Neural correlates of perceptual rivalry in the human brain

Science

Volumn 280, Issue 5371, 1998, Pages 1930-1934

  Lumer, Erik D ^a   Friston, Karl J ^a   Rees, Geraint ^a  

^a UNIVERSITY COLLEGE LONDON   (United Kingdom)

Author keywords

[No Author keywords available]

Indexed keywords

ARTICLE; CORRELATION FUNCTION; FRONTAL LOBE; HUMAN; IMAGE ANALYSIS; PARIETAL LOBE; PERCEPTION; PRIORITY JOURNAL; VISUAL FIELD; VISUAL STIMULATION;

BRAIN; BRAIN MAPPING; FRONTAL LOBE; HUMANS; MAGNETIC RESONANCE IMAGING; NEURONS; PARIETAL LOBE; SPACE PERCEPTION; VISION, BINOCULAR; VISUAL CORTEX; VISUAL PERCEPTION;

EID: 0032546988     PISSN: 00368075     EISSN: None     Source Type: Journal    
DOI: 10.1126/science.280.5371.1930     Document Type: Article

References (54)

1. N. K. Logothetis and J. D. Schall, Science 245, 761 (1989); F. Crick, Nature 379, 485 (1996).
2. N. K. Logothetis and J. D. Schall, Science 245, 761 (1989); F. Crick, Nature 379, 485 (1996).
3. R. W. Lansing, Science 146, 1325 (1964).
4. D. A. Leopold and N. K. Logothetis, Nature 379, 549 (1996); D. L. Sheinberg and N. K. Logothetis, Proc. Natl. Acad. Sci. U.S.A. 94, 3408 (1997).
5. D. A. Leopold and N. K. Logothetis, Nature 379, 549 (1996); D. L. Sheinberg and N. K. Logothetis, Proc. Natl. Acad. Sci. U.S.A. 94, 3408 (1997).
6. The notion that rivalry involves central rather than peripheral processes goes back, at least, to Helmholtz [Treatise on Physiological Optics (Optical Society of America, New York, 1911)]. This view contrasts with later proposals that rivalry results from competition between monocular channels at an early stage of visual processing. For a general review of the psychophysical evidence motivating these alternative accounts, see P. Walker [Psychol. Bull. 85, 376 (1978)] and R. Blake [Psychol. Rev. 96, 145 (1989)]. Recent studies have shown that rivalry occurs normally when conflicting images are rapidly exchanged between the eyes, thus arguing against monocular theories [N. K. Logothetis, D. A. Leopold, D. L. Sheinberg, Nature 380, 621 (1996)].
7. The notion that rivalry involves central rather than peripheral processes goes back, at least, to Helmholtz [Treatise on Physiological Optics (Optical Society of America, New York, 1911)]. This view contrasts with later proposals that rivalry results from competition between monocular channels at an early stage of visual processing. For a general review of the psychophysical evidence motivating these alternative accounts, see P. Walker [Psychol. Bull. 85, 376 (1978)] and R. Blake [Psychol. Rev. 96, 145 (1989)]. Recent studies have shown that rivalry occurs normally when conflicting images are rapidly exchanged between the eyes, thus arguing against monocular theories [N. K. Logothetis, D. A. Leopold, D. L. Sheinberg, Nature 380, 621 (1996)].
8. The notion that rivalry involves central rather than peripheral processes goes back, at least, to Helmholtz [Treatise on Physiological Optics (Optical Society of America, New York, 1911)]. This view contrasts with later proposals that rivalry results from competition between monocular channels at an early stage of visual processing. For a general review of the psychophysical evidence motivating these alternative accounts, see P. Walker [Psychol. Bull. 85, 376 (1978)] and R. Blake [Psychol. Rev. 96, 145 (1989)]. Recent studies have shown that rivalry occurs normally when conflicting images are rapidly exchanged between the eyes, thus arguing against monocular theories [N. K. Logothetis, D. A. Leopold, D. L. Sheinberg, Nature 380, 621 (1996)].
