| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
The Journal of Neurophysiology Vol. 79 No. 6 June 1998, pp. 3257-3265
Copyright ©1998 by the American Physiological Society
RAPID COMMUNICATION
Section on Functional Brain Imaging, Laboratory of Brain and Cognition, National Institute of Mental Health, National Institutes of Health, Bethesda, Maryland 20892-1366
 |
ABSTRACT |
|---|
 Abstract
 Introduction
Methods
Results
Discussion
References
|
|---|
Clark, Vincent P., Jose M. Maisog, and James V. Haxby. fMRI study of face perception and memory using random stimulus sequences. J. Neurophysiol. 79: 3257-3265, 1998. A new functional magnetic resonance imaging (fMRI) method was used to investigate the functional neuroanatomy of face perception and memory. Whole-brain fMRI data were acquired while four types of stimuli were presented sequentially in an unpredictable pseudorandom order at a rate of 0.5 Hz. Stimulus types were a single repeated memorized target face, unrepeated novel faces, nonsense scrambled faces, and a blank screen. Random stimulus sequences were designed to generate a functional response to each stimulus type that was uncorrelated with responses to other stimuli. This allowed fMRI responses to each stimulus type to be examined separately using multiple regression. Signal increases were found for all stimuli in ventral posterior cortex. Responses to intact faces extended to more anterior locations of occipitotemporal cortex than did responses to scrambled faces, consistent with previous studies of face perception. Responses evoked by novel faces were in regions of ventral occipitotemporal cortex medial to regions in which significant responses were evoked by the target face. The repeated target face stimulus also evoked activity in widely distributed regions of frontal and parietal cortex. These results demonstrate that cortical hemodynamic responses to interleaved novel and repeated stimuli can be distinguished and measured using fMRI with appropriate stimulus sequences and data analysis methods. This method can now be used to examine the neural systems involved in cognitive tasks that were previously impossible to study using positron emission tomography or fMRI.
In many electrophysiological and psychophysical experiments of attention, perception, and memory, randomized stimulus orders are used with target stimuli presented at unpredictable times (Hillyard and Picton 1987 Seven healthy right-handed volunteers (4 women, ages 22-32 yr) participated in this study. All subjects gave written informed consent.
The mean accuracy for detection of the memorized target face was 95%. The average reaction time for the correct detection was 641 ± 103 ms.
These results demonstrate that fMRI can be used to detect and discriminate the responses to different types of stimuli presented in randomized, interleaved sequences with rapid rates of stimulus presentation. Nonspecific responses to all stimuli were found in posterior occipital cortex. The presentation of intact faces (target and novel faces) evoked activity in more anterior locations of occipitotemporal cortex. This general pattern of response topography was consistent across the seven subjects studied and agrees with previous fMRI experiments of face perception using a blocked stimulus design (Clark et al. 1996
INTRODUCTION
Abstract
Introduction
Methods
Results
Discussion
References
; Regan 1989). By contrast, most previous studies of object perception and memory using positron emission tomography (PET) and functional magnetic resonance imaging (fMRI) have used experimental paradigms in which trains of similar stimuli are presented in repeated blocks (e.g., Clark et al. 1996, 1997; Haxby et al. 1994, 1996; Kanwisher et al. 1997; McCarthy et al. 1997b; Puce et al. 1996; Tulving et al. 1996). In the typical blocked experimental design, the functional responses evoked by individual stimulus types presented within blocks are not resolved. Because of this limitation, responses to unpredictable presentations of individual novel stimuli cannot be discriminated from responses to familiar stimuli presented in the same sequence. This comparison is important for understanding the neural mechanisms supporting memory and recognition. The use of blocked stimulus
designs also makes it difficult to compare the results of these fMRI investigations with those of previous electrophysiological and psychophysical research that have used randomized designs. Furthermore, the specificity of functional data obtained using blocked designs may be compromised by a number of factors, including the subjects' expectancies regarding upcoming stimuli within and between blocks, and changes in subjects' attentional state and level of arousal between blocks.
; Buckner et al. 1996; McCarthy et al. 1997a; Zarahn et al. 1997). This method limits the number of trials that may be collected in individual experiments, and the use of long intervals without stimuli may influence subjects' level of attention and arousal. A recent paper by Dale and Buckner (1997) used stimulus presentation rates as fast as one stimulus every 2 s and showed that responses in retinotopically organized lower-order visual cortical areas of the right and left hemispheres responded selectively to stimuli positioned in the left and right visual fields, respectively. In the present study, we investigated whether differential responses in higher-order visual cortical areas to different types of stimuli presented rapidly in the same visual field location can be distinguished using fMRI.
