how exact spike localization and correlation with functional MR imaging was performed.
The thresholding (P < .001) of our statistical analysis was documented in the Materials and Methods section of our study. The evaluation of our functional MR imaging data, as well as the projection of the activated maps onto the structural images, was performed by using the Analysis of Functional NeuroImages software (afni.nimh.nih.gov/afni) as stated in our article. This software was used to calculate, voxel by voxel, cross-correlation maps between the time series data and a reference function that represents the time course of the spike activity (figure 3c). The cross-correlation coefficient represents the similarity of the measured signal-time course to the reference function. If the noise in the time series data is assumed to be Gaussian and uncorrelated, the correlation coefficient can be converted to an equivalent t statistic:
|
The Spearman rank correlation test was used to evaluate the correlations between EEG amplitude, signal intensity, and activation volume, because the sample size was small and the Spearman rank correlation test is robust if binormal distributions are not present. The accuracy and precision of the statistical evaluation of our data was confirmed in the separate statistical review of our study.
Graphic presentation of correlation between activated volume and spike amplitude is redundant, since the data are clearly visible in table 2. Furthermore, as was stated in the Results section and as was shown in table 2, the mean amplitude of the spikes of the individual subjects significantly correlates with the volume of activation. In the submitted manuscript, error bars were present in figure 4 but were deleted per the request of one reviewer. Figure 4 shows the signal intensity increase with a peak value of approximately 6 seconds, according to the applied technique. Signal intensity changes within the first 6 seconds could not be evaluated, however, and are not displayed. No conclusion was drawn regarding this time period.
The content of figure 3c seems to have been misunderstood. The figure shows the signal intensity (y axis) of 8 x 8 voxels over all images (x axis). On each x axis, a series of baseline data is followed by a series of data obtained after interictal spikes and is finally followed by a second series of baseline data. This time series corresponds to the reference function underlying the statistical evaluation. Variations in the reference function reflect the variation of signal intensity changes of different spikes in the subject as the time course of signal intensity changes of the spikes acquired sequentially in a patient study is displayed. The rise of signal intensity in figure 4 is in the range of the SD of our data. A statistically significant plateau of signal intensity remains. However, the cause of this plateau is unknown to the authors.
In conclusion, the technique of spike-related functional MR imaging that we described is a promising method with which to locate the anatomic source for focal spike activity, which is mandatory for a neurosurgical approach.
REFERENCES
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
| RADIOLOGY | RADIOGRAPHICS | RSNA JOURNALS ONLINE |