9. The notion that rivalry involves central rather than peripheral processes goes back, at least, to Helmholtz [Treatise on Physiological Optics (Optical Society of America, New York, 1911)]. This view contrasts with later proposals that rivalry results from competition between monocular channels at an early stage of visual processing. For a general review of the psychophysical evidence motivating these alternative accounts, see P. Walker [Psychol. Bull. 85, 376 (1978)] and R. Blake [Psychol. Rev. 96, 145 (1989)]. Recent studies have shown that rivalry occurs normally when conflicting images are rapidly exchanged between the eyes, thus arguing against monocular theories [N. K. Logothetis, D. A. Leopold, D. L. Sheinberg, Nature 380, 621 (1996)].
10. A. Borsellino, A. De Marco, A. Allazetta, S. Rinesi, B. Bartolini, Kybernetik 10, 139 (1972).
11. J. M. Wolfe, Nature 380, 587 (1996); T. J. Andrews and D. Purves, Proc. Natl. Acad. Sci. U.S.A. 94, 9905 (1997).
12. J. M. Wolfe, Nature 380, 587 (1996); T. J. Andrews and D. Purves, Proc. Natl. Acad. Sci. U.S.A. 94, 9905 (1997).
13. Participants wore nonmetallic stereoscopic glasses and viewed a small projection screen through a mirror mounted on top of the RF coil above the participant's head. Head movements were restrained by foam pads. Stimuli were projected onto the screen by means of an LCD projector. They consisted of pairs of square images, each subtending approximately 3.5° of visual angle. Before scanning began, participants used a keypad to modify the lateral separation of the two images used during dichoptic stimulation so that each image was seen through only one eye and that stereoscopic fusion and binocular rivalry could be comfortably attained. All subsequent stimuli presented during that scanning run were then presented at those locations. Subjects indicated with two keys perceptual transitions from face to grating or vice versa, using their dominant hand. A Siemens VISION (Siemens, Erlangen) operating at 2 T was used to acquire BOLD contrast functional images. Image volumes were acquired continuously every 400 ms, each comprising 48 contiguous 3-mm-thick slices to give whole-brain coverage with an in-plane resolution of 3 mm by 3 mm. Functional imaging was performed in two scanning runs comprising 496 volumes in total. In each scanning run, after eight image volumes were discarded to allow for T1 equilibration effects, the rivalry experimental condition was presented for 41 s (10 scans) followed by the replay condition for 41 s followed by 41 s of rest. Each condition was then repeated for a total of eight repetitions per run. At the beginning of each experimental session a T1-weighted anatomical image was acquired for coregistration with the functional images.
14. Because of the slow time constants of BOLD responses (>2 s), contrasts of the face and grating stimuli were manipulated to promote long intervals between consecutive perceptual alternations, and therefore optimize the conditions for detecting neurophysiological correlates of transition events by fMRI. Frequency histograms of dominance time for the face and grating were constructed for each participant from the rivalry reports collected during the scanning session. Of the 10 volunteers that were scanned, 6 had long mean dominance times and were retained for analysis of brain activity (4 males and 2 females; mean age, 31 years; age range, 27 to 34 years; 5 right-handed and 1 left-handed; mean face dominance, 2.9 s; mean grating dominance, 5.7 s).
15. 2 = 1).
16. Analysis was carried out using Statistical Parametric Mapping software (SPM96, http//:www.fil.ion.ucl. ac.uk/spm). The imaging time series was realigned, spatially normalized to the stereotactic space of Talairach and Tournoux, and smoothed with a Gaussian kernel of 8 mm full width half maximum [J. Talairach and P. Tournoux, Co-Planar Stereotaxic Atlas of the Human Brain (Thieme, New York, 1988); K. J. Friston et al., Hum. Brain Mapping 3, 165 (1995); K. J. Friston et al., ibid. 2, 189 (1995); K. J. Friston et al., NeuroImage 2, 157 (1995); K. J. Friston et al., Magn. Reson. Med. 35, 346 (1996)]. Voxels that were activated during the rivalry and replay conditions were identified by means of a statistical model containing two components that represented the transient responses produced by the transition events in each condition, together with two boxcar wave forms that modeled and removed the condition-specific differences in mean evoked activity. The event-related changes in evoked activity were modeled by convolving an empirically derived hemodynamic impulse response function with trains of unitary events that were aligned on the reported perceptual transitions [O. Josephs et al., Hum. Brain Mapping 5, 243 (1997); B. R. Rosen et al., Proc. Natl. Acad. Sci. U.S.A. 95, 773 (1998)]. In addition, low-frequency sine and cosine waves modeled and removed subject-specific low-frequency drifts in signal [A. P. Holmes et al., Neurolmage 5, S480 (1997)] and global changes in activity were removed by proportional scaling. Each component of the model served as a regressor in a multiple regression analysis. The event-related components, which constitute the effects of interest, were tested to see whether they could account for a significant portion of the variance, independent of the variance attributable to the other regressors. All statistical results are based on a single-voxel Z threshold of 3.09 (corresponding to P < 0.001, uncorrected for multiple comparisons). Resultant regions of activation were characterized in terms of their peak heights. In assessing statistical significance, we made a correction (based on the theory of random Gaussian fields) for multiple comparisons across the whole-brain volume examined and report only regions of activation above a threshold corresponding to P < 0.05, corrected [K. J. Friston et al., Hum. Brain Mapping 1, 210 (1994)].