, 1997; Haxby et al. 1994; Kanwisher et al. 1997; McCarthy et al. 1997b; Puce et al. 1996; Sergent et al. 1992). No previous PET or fMRI studies have examined responses to novel versus repeated objects presented in interleaved
Ê
sequences. A previous PET study comparing novel versus repeated objects using a blocked stimulus design (Tulving et al. 1996) found responses in medial occipitotemporal areas to novel stimuli. It was expected that the results of the present study would be consistent with these previous studies, and that the use of interleaved stimulus sequences would provide information on differential responses to novel and repeated face stimuli when presented in the same block of stimuli in an unpredictable sequence.
METHODS
Abstract
Introduction
Methods
Results
Discussion
References
; Haxby et al. 1996). Also, stimulus onset was synchronized to the onset of MRI data acquisition, which facilitated data analysis. Subjects were trained before scanning to make a speeded right handed button press to the target face.
; Maisog et al. 1995). Examples of predicted response waveforms for scrambled faces, novel faces, and the target face are illustrated in Fig. 1. The cross-correlations among these models of fMRI response to each stimulus type were then computed for each random stimulus sequence. Five different stimulus sequences were selected for use in this experiment that had nonsignificant correlations of predicted hemodynamic responses to target, novel, and scrambled faces (|r| < 0.05). Thus the predicted responses to the three types of stimuli embedded in each sequence were essentially orthogonal.
View larger version (43K):
[in a new window]
FIG. 1.
Modeled hemodynamic response time series (black lines) shown for the scrambled (S), novel (N), and memorized target (T) face stimuli. The time of appearance of stimuli are indicated at the bottom of the figure, and by the location of the gray bars. Sixty seconds of 1 experimental run is shown. Amplitude of gray bars indicates asymptotic maximum of functional magnetic resonance imaging (fMRI) response. Stimulus sequences were selected for each 3-min experimental run that generated hemodynamic responses that were uncorrelated between the 3 stimulus conditions.
). The statistical significance of stimulus-evoked changes in MRI signal intensity was evaluated using multiple regression (Friston et al. 1995b; Haxby et al. 1997; Maisog et al. 1996; Neter et al. 1990). For this
analysis, the delayed and smoothed time series computed for the three stimulus types were used as regressors. Multiple regression finds the weighted sum of these regressors that best fits the obtained fMRI time series, using a least-squares test of fit. Insofar as overlapping responses are approximately additive (Boynton et al. 1996; Dale and Buckner 1997), significant responses evoked by any combination of the three stimulus types can be identified within the same voxel location. However, because the regressors were uncorrelated, significant activity evoked by one stimulus condition could not be modeled using any linear combination of the other stimuli's regressors, and thus could only reach significance for the regressor representing the response to that stimulus. Additional regressors were included in the analysis to partial out variance due to baseline shifts between time series and linear drift within time series. The significance of the regression weight for each stimulus type was tested using the extra sums of squares test, which generated a Wilks' 
map (Neter et al. 1990). The Wilks' map was then converted to an F-test map by an exact transformation(Rencher 1995). The F ratio gives the ratio of the magnitude of signal variance explained by the statistical model to the magnitude of the unexplained signal variance. A Z statistic was then formed from the F ratio, using a probability integral transformation (Friston et al. 1991). To form the Z statistic, the magnitude of the F ratio and the degrees of freedom in the denominator of the F ratio were multiplied by an experimentally derived correction factor (0.42) (Maisog et al. 1995) to correct for temporal autocorrelations due to the smoothness of the hemodynamic response. Regions of interest (ROIs) were located by identifying clusters of three or more contiguous voxels where MRI signal increased in response to a stimulus condition and that were significant at P < 0.05. ROIs were also located by identifying clusters of three or more contiguous voxels where MRI signal increased in response to a specific combination of stimulus conditions. ROIs were spatially normalized using SPM96 (Friston et al. 1995a) for use in subsequent multisubject comparisons.
View larger version (136K):
[in a new window]
FIG. 2.
