Abstract

Pre-operative functional magnetic resonance imaging (fMRI), cortical evoked potentials (EPs) and intraoperative optical imaging of intrinsic signals (iOIS) were employed to relate the temporal–spatial characteristics of sensorimotor responses in human brain. Peripheral somasthetic stimulation (2 s) was provided either by a 110 Hz finger vibrator or transcutaneous median nerve stimulation in eight patients undergoing neurosurgical procedures. Each technique provided unique spatial patterns and temporal response profiles. EPs and iOIS activities were observed over the surface of pre- and post-central gyri (at the level of the superior genu) with very similar spatial distributions. In contrast, fMRI spatial distributions depended upon the model used for statistical correlation analysis. Using a monophasic response model, fMRI primarily localized within the central sulcus and did not demonstrate large signal changes over the pre- and post-central gyri (areas with iOIS/EP activity). However, as initial negative responses were incorporated into the response model, fMRI progressively localized closer to the iOIS and somatosensory EP maps. Temporally, responses to single stimuli differed between the fMRI and iOIS techniques. Using a monophasic model for fMRI analysis, the total fMRI response was delayed by 2–3 s relative to iOIS. As initial negative responses were incorporated in the analysis, the fMRI time course developed temporal characteristics similar to iOIS. Ultimately, when fMRI time courses were examined over pixels co-localizing with iOIS activation (without using statistical correlation analysis), the fMRI temporal profile included an initial decrease in signal (an initial dip) that closely resembled the time course of iOIS response. This is the first study to experimentally co-localize (temporally and spatially) iOIS and fMRI signals in human subjects. The spatial/temporal differences in this study likely reflect the capillary versus venous contributions of iOIS and fMRI, respectively. The temporal/spatial co-localization of the iOIS signal and the fMRI initial dip suggests the initial fMRI dip and the iOIS signal may result from similar physiologic events.

Introduction

Across modalities, perfusion-based measurements are complicated by multifaceted signals observed in the response profile. Of these signals, the initial and post-stimulus negative responses in functional magnetic resonance imaging (fMRI) have been the subject of numerous studies. However, the origin and existence of these response characteristics (the ‘initial dip’ and the ‘undershoot’) are still under debate. Functional spectroscopy first reported a small statistically significant decrease within 500 ms of activation (Ernst and Hennig, 1994). Later Menon et al. and Hu et al. (Menon et al., 1995; Hu et al., 1996) both observed robust dips in fMRI by selectively choosing pixels for this feature. However, Marota et al. (Marota et al., 1999) was unable to demonstrate an initial dip at either 2 T or 4.7 T, while other recent high-field-strength fMRI studies have also failed to identify the dip in cats (Jezzard et al., 1998) and in humans (Chen et al., 1998). Of the studies that have identified negative responses, several have reported the early response may be an artifact (McIntosh et al., 1996; Fransson et al., 1998), while others have suggested the initial dip may result from changes not related to a T2 blood oxygen level dependent (BOLD) effect (Hennig et al., 1995; Janz et al., 1997). In T2 BOLD fMRI the initial dip is commonly attributed to a localized increase in deoxyhemoglobin from increased oxygen metabolism. Optical intrinsic spectroscopy data (Malonek and Grinvald, 1996) are often cited for support; however, the linear modeling in this study has been recently challenged (Mayhew et al., 1999). Furthermore, Vanzetta and Grinvald (Vanzetta and Grinvald, 1999) have recently shown early increases in cortical oxygen metabolism by using oxygen tension studies and have correlated this to OIS (spectroscopic) findings. Despite these efforts, to date there have been few experimental correlations of optical imaging of intrinsic signals (OIS) and fMRI in animals (Hess et al., 2000) or humans (Pouratian et al., 2000). The establishment of the temporal and spatial co-localization of the OIS and fMRI response functions is a necessary step toward characterizing these signals.

The spatial congruence between neuronal and vascular responses is also poorly defined (Lou et al., 1987; Lindauer et al., 1993), primarily due to differing etiologies of mapping techniques and resolution differences. For example, the BOLD fMRI signal is thought to originate from venous capillary oximetry changes and therefore represents a venous blood- dependent indicator of brain activation (Cohen and Bookheimer, 1994). In contrast, O15 positron emission tomography studies often measure changes in blood flow to produce functional maps (Phelps and Mazziotta, 1985). Intraoperative OIS (iOIS), on the other hand, produces complex activation maps (Frostig et al., 1990; Narayan et al., 1995) with several underlying etiologies (metabolic and vascular). Spatial maps may therefore differ across techniques. Indeed, Rao et al. (Rao et al., 1993) indicated that fMRI often centers over a sulcus or includes it, while we have shown iOIS signal often centers over gyri (Toga et al., 1995; Cannestra et al., 1998a; Pouratian et al., 2000). Narayan et al. (Narayan et al., 1995) reported that OIS and intravascular fluorescent dye maps overspilled regions of electrophysiologic activity (using single unit recordings) by ~20%. Hess et al. (Hess et al., 2000) recently described similar spatial relationships between fMRI, 2-deoxyglucose (2DG) and OIS using block paradigms in rodents; however, few multi- modality functional imaging studies exist.

The relationship between the fMRI, OIS and electrical recordings in humans has yet to be established. The purpose of the following study is to define the temporal and spatial characteristics of activity-related somatosensory evoked potential (SSEP), fMRI and OIS signals in humans. In particular, our study was performed to establish the temporal–spatial relationship of the OIS and fMRI signals in humans, and correlate across modalities to examine the initial negative response. We hypothesized that the initial dip of fMRI is closely related to the OIS signal. We therefore expected a robust initial negative response when co-localized OIS response pixels were examined for fMRI response. Additionally, we hypothesized that there would be significant differences between the fMRI response obtained by correlation analysis using a monophasic response model and the iOIS and SSEP signals obtained intraoperatively. To address these temporal and spatial issues, we used a multimodality imaging approach. Sensorimotor cortex was studied during 110 Hz index finger vibration in eight human subjects undergoing neurosurgical procedures for removal of intracranial pathology [tumors or arterial–venous malformations (AVMs)]. Intraoperative optical imaging revealed rapid and transient signal changes by measuring changes in vascular and metabolic activity. Intraoperative cortical surface evoked potentials (SSEPs) were used to monitor electrical activity. Preoperative BOLD fMRI (at 3 T) was used with the same experimental parameters as iOIS to interrelate findings to other non-invasive mapping studies. Postoperatively, all three modalities were co-registered in three-dimensional space using warping algorithms. Spatial and temporal statistics were then calculated and compared across modalities.

Materials and Methods

We measured functional magnetic resonance, evoked potential and optical reflectance changes over sensorimotor cortex in eight patients undergoing surgical resection of fronto-parietal masses (tumors or AVMs). Informed consent was obtained from all patients prior to pre-operative imaging. fMRI was performed pre-operatively 24–72 h before surgery. During surgery, the patient's head was fixed, via a Mayfield apparatus, to the operating table. After site localization (via a frameless stereotaxic system), the craniotomy and reflection of the dura was performed. SSEPs and iOIS recordings were then obtained. Post-operatively, all images were co-registered in three-dimensional space using three-dimensional warping algorithms. Volumes were also registered into the Talairach atlas (Talairach and Tournoux 1988), for stereotaxic coordinates.