17. Analysis was carried out using Statistical Parametric Mapping software (SPM96, http//:www.fil.ion.ucl. ac.uk/spm). The imaging time series was realigned, spatially normalized to the stereotactic space of Talairach and Tournoux, and smoothed with a Gaussian kernel of 8 mm full width half maximum [J. Talairach and P. Tournoux, Co-Planar Stereotaxic Atlas of the Human Brain (Thieme, New York, 1988); K. J. Friston et al., Hum. Brain Mapping 3, 165 (1995); K. J. Friston et al., ibid. 2, 189 (1995); K. J. Friston et al., NeuroImage 2, 157 (1995); K. J. Friston et al., Magn. Reson. Med. 35, 346 (1996)]. Voxels that were activated during the rivalry and replay conditions were identified by means of a statistical model containing two components that represented the transient responses produced by the transition events in each condition, together with two boxcar wave forms that modeled and removed the condition-specific differences in mean evoked activity. The event-related changes in evoked activity were modeled by convolving an empirically derived hemodynamic impulse response function with trains of unitary events that were aligned on the reported perceptual transitions [O. Josephs et al., Hum. Brain Mapping 5, 243 (1997); B. R. Rosen et al., Proc. Natl. Acad. Sci. U.S.A. 95, 773 (1998)]. In addition, low-frequency sine and cosine waves modeled and removed subject-specific low-frequency drifts in signal [A. P. Holmes et al., Neurolmage 5, S480 (1997)] and global changes in activity were removed by proportional scaling. Each component of the model served as a regressor in a multiple regression analysis. The event-related components, which constitute the effects of interest, were tested to see whether they could account for a significant portion of the variance, independent of the variance attributable to the other regressors. All statistical results are based on a single-voxel Z threshold of 3.09 (corresponding to P < 0.001, uncorrected for multiple comparisons). Resultant regions of activation were characterized in terms of their peak heights. In assessing statistical significance, we made a correction (based on the theory of random Gaussian fields) for multiple comparisons across the whole-brain volume examined and report only regions of activation above a threshold corresponding to P < 0.05, corrected [K. J. Friston et al., Hum. Brain Mapping 1, 210 (1994)].
18. Analysis was carried out using Statistical Parametric Mapping software (SPM96, http//:www.fil.ion.ucl. ac.uk/spm). The imaging time series was realigned, spatially normalized to the stereotactic space of Talairach and Tournoux, and smoothed with a Gaussian kernel of 8 mm full width half maximum [J. Talairach and P. Tournoux, Co-Planar Stereotaxic Atlas of the Human Brain (Thieme, New York, 1988); K. J. Friston et al., Hum. Brain Mapping 3, 165 (1995); K. J. Friston et al., ibid. 2, 189 (1995); K. J. Friston et al., NeuroImage 2, 157 (1995); K. J. Friston et al., Magn. Reson. Med. 35, 346 (1996)]. Voxels that were activated during the rivalry and replay conditions were identified by means of a statistical model containing two components that represented the transient responses produced by the transition events in each condition, together with two boxcar wave forms that modeled and removed the condition-specific differences in mean evoked activity. The event-related changes in evoked activity were modeled by convolving an empirically derived hemodynamic impulse response function with trains of unitary events that were aligned on the reported perceptual transitions [O. Josephs et al., Hum. Brain Mapping 5, 243 (1997); B. R. Rosen et al., Proc. Natl. Acad. Sci. U.S.A. 95, 773 (1998)]. In addition, low-frequency sine and cosine waves modeled and removed subject-specific low-frequency drifts in signal [A. P. Holmes et al., Neurolmage 5, S480 (1997)] and global changes in activity were removed by proportional scaling. Each component of the model served as a regressor in a multiple regression analysis. The event-related components, which constitute the effects of interest, were tested to see whether they could account for a significant portion of the variance, independent of the variance attributable to the other regressors. All statistical results are based on a single-voxel Z threshold of 3.09 (corresponding to P < 0.001, uncorrected for multiple comparisons). Resultant regions of activation were characterized in terms of their peak heights. In assessing statistical significance, we made a correction (based on the theory of random Gaussian fields) for multiple comparisons across the whole-brain volume examined and report only regions of activation above a threshold corresponding to P < 0.05, corrected [K. J. Friston et al., Hum. Brain Mapping 1, 210 (1994)].