Color-coded percentage changemaps from 3 subjects demonstrating the response to scrambled, novel, and target face stimuli, plotted onto coplanar normalized coronal structural images of right hemisphere located at 
20z mm in the Talairach coordinate system (Talairach and Tournoux 1988). The right side of the brain is displayed on the left, and left on the right. Areas showing increased MR signal during face-matching relative to control are shown in red and yellow, and areas showing decreased signal are shown in blue and green. Color intensity ranges from +0.8 to 0.8%, representing average percentage change in response to a single stimulus.
. 2) 
2 distribution maps were obtained for each stimulus condition separately by obtaining the sum of the squares of all seven subjects' Z score probability maps, which had been thresholded to include only voxels with positive signal change. The sum of squared Z score values has been shown to have a 2 distribution (Hugill 1985; Papoulis 1991). This was done separately for the statistical probability maps of the responses to scrambled faces, novel faces, and the target face. These three group 2 maps were then thresholded at P < 0.05(df = 7).
View larger version (131K):
[in a new window]
FIG. 3.
For all slices, regions of significantly increased signal evoked by scrambled faces (S; green), novel faces (N; red), and the target face (T; blue) are shown. Results are plotted onto coplanar axial structural images. The location of the images along the z dimension in Talairach space (ranging from 20 to + 50 mm) are shown. The right side of the brain is displayed on the left, and left on the right. Rows marked "O" show regions with a signifcant overlap of response, defined as a minimum of 3 subjects showing a significant response for the same condition in the same normalized voxel location. Voxel color determined by relative number of subjects showing significant response to each condition at that normalized voxel location. Rows marked "2" show regions with a significant 2 amplitude, obtained by summing the squared Z score values across subjects. Voxel color determined by relative amplitude of 2 test of each condition at that normalized voxel location.
2 s to 0 s).
RESULTS
Abstract
Introduction
Methods
Results
Discussion
References
2 maps. Both analyses confirmed that the different patterns of response to each stimulus type were consistent across subjects, as can be seen in Fig. 3, but areas of overlap and significant 2 were less extensive than areas of activation found in results of individual subjects. As the overlap analysis proved to be more conservative than the 2 analysis, the overlap maps were examined for volume, anatomic center of mass, and BA.
4x, 95y, 16z) and in the fusiform gyrus bilaterally (32x, 74y, 19z and 36x, 74y, 19z). Additional regions in posterior occipital cortex (posterior to 80y) were found that responded to scrambled and novel faces, but not to the target face. Some voxels were found in individual subjects that responded to scrambled and target faces, but not to novel faces. However, none of these voxels overlapped across subjects.
56y, 22z). This is nearby to foci found in many other studies of face perception using a blocked stimulus design (Clark et al. 1996: 37x, 55y, 10z; Haxby et al. 1994: 38x, 58y, 0z; Kanwisher et al. 1997: 40x, 55y, 10z; McCarthy et al. 1997b: 40x, 59y, 22z; Sergent et al. 1992: 37x, 55y, 11z). Activity evoked by novel faces in anterior ventral occipitotemporal cortex (anterior to 60y) was situated medially to activity evoked by the target face in both the left hemisphere [35x for novel faces vs. 41x for target face, t = 7.7, df(1,28), P < 10
5] and the right hemisphere [38x for novel faces vs. 49x for target face; t = 5.8, df(1,99), P < 107].
View larger version (103K):
[in a new window]
FIG. 4.
Stimulus locked average time series are shown for 4 groups of contiguous voxels of 1 subject. The white region was found to respond significantly to all 3 stimulus types, the purple region responded to novel and target faces, but not the scrambled face, the red region responded only to novel faces, and the blue region responded only to the target face. Averaged evoked fMRI time series responses to scrambled faces are shown in green, responses to novel faces shown in red, and responses to target face shown in blue. Time series plotted from 4 s before stimulus onset to 18 s after stimulus onset. Amplitude of percentage change relative to prestimulus baseline is shown.
View larger version (136K):
[in a new window]
FIG. 5.
Group mean evoked fMRI responses to the target face stimulus, normalized to the peak response amplitude. Mean response waveforms were averaged across subjects for 8 separate regions, determined by location relative to the AC-PC line. This included right hemisphere (R, dotted line) vs. left hemisphere (L, dashed line), dorsal (D, thin line) vs. ventral (V, thick line), and anterior (A, green line) vs. posterior (P, red line). Voxels from all brain regions peaked at the same time (6 s).
DISCUSSION
Abstract
Introduction
Methods
Results
Discussion
References
, 1997; Kanwisher et al. 1997; McCarthy et al. 1997b; Puce et al. 1996).