Human iOIS

All iOIS recordings were performed with the subject under general anesthesia. The human intraoperative optical imaging system has been described in previous publications (Toga et al., 1995; Cannestra et al., 1996, 1998a, b, 2000; Pouratian et al., 2000). Briefly, we utilized a slowscan charge coupled device (CCD) camera (TE/CCD-576EFT, Princeton Instruments, Trenton, NJ) mounted via a custom adapter onto the video monitor port of a Zeiss operating microscope. Images were acquired through a transmission filter at 610 (600–620) nm (Corion Corp., Holliston, MA). Circular polarizer and heat filters were placed under the main objective of the operating microscope to reduce glare artifacts from the cortical surface. White light illumination was provided by the Zeiss operating microscope light source, through a fiber optic illuminator. Since the microscope view was determined by the surgeon's choice of magnifications and microscope placement, the experimental field of view (FOV) varied, depending on lens distance from cortex and specific orientation. It varied from 2.75 to 7.50 cm, resulting in spatial resolution of 110–425 μm.

iOIS acquisitions consisted of CCD exposure of the cortex during a control state (no stimulation) followed by a subsequent stimulated state (see below). Control trials and experimental trials were interleaved. During stimulation trials, baseline CCD images were acquired, then physiologically synchronized PC software triggered the stimulations (2 s duration). Experiments contained 10–20 stimulation trials and included an equal number of non-stimulated interleaved controls. iOIS frame rate was 250 ms.

Human iOIS Image Alignment

Image alignment is critical to measure optical changes within the operative setting. Optical changes are exceedingly small, and may be attenuated by even the slightest motion artifact. Image alignment was achieved with synchronization to electrocardiographic and pneumographic waveforms (Toga et al., 1995), and by utilizing post-acquisition automated image registration (AIR) algorithms (Cannestra et al., 1998a; Woods, 1998).

When possible, data acquisition was synchronized with monitored electrocardiographic and pneumographic waveforms to minimize cortical movement artifacts. Data acquisition at similar time points during the respiration cycle minimized motion artifact caused by periodic movement of the brain (Toga et al., 1995; Cannestra et al., 1996, 1998b). Synchronized image acquisition only occurred during the expiration portion of the respiratory cycle. The pneumographic waveform triggered the beginning of the imaging cycle, after which acquisition was controlled from synchronization to the cardiac cycle (250–500 ms post-R-wave). Experimental trials always began at the same point in time during the respiration cycle and after each R wave (as monitored by ECG). Control trials and experimental trials are interleaved during sequential expirations utilizing the same synchronization procedure. Control and experimental images are therefore matched in time from the beginning of the expiratory phase, i.e. each experimental image has a separate control image. Since data acquisition occurred at similar time points during every respiration cycle, all images were collected with the brain in a similar position (± 0.5 of a heart beat), minimizing the effect of periodic motion of the brain.

Alignment and warping algorithms were used to superimpose sequential acquired images prior to image averaging and analysis. Images were co-registered post-operatively utilizing non-linear AIR algorithms implemented on a UNIX workstation. This technique has proven successful in removing motion artifact and subsequent noise in previous publications (Cannestra et al., 1998a, 2000; Woods, 1998; Pouratian et al., 2000).

Human Electrophysiology

Quantitative evoked potentials (EPs) were recorded to help target the area for imaging and for correlation with the resulting optical signals. All EP recordings were performed with the subject under general anesthesia. EPs were obtained from a 20 electrode cortical surface recording (4 × 5 electrode grid, 1.0 cm inter-electrode spacing). Topographical color maps represent voltages obtained ~20 ms after stimulation (Figs 1 and 2). Motor cortex was identified by a maximal amplitude negative peak. Somatosensory cortex was identified by phase reversal of the negative peak across the central sulcus where the negative end of the dipole was on the somatosensory cortex (Nuwer et al., 1992). SSEPs were performed by the UCLA clinical EEG team under the supervision of a neurologist, as part of the surgical procedure. They were performed by transcutaneous median nerve stimulation using a standard clinical protocol (Nuwer et al., 1992). The resulting EP maps were superimposed over images of the exposed cortex. Before removal of the SSEP electrode grid from the cortical surface, an image was acquired to register SSEP and iOIS maps. Electrode dimensions and orientations were used to determine scaling and estimate cortical curvature. These results agree with previous studies relating iOIS with SSEP and ESM measures over somatosensory cortex (Toga et al., 1995; Cannestra et al., 1998a).

Human fMRI

The most common fMRI technique exploits changes in magnetic resonance signal intensity with blood oxygenation. This intrinsic contrast method has been termed blood oxygen level dependent (BOLD) contrast (Ogawa et al., 1990) and may be used to dynamically measure brain activity non-invasively (Belliveau et al., 1991, 1992; Frahm et al., 1992; Kwong et al., 1992; Bandettini et al., 1993, 1997). [For a review see Sorensen et al. (Sorensen et al., 1997).] BOLD fMRI data are obtained routinely for pre-operative clinical assessment of UCLA Comprehensive Brain Tumor program patients. All fMRI recordings were performed with the subject awake. Preoperative fMRI was performed on a 3 T GE scanner in patients undergoing surgical resection for parietal tumors. For the fMRI activation studies, multi-slice echo planar imaging (EPI; modifications by Advanced NMR Systems) was used with a gradient echo spin sequence (TR = 2000 ms, TE = 45 ms, Flip = 79, Nex = 1.0, FOV 64 × 64), yielding an in-plane resolution of 3.1 mm and slice thickness of 4 mm (1 mm interslice gap). Fifteen axial slices were acquired through the brain, covering from the most superior cortex through the Sylvian fissure. A total of 152 images were obtained over a 5 min 12 s impulse response paradigm. Each experimental trial began and ended with 20 s of rest to establish and assess baseline MR signal. Stimulations were triggered every 16 s for a total of 16 stimulations per experimental trial. This period allowed for a complete return to baseline of the fMRI signal.Two to three trials were performed in each patient.

Since the fMRI scans were 5 min in duration, there was the potential for patient movement. During fMRI scans, the subject's head was stabilized posteriorly by a deformable cushion and anteriorly by adhesive tape. Residual movement artifacts were corrected post-acquisition using automatic image registration algorithms (Woods et al., 1998a,b).

Stimulation

Finger tip vibratory stimulations (index finger) were administered using a custom-built 110 Hz pneumatically driven vibrator (MRI compatible). When placed at the finger tip, the amplitude of the device was 0.5–1 mm. Subjects reported vibratory sensation but did not report stimulus related finger movement. All stimulations were 2 s in duration. Interstimulus intervals were 16–30 s in duration. In the fMRI scanner, the pneumatic vibrator was synchronized and phase lagged to image acquisition. Each fMRI stimulation was lagged 500 ms in successive trials (16 trials) to obtain average data points every 500 ms post-stimulation (at a TR of 2 s, each 16 trial set included four trials acquired at identical 500 ms time points). During OIS stimulation trials, baseline CCD images were acquired, then physiologically synchronized PC software triggered the stimulations. SSEPs consisted of transcutaneous electrical stimulation (12.5–17.5 mA) of the median nerve. These parameters were used as part of the clinical SSEP protocol.

Three-dimensional Image Registration

After acquisition, brain mapping data were co-registered in three- dimensional space using an in-house software package implemented on an Silicon Graphics UNIX workstation. Pre-operative T1, fast spin echo and functional MRI data sets were reconstructed in three dimensions using a software package developed by the Montreal Neurologic Institute (Display, MNI). Co-registration of the MRI volumes were then perfomed by using the AIR warping algorithms described above. Cortical surface models were then extracted from the T1 volume by exploiting voxel intensity changes at the cerebrospinal fluid–gray matter interface [soft- ware developed by MacDonald (MacDonald, 1998)]. OIS and SSEP data were co-registered with the resultant volumetric/surface model by first reconstructing the two-dimensional images into three-dimensional space and applying the AIR warping algorithms to warp corresponding sulcal and gyral landmarks (Pouratian et al., 2000). Once complete, this multimodality data set was then placed in Talairach space (Talairach and Tournoux, 1988) by registering landmarks with the International Consortium for Brain Mapping brain atlas. All localization statistics were therefore produced in Talairach coordinates. Other statistics (size and intensity) were calculated in original patient image space.

iOIS Analysis

Optical reflectance images were analyzed by pixel-by-pixel subtraction of a control trial and a stimulation trial. This subtracted image was then divided by the control image to normalize for differences between subjects and trials (Narayan et al., 1994). These ratios represent proportional changes from baseline. Ratio images were averaged across trials at each time epoch to increase the signal-to-noise ratio (Janesick et al., 1987). No other digital image processing was performed.