19. Analysis was carried out using Statistical Parametric Mapping software (SPM96, http//:www.fil.ion.ucl. ac.uk/spm). The imaging time series was realigned, spatially normalized to the stereotactic space of Talairach and Tournoux, and smoothed with a Gaussian kernel of 8 mm full width half maximum [J. Talairach and P. Tournoux, Co-Planar Stereotaxic Atlas of the Human Brain (Thieme, New York, 1988); K. J. Friston et al., Hum. Brain Mapping 3, 165 (1995); K. J. Friston et al., ibid. 2, 189 (1995); K. J. Friston et al., NeuroImage 2, 157 (1995); K. J. Friston et al., Magn. Reson. Med. 35, 346 (1996)]. Voxels that were activated during the rivalry and replay conditions were identified by means of a statistical model containing two components that represented the transient responses produced by the transition events in each condition, together with two boxcar wave forms that modeled and removed the condition-specific differences in mean evoked activity. The event-related changes in evoked activity were modeled by convolving an empirically derived hemodynamic impulse response function with trains of unitary events that were aligned on the reported perceptual transitions [O. Josephs et al., Hum. Brain Mapping 5, 243 (1997); B. R. Rosen et al., Proc. Natl. Acad. Sci. U.S.A. 95, 773 (1998)]. In addition, low-frequency sine and cosine waves modeled and removed subject-specific low-frequency drifts in signal [A. P. Holmes et al., Neurolmage 5, S480 (1997)] and global changes in activity were removed by proportional scaling. Each component of the model served as a regressor in a multiple regression analysis. The event-related components, which constitute the effects of interest, were tested to see whether they could account for a significant portion of the variance, independent of the variance attributable to the other regressors. All statistical results are based on a single-voxel Z threshold of 3.09 (corresponding to P < 0.001, uncorrected for multiple comparisons). Resultant regions of activation were characterized in terms of their peak heights. In assessing statistical significance, we made a correction (based on the theory of random Gaussian fields) for multiple comparisons across the whole-brain volume examined and report only regions of activation above a threshold corresponding to P < 0.05, corrected [K. J. Friston et al., Hum. Brain Mapping 1, 210 (1994)].