60y), whereas greater responses to the memorized target face were found in lateral portions of ventral temporal cortex. This difference in response topography suggests that stimulus evoked responses in anterior occipitotemporal cortex may be segregated depending on stimulus familiarity or behavioral relevance. This conjecture is consistent with the work of Tulving et al. (1996), who found a temporal/limbic "novelty detection" region in a PET study of novel versus previously seen complex pictures. This is also consistent with other studies where stimulus novelty was examined using single word stimuli (Kapur et al. 1995) and for auditorily presented sentences (Tulving et al. 1994). The finding of more lateral activity evoked by the target face is consistent with lesion studies in nonhuman primates, which have shown that lesions in ventrolateral extrastriate cortex, near or within area TEO, greatly decreased performance on discrimination learning tasks where fine visual discriminations must be made (Iwai and Mishkin 1968, 1969; Kikuchi and Awai 1980). Many regions of frontal and parietal cortex responded to the target face but did not respond to the other stimuli. These results and those of McCarthy et al. (1997a) and Menon et al. (1997) suggest that the detection of infrequent target stimuli evokes activity in a number of widely distributed cortical regions.
; Buckner et al. 1996; McCarthy et al. 1997a; Zarahn et al. 1997). Although useful for some applications, the use of slow rates of stimulus presentation limits the number of stimuli that can be presented within a single experiment and may influence subjects' level of attention and arousal. Also, this makes it difficult to compare findings with previous electrophysiological and psychophysical studies, which have typically used more rapid rates of stimulus presentation (Hillyard and Picton 1987; Regan 1989). Dale and Buckner (1997) used stimulus presentation rates that ranged between 1 stimulus per 20 s and 1 stimulus per 2 s, using a simple reversing checkerboard stimulus presented sequentially in the right and left visual fields. This study employed the careful use of stimulus randomization, so that when time-locked averaging was employed, the number of each stimulus type occurring at each relative time point before and after the stimuli of interest were approximately equal throughout the epoch. This ensured that the other stimulus types did not contribute to the average evoked response of the stimulus of interest. This study showed that responses to left and right field stimuli could be distinguished in retinotopically organized lower order visual areas. This study also showed that stimuli presented in a series evoked hemodynamic activity that was nearly identical to the linear addition of activity evoked by individual stimuli, which was also found in the study of Boynton et al. (1996).
Ê
The present study investigated whether fMRI can be used to distinguish responses in higher-order cortical areas to different types of randomly presented stimuli based on differences in stimulus form and familiarity. In contrast to the method of stimulus randomization used by Dale and Buckner (1997), the method employed here produced orthogonal response waveforms between the three stimulus types used, but did not require constraints in the relative proportions of different stimulus types required for time-locked averaging. The multiple regression analysis did not require that the response to each individual stimulus be examined, but it instead examined the summed response across all stimuli of the same type within individual runs. Because the fMRI signal is additive and varies in amplitude with variations in stimulus density over the preceding 8-10 s, the methods presented here could be used to examine responses to stimuli presented using any combination of stimulus presentation rate and ISI. This is true given that an accurate model of neural and hemodynamic responses to the stimuli can be constructed, and the stimulus sequences can be arranged such that orthogonal regressors for each stimulus type can be obtained for use in the multiple regression analysis.
). Such differences could reduce the sensitivity of this method, as the response waveforms would be shifted in time relative to the regressors. However, differences in response latency were analyzed across a range of brain regions and were not found to vary by more than a small amount. Future improvements in the modeling of the neural and hemodynamic responses to stimuli will serve to reduce statistical error and improve signal detection.
| |
ACKNOWLEDGEMENTS |
|---|
We are very grateful to Drs. J. Ingeholm, A. Song, and P. Jezzard, and to E. Hoffman, J. Schouten, and the staff of the National Institutes of Health In Vivo NMR Research Center for assistance, and to Drs. L. Ungerleider, R. Parasuraman, M. Beauchamp, L. Petit, and J. Van Horn for comments. We thank Dr. R. Woods for providing AIR software for image registration and Dr. K. Friston for providing the software for spatial normalization.
| |
FOOTNOTES |
|---|
Address for reprint requests: V. P. Clark, Dept. of Psychiatry, MC2017, University of Connecticut Health Center, 263 Farmington Ave., Farmington, CT 06030.