Signal magnitude was defined as the average pixel intensity in a statistically defined region of interest (ROI). ROIs were first created from the averaged ratio images at each time point by performing a three-pixel Gaussian blur, equalizing and thresholding at the mean pixel value + 1 STD. This ROI was then superimposed upon the original averaged ratio image and the magnitude calculated (Blood et al., 1995; Cannestra et al.,1996). The time point with the largest calculated magnitude was considered to be the maximal optical response for the trial. This ‘maximal’ ROI was then superimposed upon the other images of the trial to calculate the respective magnitude at each time point, thereby yielding the corresponding OIS time course. The ROI was then applied to provide the iOIS/fMRI co-localized pixels and subsequently the iOIS/fMRI colocalized time course (see below).

fMRI Analysis

The fMRI images were reconstructed, analyzed and displayed as original slice data (except for localization center of mass calculations). To analyze the data, paradigms were convolved with either (i) fMRI Monophasic: a model of hemodynamic response function (Savoy et al., 1994; Cohen, 1997); (ii) fMRI Small Dip: a model that includes a small initial negative response observed by previous investigators (Menon et al., 1995); or (iii) fMRI Correlation: a model empirically obtained from co-localized fMRI and iOIS pixels (see below). The fMRI Monophasic model assumes a delayed onset of blood flow (seconds) followed by a slow rise and fall beginning 2 s after the activation task. Specifically, we used an impulse response model based on the temporal response of single visual flashes in V1 by Savoy et al. (Savoy et al., 1994). This model assumes a hemo- dynamic delay of 2 s, a peak response of 5 s and a fall to baseline that is complete by 11 s. Note that neither an initial signal decrease nor a post-stimulus decrease is modeled in this equation; however, the data from Savoy et al. (Savoy et al., 1994) demonstrate negative dips at 1.5 T with this model. The second model (fMRI Small Dip) assumes an initial signal decrease at 2.0 s and subsequent inversion to a peak positive response at 8.0 s. This model was obtained by emperic data from Menon et al. (Menon et al., 1995) at 4 T. The third model (iOIS/fMRI Correlation) was obtained from the time course obtained by measuring the activity of the co-localized iOIS and fMRI pixels (see below; Table 2). fMRI analysis and statistical comparisons were restricted to the subsequent craniotomy FOV. The magnetic resonance signal intensity was then correlated (Pearson's) at every pixel (within the FOV) with the resulting input– response function (the convolved boxcar function of task versus rest, as above). All pixels exceeding a correlation coefficient of 0.3 were considered significant (corresponding to a P < 0.01). Since the analysis was restricted to the FOV, no statistical corrections for multiple comparisons were needed. In this study, no cluster threshold was applied; however, the smallest cluster was 34 pixels (1225 mm3). Overall the clusters were quite large in comparison to the FOV (~2000 mm2), leaving a low probability for false positives. The time series for every pixel was then calculated and averaged across each activation cycle. To confirm that the origin of the signal was not due to movement artifact (such as head motion or drift), fMRI data were motion corrected using an AIR algorithm (Woods et al., 1998a,b). Resultant pixels were then registered to the EPI co-planar high-resolution images using the AIR algorithms to provide accurate co-localization.

We calculated the percent signal change (of each pixel) from the averaged signal intensity across the time series, and then calculated the mean percent change within each cluster of pixels with a correlation coefficient of 0.3 or greater. This magnitude was then normalized and averaged across subjects. The ‘fMRI monophasic’ magnitude (Fig. 3) was defined as the average change in intensity within the statistical correlation analysis ROI using the model based on Savoy et al. (Savoy et al., 1994). Similarly, the ‘fMRI small dip’ magnitude (Fig. 3) was defined as the average intensity change within the statistical correlation analysis ROI using the model based on Menon et al. (Menon et al., 1995). In contrast, the ‘iOIS/fMRI correlation’ magnitude (Fig. 3) was defined as the average intensity change within the statistical correlation analysis ROI using a model based on the time course data from this study (see below).

The ‘iOIS/fMRI co-localized’ magnitude (Fig. 3) was defined as the average change in intensity within the ROI as defined by the iOIS ROI (as noted above). The ‘iOIS/fMRI co-localized’ time course therefore did not utilize the convolution noted above, but rather was calculated from the fMRI and iOIS co-localized pixels. Once this time course was empirically obtained, we modeled and applied this response to provide a hemo- dynamic model for an additional correlation analysis (noted above). Due to similarities with oscillating systems, we modeled the response profile R(t) of Figure 3 as a damped oscillating circuit responding to a step input (Table 2). The model produces a response profile which accurately reflects the initial dip, peak and decay of the fMRI cortex signal in Figure 3. This model also provides for a post-stimulus undershoot that may be represented in Figure 3, consistent with recent fMRI single trial studies (Menon et al., 1995; Bandettini et al., 1997). In comparison to the literature, this model differs mainly in the amplitude of the initial dip.

Center of Mass and Localization.

Center of mass and principal axis were also calculated for functional maps. The center of mass was computed independent from the magnitude and ROI calculations utilizing intensity and pixel location:  

\[\mathit{C}(\mathit{x,y}):\ \mathit{C}(\mathit{x})\ =\ \frac{{{\sum}_{\mathit{i,j}}}\mathit{ig_{ij}}}{{{\sum}_{\mathit{i,j}}}\mathit{g_{ij}}}\ \mathit{C}(\mathit{y})\ =\ \frac{{{\sum}_{\mathit{i,j}}}\mathit{jg_{ij}}}{{{\sum}_{\mathit{i,j}}}\mathit{g_{ij}}}\]
Where gij is the gray scale value of the image pixel at the location (i,j). Similar calculations are commonly used in functional imaging and image co-registration (Toga and Banerjee, 1993; Cannestra et al., 1998b). The above calculations were performed for both the iOIS and fMRI data sets after co-registration and placement into Talairach space (as above).

Results

Spatial and temporal statistics were compiled from iOIS, fMRI and SSEPs maps in eight human subjects undergoing neuro- surgical procedures. Overall, fMRI, iOIS and SSEP activities were observed with very similar spatial distributions. However, different spatial patterns and temporal response profiles were observed depending upon analysis. Spatially, when the fMRI signal was obtained from a monophasic response model, the fMRI and iOIS respective centers of mass (COM) were separated by statistically significant distances. When an initial dip time course was applied, the respective COMs were closer. Ultimately, when the iOIS/fMRI co-localization model was used to calculate the fMRI signal, the COM's were nearly co-localized. Temporally, when the monophasic response model was used to obtain the fMRI signal (using statistical correlation analysis) the time course was delayed by 2–3 s relative to the iOIS. However, the temporal profiles of these two modalities became more similar when time courses including an initial dip were applied. Ultimately, when only co-localizing (iOIS and fMRI) pixels were examined, the resultant fMRI time course was biphasic, with an initial dip very similar to the iOIS time course.

Spatial Localizations

All modalities (fMRI, iOIS, SSEPs) showed responses near the level of the superior genu of the central sulcus irrespective of analyses (Figs 1 and 2). SSEPs and iOIS activity was observed over the surface of pre- and post-central gyri with very similar spatial distributions. The iOIS map co-localized with the SSEP map, corresponding to a 13 mV potential change (within the 0.5 cm interpolation distance required for SSEPs). Figures 1 and 2 detail the co-localization of iOIS (middle panels) and SSEP (left panels) maps near the superior genu of the central sulcus. The co-localization of iOIS and SSEP is detailed in the magnified view of the central sulcus (Fig. 1; lower left panel). iOIS responses straddled across the central sulcus and included larger responses emanating from the exposed crest of the gyri. The overall activation roughly corresponded to the anatomic pattern of the superior genu. The SSEP dipole also localized across the superior genu (Nuwer et al., 1992), with large areas of negativity and positively corresponding to the same gyral crests observed with iOIS. This co-localization pattern is consistent with close SSEP/iOIS spatial coupling observed in previous experiments (Toga et al., 1995; Cannestra et al., 1998a).