20. Analysis was carried out using Statistical Parametric Mapping software (SPM96, http//:www.fil.ion.ucl. ac.uk/spm). The imaging time series was realigned, spatially normalized to the stereotactic space of Talairach and Tournoux, and smoothed with a Gaussian kernel of 8 mm full width half maximum [J. Talairach and P. Tournoux, Co-Planar Stereotaxic Atlas of the Human Brain (Thieme, New York, 1988); K. J. Friston et al., Hum. Brain Mapping 3, 165 (1995); K. J. Friston et al., ibid. 2, 189 (1995); K. J. Friston et al., NeuroImage 2, 157 (1995); K. J. Friston et al., Magn. Reson. Med. 35, 346 (1996)]. Voxels that were activated during the rivalry and replay conditions were identified by means of a statistical model containing two components that represented the transient responses produced by the transition events in each condition, together with two boxcar wave forms that modeled and removed the condition-specific differences in mean evoked activity. The event-related changes in evoked activity were modeled by convolving an empirically derived hemodynamic impulse response function with trains of unitary events that were aligned on the reported perceptual transitions [O. Josephs et al., Hum. Brain Mapping 5, 243 (1997); B. R. Rosen et al., Proc. Natl. Acad. Sci. U.S.A. 95, 773 (1998)]. In addition, low-frequency sine and cosine waves modeled and removed subject-specific low-frequency drifts in signal [A. P. Holmes et al., Neurolmage 5, S480 (1997)] and global changes in activity were removed by proportional scaling. Each component of the model served as a regressor in a multiple regression analysis. The event-related components, which constitute the effects of interest, were tested to see whether they could account for a significant portion of the variance, independent of the variance attributable to the other regressors. All statistical results are based on a single-voxel Z threshold of 3.09 (corresponding to P < 0.001, uncorrected for multiple comparisons). Resultant regions of activation were characterized in terms of their peak heights. In assessing statistical significance, we made a correction (based on the theory of random Gaussian fields) for multiple comparisons across the whole-brain volume examined and report only regions of activation above a threshold corresponding to P < 0.05, corrected [K. J. Friston et al., Hum. Brain Mapping 1, 210 (1994)].
21. Analysis was carried out using Statistical Parametric Mapping software (SPM96, http//:www.fil.ion.ucl. ac.uk/spm). The imaging time series was realigned, spatially normalized to the stereotactic space of Talairach and Tournoux, and smoothed with a Gaussian kernel of 8 mm full width half maximum [J. Talairach and P. Tournoux, Co-Planar Stereotaxic Atlas of the Human Brain (Thieme, New York, 1988); K. J. Friston et al., Hum. Brain Mapping 3, 165 (1995); K. J. Friston et al., ibid. 2, 189 (1995); K. J. Friston et al., NeuroImage 2, 157 (1995); K. J. Friston et al., Magn. Reson. Med. 35, 346 (1996)]. Voxels that were activated during the rivalry and replay conditions were identified by means of a statistical model containing two components that represented the transient responses produced by the transition events in each condition, together with two boxcar wave forms that modeled and removed the condition-specific differences in mean evoked activity. The event-related changes in evoked activity were modeled by convolving an empirically derived hemodynamic impulse response function with trains of unitary events that were aligned on the reported perceptual transitions [O. Josephs et al., Hum. Brain Mapping 5, 243 (1997); B. R. Rosen et al., Proc. Natl. Acad. Sci. U.S.A. 95, 773 (1998)]. In addition, low-frequency sine and cosine waves modeled and removed subject-specific low-frequency drifts in signal [A. P. Holmes et al., Neurolmage 5, S480 (1997)] and global changes in activity were removed by proportional scaling. Each component of the model served as a regressor in a multiple regression analysis. The event-related components, which constitute the effects of interest, were tested to see whether they could account for a significant portion of the variance, independent of the variance attributable to the other regressors. All statistical results are based on a single-voxel Z threshold of 3.09 (corresponding to P < 0.001, uncorrected for multiple comparisons). Resultant regions of activation were characterized in terms of their peak heights. In assessing statistical significance, we made a correction (based on the theory of random Gaussian fields) for multiple comparisons across the whole-brain volume examined and report only regions of activation above a threshold corresponding to P < 0.05, corrected [K. J. Friston et al., Hum. Brain Mapping 1, 210 (1994)].