Received 2 September 1997; accepted in final form 3 February 1998.
| |
REFERENCES |
|---|
|
Abstract
Introduction
Methods
Results
Discussion
References
|
|---|
This article has been cited by other articles:
|
|
 |
|
  L. Tamm, V. Menon, and A. L. Reiss Parietal Attentional System Aberrations During Target Detection in Adolescents With Attention Deficit Hyperactivity Disorder: Event-Related fMRI Evidence Am J Psychiatry, June 1, 2006; 163(6): 1033 - 1043. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 R. L. Buckner The hemodynamic inverse problem: Making inferences about neural activity from measured MRI signals PNAS, March 4, 2003; 100(5): 2177 - 2179. [Full Text] [PDF]
|
|
|
|
 |
|
 K. J. Meador, P. G. Ray, J. R. Echauz, D. W. Loring, and G. J. Vachtsevanos Gamma coherence and conscious perception Neurology, September 24, 2002; 59(6): 847 - 854. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 W. Staffen, A. Mair, H. Zauner, J. Unterrainer, H. Niederhofer, A. Kutzelnigg, S. Ritter, S. Golaszewski, B. Iglseder, and G. Ladurner Cognitive function and fMRI in patients with multiple sclerosis: evidence for compensatory cortical activation during an attention task Brain, June 1, 2002; 125(6): 1275 - 1282. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
B. A. Arnow, J. E. Desmond, L. L. Banner, G. H. Glover, A. Solomon, M. L. Polan, T. F. Lue, and S. W. Atlas Brain activation and sexual arousal in healthy, heterosexual males Brain, May 1, 2002; 125(5): 1014 - 1023. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 R. L. Buckner, A. Z. Snyder, A. L. Sanders, M. E. Raichle, and J. C. Morris Functional Brain Imaging of Young, Nondemented, and Demented Older Adults J. Cogn. Neurosci., November 1, 2000; 12(90002): 24S - 34. [Abstract] [Full Text]
|
|
|
|
|
|
B. Rossion, L. Dricot, A. Devolder, J.-M. Bodart, M. Crommelinck, B. de Gelder, and R. Zoontjes Hemispheric Asymmetries for Whole-Based and Part-Based Face Processing in the Human Fusiform Gyrus J. Cogn. Neurosci., September 1, 2000; 12(5): 793 - 802. [Abstract] [Full Text]
|
|
|
|
 |
|
 T. P.L. Roberts, E. A. Disbrow, H. C. Roberts, and H. A. Rowley Quantification and Reproducibility of Tracking Cortical Extent of Activation by Use of Functional MR Imaging and Magnetoencephalography AJNR Am. J. Neuroradiol., August 1, 2000; 21(8): 1377 - 1387. [Abstract] [Full Text]
|
|
|
|
 |
|
 V. P. Clark, S. Fannon, S. Lai, R. Benson, and L. Bauer Responses to Rare Visual Target and Distractor Stimuli Using Event-Related fMRI J Neurophysiol, May 1, 2000; 83(5): 3133 - 3139. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
S. Pollmann, R. Weidner, H.J. Müller, and D.Y. von Cramon A Fronto-Posterior Network Involved in Visual Dimension Changes J. Cogn. Neurosci., May 1, 2000; 12(3): 480 - 494. [Abstract] [Full Text]
|
|
|
|
 |
|
 Y. Jiang, J. V. Haxby, A. Martin, L. G. Ungerleider, and R. Parasuraman Complementary Neural Mechanisms for Tracking Items in Human Working Memory Science, January 28, 2000; 287(5453): 643 - 646. [Abstract] [Full Text]
|
|
|
|
 |
|
 C. L. Leveroni, M. Seidenberg, A. R. Mayer, L. A. Mead, J. R. Binder, and S. M. Rao Neural Systems Underlying the Recognition of Familiar and Newly Learned Faces J. Neurosci., January 15, 2000; 20(2): 878 - 886. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
R. Cabeza and L. Nyberg Imaging Cognition II: An Empirical Review of 275 PET and fMRI Studies J. Cogn. Neurosci., January 1, 2000; 12(1): 1 - 47. [Abstract] [Full Text]
|
|
|
|
|
|
L. Petit, S. M. Courtney, L. G. Ungerleider, and J. V. Haxby Sustained Activity in the Medial Wall during Working Memory Delays J. Neurosci., November 15, 1998; 18(22): 9429 - 9437. [Abstract] [Full Text] [PDF]
|
|
| |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