The localization of the BOLD fMRI signal varied dependent upon analysis. Using statistical correlation analysis and a monophasic hemodynamic model (based on Savoy et al., 1994), the fMRI signal primarily measured responses within the central sulcus, deep and superior to the activations observed by iOIS and SSEPs. Figure 1 shows the spatial involvement of fMRI signal modeled as a surface under a translucent cortex (upper right). The fMRI signal is observed extending deep within and extending superior along the central (and post central) sulcus relative to the iOIS (upper middle) and SSEP surface maps (upper left). Figure 1 (lower right panel) details the relationship between iOIS and fMRI in a magnified view of the central sulcus. The largest signals were observed at the superior genu (within the central sulcus) and tracked ~1 cm superior above the level of the superior genu. Averaged across all subjects, this fMRI COM was ~14.9 mm deep and 6.6 mm superior to the OIS COM (statistically significant differences, P < 0.005 and P < 0.035, respectively; Table 1).

However, when initial dip time courses were applied for statistical correlation analysis, the resultant fMRI signals encompassed less area and localized closer to the iOIS signal (Table 1). Using a time course providing a small initial dip [based on Menon et al. (Menon et al., 1995)] yielded a decrease in space (4%), and a closer localization (0.85, 1.44 and 1.16 mm in X, Y and Z, respectively) to iOIS. When the biphasic time course from examining iOIS/fMRI co-localized pixels (see below) was applied, the spatial involvement of fMRI decreased by 35%. The fMRI signal and the iOIS COM also were much closer (2.35, 3.01, 4.52mm in X, Y and Z, respectively), essentially co-localized in the Z plane. Figure 2 (different subject from Fig. 1) illustrates a magnified view of the relationship of the fMRI statistical T maps (right) using the biphasic time course with the iOIS (middle) and SSEP (left) maps. In Figure 2, fMRI responses are much smaller than observed in Figure 1 and are localized closer to the iOIS and SSEP maps. Although the fMRI signal still includes large responses within the sulci, significant statistical responses are observed on the surface of the cortex corresponding to the iOIS and SSEP ROIs. Table 1 gives the spatial statistics averaged across all subjects. These spatial relationships were consistent across all subjects.

Temporal Characteristics

Temporal patterns were of similar shape and duration (Fig. 3; averaged across all eight subjects); however, the fMRI time course was dependent upon analysis method. iOIS responded within 500 ms (16.8 ± 5.5% change), peaked at 2.0 s and returned to baseline at ~6 s post-stimulation. Using the monophasic model for correlation analysis (Savoy et al., 1994), the fMRI response was delayed by 2–3 s relative to the iOIS. The signal occurred within 1.5 s post-stimulation onset, peaked at ~6 s and returned to baseline by 10.0 s post-stimulation. However, when the fMRI analysis was performed using a hemodynamic model that included a small initial dip [similar to Menon et al. (Menon et al., 1995)] the resultant time course began with a small initial negative response with temporal characteristics similar to iOIS before being overshadowed by the later positive signal.

To investigate the relationship further, the fMRI response was calculated over only the pixels that co-localized with the optical signal. When this portion of the fMRI signal (iOIS/fMRI co-localized) was measured alone, a large additional early response was observed. This response had a very similar time course to the observed iOIS signal., appearing as early as 500 ms and peaking at ~1.5 s. After 1.5 s, the time course changed phase and became overshadowed by the later positive fMRI signal (crossing baseline at ~2.75 s). The time course then reached a positive peak at ~6 s and returned to baseline by 10 s, similar to the ‘monophasic fMRI’ profile described above. The pre- and post-stimulus baseline level was 3.0 ± 3.0%. Initial dip time points at 500 ms, 1.5 s and 2.0 s were statistically significant (P < 0.001) from this baseline. As an additional measure, the time course obtained from this iOIS/fMRI co-localization was then modeled (Table 2) and applied to provide an additional correlation analysis for the fMRI data (iOIS/fMRI correlation, also is provided in Fig. 3). The measured time course (and corresponding spatial statistics in Table 1) was more similar to the iOIS and iOIS/ fMRI co-localized time course. This ‘iOIS/fMRI correlation’ time course had an initial dip amplitude larger than the time course obtained by the small initial dip model (Menon et al., 1995) but smaller than the ‘iOIS/fMRI co-localized’ time course. Time courses depicted in Figure 3 were normalized (0–100% change) between modality and analysis method.

Discussion

In this study we measured fMRI, optical and evoked potential changes in patients undergoing intracranial neurosurgery. We found good correlation between SSEPs and iOIS. fMRI localized near the SSEP/iOIS responses; however, spatial and temporal statistics varied dependent upon analysis. Analysis using a monophasic model revealed fMRI signals deep and superior (to iOIS and SSEPs), with statistically significant COM differences; however, as initial negative responses were applied for analysis these differences decreased. Ultimately, when correlating to an iOIS/fMRI co-localized time course, fMRI closely localized with iOIS responses. Temporal differences were also observed between the iOIS and fMRI responses, dependent upon analyses. As initial negative responses were applied for fMRI analysis, the fMRI signal began to resemble the iOIS time course. When responses over co-localizing pixels were measured, fMRI revealed an earlier response with temporal characteristics very similar to iOIS.

Spatial Localizations

Similar to previous studies, we found very good correlation between SSEP and iOIS maps regardless of the differing modality and activation (Toga et al., 1995; Cannestra et al., 1998a). SSEP maps result from activation in the areas 3a and 3b within the central sulcus (Nuwer et al., 1992). In contrast, the iOIS signal is recorded from the exposed cortical/gyral surface commonly corresponding to area 1 of human somatosensory cortex (Geyer et al., 1999). Therefore the anatomic relationship between areas 1, 3a and 3b, and the co-localization of SSEP and iOIS signals demonstrate the close relationship between iOIS signals and underlying neuronal activity (Frostig et al., 1990; Malonek and Grinvald, 1996).

The relationship between fMRI and the iOIS/SSEP co-localization was dependent upon the method of analysis. These spatial differences may be explained from the presumed origin of BOLD fMRI. These signals are thought to originate from venous capillary oximetry changes and therefore give a venous blood-dependent indicator of brain activation. However, the BOLD signal is not necessarily restricted to small vessels around activated tissue, but may also occur in downstream vessels (Frahm et al., 1992; Lai et al., 1993; Cohen and Bookheimer, 1994; Gao et al., 1996). This may result in signal considerably distant from activated cortex (Lai et al., 1993). In our study correlation analysis using the monophasic hemodynamic model yielded the largest contributions deep and superior to the iOIS/SSEP map. Since the venous vasculature in the sensorimotor region lies within sulci and drains superiorly toward the superior saggittal sinus, these results can be explained by significant large venous contributions to the fMRI (all) signal (likely emanating from the Rolandic vein). As increasing initial negative responses were used for correlation analysis, however, the fMRI spatial statistics were closer to the iOIS/SSEP COM (fMRI small dip and iOIS/fMRI correlation, Table 1). This progressive co-localization provides evidence that the fMRI initial negative response is closely related to negative 610 nm iOIS signals. Indeed, reports from the literature indicate 610 nm iOIS is exquisitely sensitive to changes in oxyhemoglobin (Frostig et al., 1990; Malonek and Grinvald, 1996) and blood volume (Narayan et al., 1995). Furthermore, recent studies suggest the initial negative signal may relate to initial increased cortical oxidative metabolism [as observed from oxygen detection schemes (Vanzetta and Grinvald, 1999)]. Our study therefore demonstrates the close relationship between neuronal activity (SSEP), oxidative metabolism (iOIS) and the fMRI initial negative response.