22. Analysis was carried out using Statistical Parametric Mapping software (SPM96, http//:www.fil.ion.ucl. ac.uk/spm). The imaging time series was realigned, spatially normalized to the stereotactic space of Talairach and Tournoux, and smoothed with a Gaussian kernel of 8 mm full width half maximum [J. Talairach and P. Tournoux, Co-Planar Stereotaxic Atlas of the Human Brain (Thieme, New York, 1988); K. J. Friston et al., Hum. Brain Mapping 3, 165 (1995); K. J. Friston et al., ibid. 2, 189 (1995); K. J. Friston et al., NeuroImage 2, 157 (1995); K. J. Friston et al., Magn. Reson. Med. 35, 346 (1996)]. Voxels that were activated during the rivalry and replay conditions were identified by means of a statistical model containing two components that represented the transient responses produced by the transition events in each condition, together with two boxcar wave forms that modeled and removed the condition-specific differences in mean evoked activity. The event-related changes in evoked activity were modeled by convolving an empirically derived hemodynamic impulse response function with trains of unitary events that were aligned on the reported perceptual transitions [O. Josephs et al., Hum. Brain Mapping 5, 243 (1997); B. R. Rosen et al., Proc. Natl. Acad. Sci. U.S.A. 95, 773 (1998)]. In addition, low-frequency sine and cosine waves modeled and removed subject-specific low-frequency drifts in signal [A. P. Holmes et al., Neurolmage 5, S480 (1997)] and global changes in activity were removed by proportional scaling. Each component of the model served as a regressor in a multiple regression analysis. The event-related components, which constitute the effects of interest, were tested to see whether they could account for a significant portion of the variance, independent of the variance attributable to the other regressors. All statistical results are based on a single-voxel Z threshold of 3.09 (corresponding to P < 0.001, uncorrected for multiple comparisons). Resultant regions of activation were characterized in terms of their peak heights. In assessing statistical significance, we made a correction (based on the theory of random Gaussian fields) for multiple comparisons across the whole-brain volume examined and report only regions of activation above a threshold corresponding to P < 0.05, corrected [K. J. Friston et al., Hum. Brain Mapping 1, 210 (1994)].
23. Analysis was carried out using Statistical Parametric Mapping software (SPM96, http//:www.fil.ion.ucl. ac.uk/spm). The imaging time series was realigned, spatially normalized to the stereotactic space of Talairach and Tournoux, and smoothed with a Gaussian kernel of 8 mm full width half maximum [J. Talairach and P. Tournoux, Co-Planar Stereotaxic Atlas of the Human Brain (Thieme, New York, 1988); K. J. Friston et al., Hum. Brain Mapping 3, 165 (1995); K. J. Friston et al., ibid. 2, 189 (1995); K. J. Friston et al., NeuroImage 2, 157 (1995); K. J. Friston et al., Magn. Reson. Med. 35, 346 (1996)]. Voxels that were activated during the rivalry and replay conditions were identified by means of a statistical model containing two components that represented the transient responses produced by the transition events in each condition, together with two boxcar wave forms that modeled and removed the condition-specific differences in mean evoked activity. The event-related changes in evoked activity were modeled by convolving an empirically derived hemodynamic impulse response function with trains of unitary events that were aligned on the reported perceptual transitions [O. Josephs et al., Hum. Brain Mapping 5, 243 (1997); B. R. Rosen et al., Proc. Natl. Acad. Sci. U.S.A. 95, 773 (1998)]. In addition, low-frequency sine and cosine waves modeled and removed subject-specific low-frequency drifts in signal [A. P. Holmes et al., Neurolmage 5, S480 (1997)] and global changes in activity were removed by proportional scaling. Each component of the model served as a regressor in a multiple regression analysis. The event-related components, which constitute the effects of interest, were tested to see whether they could account for a significant portion of the variance, independent of the variance attributable to the other regressors. All statistical results are based on a single-voxel Z threshold of 3.09 (corresponding to P < 0.001, uncorrected for multiple comparisons). Resultant regions of activation were characterized in terms of their peak heights. In assessing statistical significance, we made a correction (based on the theory of random Gaussian fields) for multiple comparisons across the whole-brain volume examined and report only regions of activation above a threshold corresponding to P < 0.05, corrected [K. J. Friston et al., Hum. Brain Mapping 1, 210 (1994)].
24. Analysis was carried out using Statistical Parametric Mapping software (SPM96, http//:www.fil.ion.ucl. ac.uk/spm). The imaging time series was realigned, spatially normalized to the stereotactic space of Talairach and Tournoux, and smoothed with a Gaussian kernel of 8 mm full width half maximum [J. Talairach and P. Tournoux, Co-Planar Stereotaxic Atlas of the Human Brain (Thieme, New York, 1988); K. J. Friston et al., Hum. Brain Mapping 3, 165 (1995); K. J. Friston et al., ibid. 2, 189 (1995); K. J. Friston et al., NeuroImage 2, 157 (1995); K. J. Friston et al., Magn. Reson. Med. 35, 346 (1996)]. Voxels that were activated during the rivalry and replay conditions were identified by means of a statistical model containing two components that represented the transient responses produced by the transition events in each condition, together with two boxcar wave forms that modeled and removed the condition-specific differences in mean evoked activity. The event-related changes in evoked activity were modeled by convolving an empirically derived hemodynamic impulse response function with trains of unitary events that were aligned on the reported perceptual transitions [O. Josephs et al., Hum. Brain Mapping 5, 243 (1997); B. R. Rosen et al., Proc. Natl. Acad. Sci. U.S.A. 95, 773 (1998)]. In addition, low-frequency sine and cosine waves modeled and removed subject-specific low-frequency drifts in signal [A. P. Holmes et al., Neurolmage 5, S480 (1997)] and global changes in activity were removed by proportional scaling. Each component of the model served as a regressor in a multiple regression analysis. The event-related components, which constitute the effects of interest, were tested to see whether they could account for a significant portion of the variance, independent of the variance attributable to the other regressors. All statistical results are based on a single-voxel Z threshold of 3.09 (corresponding to P < 0.001, uncorrected for multiple comparisons). Resultant regions of activation were characterized in terms of their peak heights. In assessing statistical significance, we made a correction (based on the theory of random Gaussian fields) for multiple comparisons across the whole-brain volume examined and report only regions of activation above a threshold corresponding to P < 0.05, corrected [K. J. Friston et al., Hum. Brain Mapping 1, 210 (1994)].
25. Analysis was carried out using Statistical Parametric Mapping software (SPM96, http//:www.fil.ion.ucl. ac.uk/spm). The imaging time series was realigned, spatially normalized to the stereotactic space of Talairach and Tournoux, and smoothed with a Gaussian kernel of 8 mm full width half maximum [J. Talairach and P. Tournoux, Co-Planar Stereotaxic Atlas of the Human Brain (Thieme, New York, 1988); K. J. Friston et al., Hum. Brain Mapping 3, 165 (1995); K. J. Friston et al., ibid. 2, 189 (1995); K. J. Friston et al., NeuroImage 2, 157 (1995); K. J. Friston et al., Magn. Reson. Med. 35, 346 (1996)]. Voxels that were activated during the rivalry and replay conditions were identified by means of a statistical model containing two components that represented the transient responses produced by the transition events in each condition, together with two boxcar wave forms that modeled and removed the condition-specific differences in mean evoked activity. The event-related changes in evoked activity were modeled by convolving an empirically derived hemodynamic impulse response function with trains of unitary events that were aligned on the reported perceptual transitions [O. Josephs et al., Hum. Brain Mapping 5, 243 (1997); B. R. Rosen et al., Proc. Natl. Acad. Sci. U.S.A. 95, 773 (1998)]. In addition, low-frequency sine and cosine waves modeled and removed subject-specific low-frequency drifts in signal [A. P. Holmes et al., Neurolmage 5, S480 (1997)] and global changes in activity were removed by proportional scaling. Each component of the model served as a regressor in a multiple regression analysis. The event-related components, which constitute the effects of interest, were tested to see whether they could account for a significant portion of the variance, independent of the variance attributable to the other regressors. All statistical results are based on a single-voxel Z threshold of 3.09 (corresponding to P < 0.001, uncorrected for multiple comparisons). Resultant regions of activation were characterized in terms of their peak heights. In assessing statistical significance, we made a correction (based on the theory of random Gaussian fields) for multiple comparisons across the whole-brain volume examined and report only regions of activation above a threshold corresponding to P < 0.05, corrected [K. J. Friston et al., Hum. Brain Mapping 1, 210 (1994)].
26. High Z scores in the group analysis reflect a significant activation averaged across individuals. The consistency of this activation was confirmed by examination of statistical contrasts for individual subjects.
27. The reconstructed postevent histograms represent an estimate of the evoked hemodynamic activity in a given voxel attributable to an individual transition event, after the effects of no interest have been removed. Each scan in each subject contributed as many sample points to this estimate as there were events in the 16 seconds preceding the acquisition of the scan. Each sample point was displaced along the time axis by the latency between the acquisition of the scan and the occurrence of the preceding event; its ordinate was calculated by adding to the value of the fitted hemodynamic response function at this latency the residual variance after complete model fitting. The sample points were used to construct a postevent histogram, with bins of 1.25 s.