Temporal Response

The iOIS signals in this study are consistent with our previous studies in humans (Toga et al., 1995; Cannestra et al., 1996, 1998a, b) using somatosensory stimulation. These time courses also correlate with NIRS (Villringer et al., 1993), video- microscopy (Ngai et al., 1988), vascular fluorescent dyes (Narayan et al., 1995), optical spectroscopy (Mayhew et al., 1999) and vascular bead studies (Cox et al., 1993). In addition, the temporal characteristics of our fMRI signal (both the later signals and the initial dip) are very similar to the response profiles described previously (Menon et al., 1995; Hu et al., 1996; Yacoub et al., 1999).

The different time courses we obtained from statistical correlation analysis, fMRI and iOIS also likely represent differing components of activity related perfusion response. The observed differences may result from the venous origin of the BOLD fMRI signal and temporal/spatial differences between the two techniques. If we assume the origin of iOIS is more arterial [from intravascular fluorescent dye studies (Narayan et al., 1995)] and the monophasic BOLD fMRI is more venous, then the delays observed between the iOIS and monophasic fMRI (fMRI all, Fig. 2) peak response may be due to signals emanating from mostly preversus post-capillary exchange. Interestingly, this delay is very similar to the arterial–venous transit times (~2 s) reported by Cox et al. (Cox et al., 1993). As initial negative responses were used to correlate to and calculate fMRI spatial statistics, the resultant time courses also included larger amplitudes of the initial dip. Spatially, as the initial dip became more prominent, the fMRI signals become more co-localized with the iOIS signal (fMRI small dip and iOIS/fMRI correlation, Table 1). Ultimately, the appearance of the large initial dip fMRI response when examining commonly activated pixels (iOIS/ fMRI co-localized; Fig. 3) independent of correlation analysis suggests this specific response component may produce functional maps that relate closer to electrical and optical activity. If distinguished from the later macro-venous signal, this minute signal may be exploited as an additional mapping technique. Indeed, a recent report (Kim et al., 1999) exploited the fMRI initial negative response to map iso-orientation domains of cat visual cortex.

Although the fMRI temporal response in our study (appeared within 500 ms, peaked at 1.5 s, turned positive and peaked at ~6 s) is very similar to previous reports (Ernst and Hennig, 1994; Menon et al., 1995; Yacoub et al., 1999), the amplitude of our initial dip is different and varied dependent on analysis. Previous studies have shown initial dip magnitude to vary depending on analysis technique and modality. Ernst and Hennig (Ernst and Hennig, 1994) showed a small magnitude signal (<0.2%) with MR spectroscopy. Menon et al. (Menon et al., 1995) showed a much larger signal by selectively choosing pixels with the initial dip feature (~16% of the following positive response). Fransson et al. (Fransson et al., 1998) showed negative responses of ~33% of the subsequent positive response. Yacoub et al. (Yacoub et al., 1999) showed the initial dip to be ~25% of the later positive response. In our study, the initial dip was 19, 61 and 80% magnitude of the corresponding positive response depending on the analysis technique. When increasing initial dips were applied to the hemodynamic response function, increasing spatial co-localizations also were obtained. The largest magnitude initial negative response (80%) was obtained by using iOIS as an objective ROI measure for fMRI analysis (without statistical correlation). Indeed, if the initial dip and the OIS signal share a common origin, then large magnitude negative responses should be observed in the co-localizing pixels.

The exact origin of the initial dip is still under debate. Evidence from optical intrinsic spectroscopic data based on linear modeling suggest the initial dip is due to localized increases in deoxyhemoglobin from increased oxygen metabolism (Malonek and Grinvald, 1996). However, this linear modeling has been recently challenged (Mayhew et al., 1999). Further modeling by Mayhew et al. (Mayhew et al., 1999) applied additional hemodynamic oscillations (e.g. the V-signal) to produce the initial dip in the oxy- and deoxyhemoglobin optical signals. However, we did not observe any oscillations in our study, nor have we in previous human reports (Toga et al., 1995; Cannestra et al., 1996, 1998a, b). Additionally, MRS studies (Hennig et al., 1995) have also suggested that the initial dip may result from T1 and proton density changes (and not to a BOLD effect), and may relate to the scattering changes in OIS (Janz et al., 1997). Oxygen tension studies, however, have shown an increase in cortical oxidative metabolism post-stimulation (Vanzetta and Grinvald, 1999). Our data support the hypothesis that the fMRI changes responsible for the initial dip are closely related to the processes behind the negative 610 nm iOIS. Furthermore, our study shows initial negative response co-localizes (spatially and temporally) to resultant iOIS maps and underlying EP activity as obtained by SSEPs.

Potential Confounds

The spatial differences may be due, in part, to the differences in experimental conditions during performance of these three modalities. Index finger vibration was applied in iOIS and fMRI mapping, while transcutaneous median nerve electrical stimulation was used for SSEPs. Additionally, the iOIS and SSEP mapping were performed with the subject anesthetized, while the fMRI was performed with the subject awake. Although these potential confounds could contribute to the intermodality differences observed here, the SSEP and iOIS maps had very similar mapping across stimulation modality [consistent with the index finger contributing large iOIS responses within the median nerve iOIS response (Cannestra et al., 1998a)]. Additionally, the differences between SSEP/iOIS response and the fMRI are unlikely to be due to the anesthesia since fMRI maps became progressively closer (temporally and spatially) to the SSEP/iOIS maps with changes in statistical correlation models.

The spatial differences may also be due, in part, to the volumetric versus surface acquisition of fMRI and iOIS/SSEPs, respectively (as noted above). Although the optical imaging system was placed over the sulcal region (roughly 2.5 × 2.5 cm FOV with ~100 μm spatial resolution), iOIS failed to detect signal changes within the central sulcus. Due to limited light penetration into sulci, the OIS technique is less sensitive to deeper cortex relative to surface gray matter changes. In contrast, the volumetric acquisition of fMRI allows the detection of the response within the sulci. This difference may contribute to the spatial differences, but is unlikely to result in the superior localization of the fMRI map relative to the iOIS and SSEP maps. It is also unlikely to provide the progressive temporal and spatial co-localization of the fMRI and iOIS signal with the applied analyses.

Although correlation analyses may bias reports by self-selecting pixels with a particular feature of interest, we performed five different calculations to provide the temporal and spatial data observed in Figure 3 and Table 1, respectively. These results revealed trends in both the spatial and temporal domains, independent of analyses. Spatially, the significant fMRI signal localized closer to the iOIS and SSEP COMs when increasing negative amplitude dips were applied for correlation analysis. Temporally, the initial negative dip magnitude increased as the COMs became closer. When co-localized pixels were examined (independent of correlation analysis), the fMRI time course had a large initial dip similar in magnitude to the iOIS signal. These trends are therefore unlikely to solely represent bias from correlation analyses.

The results of this study are critically dependent upon the proper co-registration of brain maps. Pre-operatively, the fMRI and anatomic data sets were obtained in immediate succession. Interscan (and intraparadigm) movements were corrected by the use of the AIR automated image registration algorithm (Woods et al., 1998a,b). All co-localizations (fMRI to T1; iOIS and SSEP to T1 cortical surface model) were obtained by exploiting the AIR registration and warping tools. This technique has proven successful in removing motion artifact and subsequent noise in human and animal imaging protocols (Cannestra et al., 1998a, 2000; Woods, 1998). The mathematics employed have been validated and the accuracy determined using phantoms and invertability tests (Woods et al., 1998a,b). However, slight misregistration could contribute to some of the discrepancies observed in this study.

Interscan variability and reproducibility could also contribute to the differences observed in this study. The interscan (intrasubject) variabilities for iOIS were 8, 10 and 6% for space, intensity and COM, respectively. For fMRI, intrasubject variabilities were 6, 12 and 1% for space, intensity and COM, respectively. SSEPs were highly stable and varied by <5% for space and intensity. Dipole location was always maintained between the same two electrodes. Although these intrasubject variabilities may contribute to the results of this study, the differences are small relative to the intersubject variability in Table 1. However, to minimize these effects, we averaged each imaging modality over 16, 10–20 and 250 acquisitions for fMRI, iOIS and SSEPs, respectively.

Conclusions

In this study, pre-operative BOLD fMRI, intraoperative SSEPs and iOIS were employed to relate temporal–spatial characteristics of sensorimotor response in human brain. Different spatial patterns and temporal response profiles were observed depending on analysis. Spatially, the closest localizations was obtained by using a statistical correlation analysis with a model that included a large initial negative response (obtained empirically from examining fMRI and iOIS co-localizing pixels). Temporally, the most similar temporal profiles were obtained when fMRI response was examined over pixels co-localizing with iOIS activation. These findings may indicate the iOIS and early BOLD fMRI signals (i.e. the initial dip) share common etiologies. This study also demontrates that later positive BOLD fMRI signals may include venous signals not present or possibly ‘downstream’ from iOIS maps.

Notes

The authors would like to thank Alan Evans and the Montreal Neurologic Institute for the use of software. We would also thank the surgical team for their willingness and cooperation. A.F.C. and N.P. were supported in part by the Training Program in Neuroimaging (MH19950). N.P. also is supported by an NIH National Research Service Award (MH12773-01). Additional support provided by research grants to A.W.T. (NIMH MH52083 and NCRR RR13642).

Address correspondence to Arthur W. Toga, Laboratory of Neuro Imaging, Department of Neurology, UCLA School of Medicine, 710 Westwood Plaza, Los Angeles, CA 90095-1769, USA. Email: toga@ loni.ucla.edu.

Table 1

iOIS, fMRI and SSEP statistics

 SSEP iOIS fMRI Monophasic fMRI Small Dip fMRI/iOIS Correlation 
After co-registration in three dimensions, activation size (region of interest), intensity and COM were calculated for each modality (SSEP, iOIS and fMRI) and fMRI analysis method (monophasic correlation, small dip correlation and iOIS/fMRI correlation). Statistics were averaged across all eight patients. Size and intensity were calculated in native patient space. COM was calculated in the Talairach coordinate system. Intensity is listed as decrease in reflectance (DIR) for iOIS and contrast units (CU) for fMRI. An intensity threshold of ±13 mV was defined for the SSEP maps. The intensity change of fMRI was 12 and 13% of the overall MR signal. 
*Statistical significance between iOIS and the fMRI Monophasic analysis. Note that the iOIS and SSEP sizes are two-dimensional (mm2), while the fMRI response is in three dimensions (mm3). iOIS and SSEP have a similar Y location due to two-dimensional surface co-localization. 
Activation size 562 ± 84 mm2 526 ± 343 mm2 8082 ± 1758 mm3 7758 ± 1697 mm3 5172 ± 1252 mm3 
Activation intensity  ± 13 mV  4.2 ± 1.8 × 10–3 DIR  32 ± 5 CU (12%)  36.1 ± 5 CU (12%)  36.8 ± 4 CU (13%) 
Activation COM (Talairach coordinates, mm):     
X (P < 0.08)  45.31 ± 4.30  46.63 ± 8.02  40.71 ± 6.73  41.56 ± 6.42  43.03 ± 7.02 
Y (P <0.005)* –47.86 ± 9.12 –48.15 ± 9.00  –34.29 ± 10.26  –35.73 ± 10.74  –37.30 ± 9.78 
Z (P <0.035)*  45.9 ± 4.65  44.71 ± 8.08  51.35 ± 4.75  49.19 ± 4.39  46.83 ± 3.95 
 SSEP iOIS fMRI Monophasic fMRI Small Dip fMRI/iOIS Correlation 
After co-registration in three dimensions, activation size (region of interest), intensity and COM were calculated for each modality (SSEP, iOIS and fMRI) and fMRI analysis method (monophasic correlation, small dip correlation and iOIS/fMRI correlation). Statistics were averaged across all eight patients. Size and intensity were calculated in native patient space. COM was calculated in the Talairach coordinate system. Intensity is listed as decrease in reflectance (DIR) for iOIS and contrast units (CU) for fMRI. An intensity threshold of ±13 mV was defined for the SSEP maps. The intensity change of fMRI was 12 and 13% of the overall MR signal. 
*Statistical significance between iOIS and the fMRI Monophasic analysis. Note that the iOIS and SSEP sizes are two-dimensional (mm2), while the fMRI response is in three dimensions (mm3). iOIS and SSEP have a similar Y location due to two-dimensional surface co-localization. 
Activation size 562 ± 84 mm2 526 ± 343 mm2 8082 ± 1758 mm3 7758 ± 1697 mm3 5172 ± 1252 mm3 
Activation intensity  ± 13 mV  4.2 ± 1.8 × 10–3 DIR  32 ± 5 CU (12%)  36.1 ± 5 CU (12%)  36.8 ± 4 CU (13%) 
Activation COM (Talairach coordinates, mm):     
X (P < 0.08)  45.31 ± 4.30  46.63 ± 8.02  40.71 ± 6.73  41.56 ± 6.42  43.03 ± 7.02 
Y (P <0.005)* –47.86 ± 9.12 –48.15 ± 9.00  –34.29 ± 10.26  –35.73 ± 10.74  –37.30 ± 9.78 
Z (P <0.035)*  45.9 ± 4.65  44.71 ± 8.08  51.35 ± 4.75  49.19 ± 4.39  46.83 ± 3.95 
Table 2

Modeling of the iOIS/fMRI response

Figure 1.

 SSEP, iOIS, and fMRI (monophasic correlation) intrasubject studies reveal differences in spatial localization. Stimulations were provided by a 110 Hz pneumatically driven finger vibrator. iOIS utilized the same stimulator (SSEPs are from median nerve stimulation). Images are co-localized on a surface model obtained from T1-weighted MRI. All signals localized at the level of the superior genu of the central sulcus. iOIS responses (middle upper) were recorded over the crest of both sensory and motor cortex. SSEPs (upper left) also localized across this region. fMRI (upper right), however, localized within the central sulcus, and extended deep and superior. The SSEP and iOIS responses are co-localized in the zoomed window (lower left). The iOIS and SSEP maps closely co-localize. The iOIS and fMRI responses are co-localized in the zoomed window (lower right). The fMRI signal clearly exists within the sulcus and extends far superior and deep (mesh cortex is translucent), while the iOIS signal localizes on the gyrus. iOIS is psuedocolored to represent reflectance change (right scale, 0.5–3.0 × 10–3) from baseline. The fMRI, demarcated in red (all statistically significant pixels P < 0.01), used the monophasic model for correlation analysis. The SSEP map indicates mV change (left scale, –26.5 to +26.5 mV). The scale bar is 1 cm.

Figure 1.

 SSEP, iOIS, and fMRI (monophasic correlation) intrasubject studies reveal differences in spatial localization. Stimulations were provided by a 110 Hz pneumatically driven finger vibrator. iOIS utilized the same stimulator (SSEPs are from median nerve stimulation). Images are co-localized on a surface model obtained from T1-weighted MRI. All signals localized at the level of the superior genu of the central sulcus. iOIS responses (middle upper) were recorded over the crest of both sensory and motor cortex. SSEPs (upper left) also localized across this region. fMRI (upper right), however, localized within the central sulcus, and extended deep and superior. The SSEP and iOIS responses are co-localized in the zoomed window (lower left). The iOIS and SSEP maps closely co-localize. The iOIS and fMRI responses are co-localized in the zoomed window (lower right). The fMRI signal clearly exists within the sulcus and extends far superior and deep (mesh cortex is translucent), while the iOIS signal localizes on the gyrus. iOIS is psuedocolored to represent reflectance change (right scale, 0.5–3.0 × 10–3) from baseline. The fMRI, demarcated in red (all statistically significant pixels P < 0.01), used the monophasic model for correlation analysis. The SSEP map indicates mV change (left scale, –26.5 to +26.5 mV). The scale bar is 1 cm.

Figure 3.

 iOIS/fMRI time course studies reveal differences in response. Values are the average across all eight subjects. Four temporal profiles are presented. (i) iOIS responded within 500 ms, peaked at 2.0 s and returned to baseline by 6.0 s. (ii) The monophasic correlation analysis for fMRI responded by 1.5 s, peaked at 4.5 s and returned to baseline at 10 s. (iii) The fMRI signal also was obtained using a small initial dip correlation model (fMRI small dip). This signal had an early response dip peaking at 1.0 s before being overshadowed by a later response peak. (iv) The fMRI signal also was obtained over the gray matter pixels co-localizing with the iOIS signal. This time course (iOIS/fMRI) included an additional early response change. The time course responded with a dip within 500 ms, peaked negatively at 1.5 s, returned to baseline at 3.0 s, rose to peak at ~5 s and returned to baseline by 10 s. The pre- and post-stimulus baseline level was 3.0 ± 3.0%. Initial dip time points at 500 ms, 1.5 s and 2.0 s were statistically significant (P < 0.001) from this baseline. (v) This iOIS/fMRI time course was then used for an additional fMRI correlation time course (iOIS/fMRI correlation). This time course also responded with a dip at 500 ms. The intensity of this dip was larger than the ‘fMRI small dip’ time course, but smaller than the empirically obtained ‘iOIS/fMRI’ time course. No oscillatory signals were observed. Stimulations (fMRI and iOIS) were provided by a 110 Hz pneumatically driven finger vibrator, synchronized and phase lagged to the scanner. The Y axis is signal change from baseline normalized to 0–100%.

Figure 3.

 iOIS/fMRI time course studies reveal differences in response. Values are the average across all eight subjects. Four temporal profiles are presented. (i) iOIS responded within 500 ms, peaked at 2.0 s and returned to baseline by 6.0 s. (ii) The monophasic correlation analysis for fMRI responded by 1.5 s, peaked at 4.5 s and returned to baseline at 10 s. (iii) The fMRI signal also was obtained using a small initial dip correlation model (fMRI small dip). This signal had an early response dip peaking at 1.0 s before being overshadowed by a later response peak. (iv) The fMRI signal also was obtained over the gray matter pixels co-localizing with the iOIS signal. This time course (iOIS/fMRI) included an additional early response change. The time course responded with a dip within 500 ms, peaked negatively at 1.5 s, returned to baseline at 3.0 s, rose to peak at ~5 s and returned to baseline by 10 s. The pre- and post-stimulus baseline level was 3.0 ± 3.0%. Initial dip time points at 500 ms, 1.5 s and 2.0 s were statistically significant (P < 0.001) from this baseline. (v) This iOIS/fMRI time course was then used for an additional fMRI correlation time course (iOIS/fMRI correlation). This time course also responded with a dip at 500 ms. The intensity of this dip was larger than the ‘fMRI small dip’ time course, but smaller than the empirically obtained ‘iOIS/fMRI’ time course. No oscillatory signals were observed. Stimulations (fMRI and iOIS) were provided by a 110 Hz pneumatically driven finger vibrator, synchronized and phase lagged to the scanner. The Y axis is signal change from baseline normalized to 0–100%.

References

Bandettini PA, Jesmanowicz A, Wong EC, Hyde JS (
1993
) Processing strategies for time-course data sets in functional MRI of the human brain.
Magn Reson Med
 
30
:
161
–173.
Bandettini PA, Kwong KK, Davis TL, Tootell RBH, Wong EC, Fox PT, Belliveau JW, Weisskoff RM, Rosen BR (
1997
) Characterization of blood oxygenation and flow changes during prolonged brain activation.
Hum Brain Mapp
 
5
:
93
–109.
Belliveau JW, Kennedy DN Jr, McKinstry RC, Buchbinder BR, Weisskoff RM, Bohen MS, Vevea JM, Brady TJ, Rosen BR (
1991
) Functional mapping of the human visual cortex by magnetic resonance imaging.
Science
 
254
:
716
–719.
Belliveau JW, Kwong KK, Kennedy DN, Baker JR, Stern CE, Benson R et al. (
1992
) Magnetic resonance imaging mapping of brain function. Human visual cortex.
Invest Radiol
 
2
:
S59
–65.
Blood AJ, Narayan SM, Toga AW (
1995
) Stimulus parameters influence characteristics of optical intrinsic responses in somatosensory cortex.
J Cereb Blood Flow Metab
 
15
:
1109
–1121.
Cannestra AF, Blood AJ, Black KL, Toga AW (
1996
) The evolution of optical signals in human and rodent cortex.
NeuroImage
 
3
:
202
–208.
Cannestra AF, Black KL, Martin NA, Cloughesy T, Burton JS, Rubinstein E, Woods RP, Toga AW, (
1998
) Topographical and temporal specificity of human intraoperative optical intrinsic signals.
NeuroReport
 
9
:
2557
–2563.
Cannestra AF, Pouratian N, Shomer MH, and Toga AW (
1998
) Refractory periods observed by intrinsic signal and fluorescent dye imaging.
J Neurophysiol
 
80
:
1522
–1532.
Cannestra AF, Bookheimer SY, O'Farrell A, Sicotte N, Martin NA, Becker D, Rubino G, Toga AW (
2000
) Temporal and topographical characterization of language cortices utilizing intraoperative optical intrinsic signals.
NeuroImage
 
12
:
41
–54.
Chen W, Kato T, Zhu XH, Strupp J, Ogawa S, Ugurbil K (
1998
) Spatial and temporal differentiation of fMRI BOLD response in promary visual cortex of human brain during sustained visual stimulation.
Magn Reson Med
 
39
:
520
–527.
Cohen MS (
1997
) Parametric analysis of fMRI data using linear systems methods.
NeuroImage
 
6
:
93
–103.
Cohen MS, Bookheimer SY (
1994
) Localization of brain function using magnetic resonance imaging.
Trends Neurosci
 
17
:
268
–276.
Cox SB, Woolsey TA, Rovainen CM (
1993
) Localized dynamic changes in cortical blood flow with whisker stimulation corresponds to matched vascular and neuronal architecture of rat barrels.
J Cereb Blood Flow Metab
 
13
:
899
–913.
ErnstT, Hennig J (
1994
) Observation of a fast response in functional MRI.
Magn Reson Med
 
32
:
146
–9.
Frahm J, Bruhn H, Merboldt K-D, Hanicke W (
1992
) Dynamic MR imaging of human brain oxygenation during rest and photic stimulation.
J Magn Reson Imag
 
5
:
501
–505.
Fransson P, Kruger G, Merboldt, KD, Frahm J (
1998
) Temporal characteristics of oxygen sensitive MRI responses to visula activation in humans.
Magn Reson Med
 
39
:
912
–919.
Frostig RD, Lieke EE, Ts'o DY, Grinvald A (
1990
) Cortical functional architecture and local coupling between neuronal activity and the microcirculation revealed by in vivo high-resolution optical imaging of intrinsic signals.
Proc Natl Acad Sci USA
 
87
:
6082
–6086.
Gao JH, Miller I, Lai S, Xiong J, Fox PT (
1996
) Quantitative assessments of blood inflow effects in functional MRI signals.
Magn Reson Med
 
36
:
314
–319.
Geyer S, Schleicher A, Zilles K (
1999
) Areas 3a, 3b, and 1 of human promary somatosensory cortex.
NeuroImage
 
10
:
63
–83.
Hennig J, Janz C, Speck O, Ernst T (
1995
) Functional spectroscopy of brain activation following a single light pulse: examinations of the mechanism of the fast initial response.
Int J Imag Syst Technol
 
6
:
203
–208.
Hess A, Stiller D, Kaulisch T, Heil P, Scheich H (
2000
) New insights into the hemodynamic blood oxygenation level-dependent response through combination of functional magnetic resonance imaging and optical recording in gerbil barrel cortex.
J Neurosci
 
20
:
3328
–3338.
Hu X, Tuong HL, Ugurbil K (
1996
) Evaluation of the early response in fMRI using short stimulus duration.
NeuroImage
 
3
:
S7
.
Janesick JR, Kleesen KP, Elliott T (
1987
) Charge-coupled-device charge- collection efficiency and the photon-transfer technique.
Optical Engng
 
26
:
972
–980.
Janz C, Speck O, Hennig J (
1997
) Time resolved measurements of brain activation after a short stimulus: new results on the physiologic mechanisms of the cortical response.
NMR Biomed
 
10
:
222
–229.
Jezzard P, Rauschecker JP, Malonek D (
1998
) An in vivo model for functional MRI in cat visual cortex.
Magn Reson Med
 
38
:
699
–705.
Kim D-S, Duong TQ, Kim S-G (
1999
) Magnetic resonance imaging of iso-orientation domains in cat visual cortex using early negative bold signals.
Neurosci Abstr
 
25
:
783
.
Kwong KK, Belliveau JW, Chesler DA, Goldberg IE, Weisskopf RM, Poncelet BP, Kennedy DN, Hoppel BE, Cohen MS, Turner R, Cheng H-M, Brady TJ, Rosen BR (
1992
) Dynamic magnetic resonance imaging of human brain activity during primary sensory stimulation.
Proc Natl Acad Sci USA
 
89
5675
–5679.
Lai S Hopkins AL, Haacke EM, Li D, Wasserman BA, Buckley P, Friedman L Meltzer H Hedera P, Friedland R (
1993
) Identification of vascular structures as a major source of signal contrast in high resolution 2D and 3D functional activation imaging of the motor cortex at 15 T: preliminary results.
Magn Reson Med
 
30
:
387
–392.
Lindauer U, Villringer A, Dirnagl U (
1993
) Characterization of CBF response to somatosensory stimulation: model and influence of anesthetics.
Am Physiol Soc
 
264
:
H1223
–H1228.
Lou HC, Edvinsson L, MacKenzie ET (
1987
) The concept of coupling blood flow to brain function: revision required?
Ann Neurol
 
22
:
289
–297.
MacDonald D, (1998) A method for identifying geometrically simple surfaces from three dimensional images. PhD Thesis, McGill University.
Malonek D, Grinvald A (
1996
) Interactions between electrical activity and cortical microcirculation revealed by imaging spectroscopy: implications for functional brain mapping.
Science
 
272
:
551
–554.
Marota JJA, Ayata C, Moskowitz MA, Weisskoff, RM, Rosen BR, Mandeville JB (
1999
) Investigation of the early response to rat forepaw stimulation.
Magn Reson Med
 
41
:
247
–252.
Mayhew J, Zheng Y, Hou Y, Vuksanovic B, Berwick J, Askew S, Coffey P, (
1999
) Spectroscopic analysis of changes in remitted illumination: the response to increased neural activity in brain.
NeuroImage
 
10
:
304
–326.
McIntosh J, Zhang Y, Kidambi S, Harshbarger T, Mason G Prohost GM, Twieg D (1996) Echo-time dependence of the functional MRI ‘fast response’. In: Proceedings of the Society of Magnetic Resonance 4th scientific meeting, New York, p. 284.
Menon RS, Ogawa S, Xiaoping H, Strupp JP, Anderson P, Ugurbil K (
1995
) BOLD based functional MRI at 4 Tesla includes a capillary bed contribution: echo-planar imaging correlates with previous optical imaging using intrinsic signals.
Magn Reson Med
 
33
:
453
–459.
Narayan SM, Santori EM, Toga AW (
1994
) Mapping functional activity in rodent cortex using optical intrinsic signals.
Cereb Cortex
 
4
:
195
–205.
Narayan SM, Esfahani P, Blood AJ, Sikkens L, Toga AW (
1995
) Functional increases in cerebral blood volume over somatosensory cortex.
J Cereb Blood Flow Metab
 
15
:
754
–765.
Ngai AC, Ko KR, Morii S, Winn HR (
1988
) Effect of sciatic nerve stimulation on pial arterioles in rats.
Am J Physiol
 
254
:
H133
–139.
Nuwer MR, Banoczi WR, Cloughesy TF, Hoch DB, Peacock W, Levesque MF, Black KL, Martin NA, Becker DP (
1992
) Topographic mapping of somatosensory evoked potential helps identify motor cortex more quickly in the operating room.
Brain Topog
 
5
:
53
–58.
Ogawa S, Lee TM, Kay AR, Tank DW (
1990
) Brain magnetic resonance imaging with contrast dependent on blood oxygenation.
Proc Natl Acad Sci USA
 
87
:
9868
–9872.
Phelps ME, Mazziotta JC (
1995
) Positron emission tomography: human brain function and biochemistry.
Science
 
228
:
799
–809.
Pouratian N, Bookheimer SY, O'Farrell AM, Sicotte NL, Cannestra AF, Becker DP, Toga AW (
2000
) Optical imaging of bilingual cortical representations.
J Neurosurg
 
93
:
676
–681.
Rao SM, Binder JR, Bandettini PA, Hammeke TA, Yetkin FZ, Jesmanowicz A, Lisk LM, Morris GL, Mueller MW, Estkowski LD (
1993
) Functional magnetic resonance imaging of complex human movements.
Neurology
 
43
:
2311
–2318.
Savoy RL, O'Craven KM, Weisskoff RM, Davis TL, Baker J, Rosen B (
1994
) Exploring the temporal boundaries of fMRI: measuring responses to very brief visual stimuli.
Soc Neurosci Abs
 
20
:
1264
.
Sorensen AG, Tievsky AL, Ostergaard L, Weisskoff RM, Rosen BR (
1997
) Contrast agents in functional MR imaging.
J Magn Reson Imag
 
7
:
47
–55.
Talairach J, Tournoux P (1988) Co-planar stereotaxic atlas of the human brain. New York: Thieme.
Toga AW, Banerjee PK (
1993
) Registration revisited.
J Neurosci Methods
 
48
:
1
–13.
Toga AW, Cannestra AF, Black KL (
1995
) The temporal/spatial evolution of optical signals in human cortex.
Cereb Cortex
 
5
:
561
–565.
Vanzetta I, Grinvald A (
1999
) Increased cortical oxidative metabolism due to sensory stimulation: implications for functional brain imaging.
Science
 
286
:
1555
–1558.
Villringer A, Planck J, Hock C, Schleinkofer L, Dirnagl U (
1993
) Near infrared spectroscopy (NIRS): a tool to study hemodynamic changes during activation of brain function in human adults.
Neurosci Lett
 
154
:
101
–104.
Woods RP (1998) Automated global polynomial warping. In: Brain warping (Toga AW, ed.), pp. 365–376. San Diego, CA: Academic Press.
Woods RP, Grafton ST, Holmes CJ, Cherry SR, Mazziotta JC (
1998
) Automated image registration. I. General methods and intrasubject, intramodality validation.
J Comput Assist Tomog
 
22
:
139
–152.
Woods RP, Grafton ST, Watson JDG, Sicotte NL, Mazziotta JC (
1998
) Automated image registration. II. Intersubject validation of linear and nonlinear models.
J Comput Assist Tomog
 
22
:
153
–165.
Yacoub E, Huu T, Ugurbil K, Hu X (
1999
) Further evaluation of the initial negative response in functional magnetic resonance imaging.
Magn Reson Med
 
41
:
436
–441.