28. D. C. Van Essen and J. L. Gallant, Neuron 13, 1 (1994); L. G. Ungerleider and J. V. Haxby, Curr. Opin. Neurobiol. 4, 157 (1994); R. Malach et al., Proc. Natl. Acad. Sci. U.S.A. 92, 8135 (1995); N. K. Logothetis and D. L. Sheinberg, Annu. Rev. Neurosci. 19, 557 (1996).
29. D. C. Van Essen and J. L. Gallant, Neuron 13, 1 (1994); L. G. Ungerleider and J. V. Haxby, Curr. Opin. Neurobiol. 4, 157 (1994); R. Malach et al., Proc. Natl. Acad. Sci. U.S.A. 92, 8135 (1995); N. K. Logothetis and D. L. Sheinberg, Annu. Rev. Neurosci. 19, 557 (1996).
30. D. C. Van Essen and J. L. Gallant, Neuron 13, 1 (1994); L. G. Ungerleider and J. V. Haxby, Curr. Opin. Neurobiol. 4, 157 (1994); R. Malach et al., Proc. Natl. Acad. Sci. U.S.A. 92, 8135 (1995); N. K. Logothetis and D. L. Sheinberg, Annu. Rev. Neurosci. 19, 557 (1996).
31. D. C. Van Essen and J. L. Gallant, Neuron 13, 1 (1994); L. G. Ungerleider and J. V. Haxby, Curr. Opin. Neurobiol. 4, 157 (1994); R. Malach et al., Proc. Natl. Acad. Sci. U.S.A. 92, 8135 (1995); N. K. Logothetis and D. L. Sheinberg, Annu. Rev. Neurosci. 19, 557 (1996).
32. N. Kanwisher, J. McDermott, M. M. Chun, J. Neurosci. 17, 4302 (1997).
33. To analyze further the hemodynamic activity associated with perceptual changes in the absence of stimulus change, we tested for the presence of activity that correlated with the perceptual dominance or suppression of the face stimulus during rivalry. Compared to our main analysis of perceptual transitions per se, we expected that such a test would emphasize areas involved in the visual representation of faces, while de-emphasizing additional areas associated with transition events that are not face-specific. Areas that were significantly more active when the face was perceived than when it was not seen (P < 0.05 corrected) were largely confined to the fusiform gyri; maximally activated foci were located at x = 36, y = -45, z = -24 (Z = 7.34); x = 36, y = -60, z = -21 (Z = 6.37); x = -33, y = -66, z= -15 (Z = 6.71); x = -36, y = -45, z = -24 (Z = 6.24).
34. Although both rivalry and replay conditions yielded similar hemodynamic responses in the anterior eingulate and SMA regions, activation in these regions was less significant during replay because of greater intersubject variability. Foci of maximal activation during replay included x = -6, y = 9, z = 45 (Z = 4.69; P = 0.06 corrected) and x = 6. y = 18. z = 48 (Z = 4.56; P = 0.1 corrected).
35. G. F. Poggio, F. Gonzalez, F. Krause, J. Neurosci. 8, 4531 (1988); F. Sengpiel and C. Blakemore, Nature 368, 847 (1994).
36. G. F. Poggio, F. Gonzalez, F. Krause, J. Neurosci. 8, 4531 (1988); F. Sengpiel and C. Blakemore, Nature 368, 847 (1994).
37. Superior parietal activity has not been observed during the performance of difficult visual tasks that do not involve spatial shifts of attention [M. Corbetta, F. M. Miezin, S. Dobmeyer, G. L. Shulman, S. E. Petersen, J. Neurosci. 8, 2383 (1991); H. J. Heinze et al., Nature 372, 543 (1995); M. Corbetta, G. L. Shulman, F. M. Miezin, S. E. Petersen. Science 270, 802 (1995); G. Rees, C. D. Frith, N. Lavie, ibid. 278, 1616 (1997).
38. Superior parietal activity has not been observed during the performance of difficult visual tasks that do not involve spatial shifts of attention [M. Corbetta, F. M. Miezin, S. Dobmeyer, G. L. Shulman, S. E. Petersen, J. Neurosci. 8, 2383 (1991); H. J. Heinze et al., Nature 372, 543 (1995); M. Corbetta, G. L. Shulman, F. M. Miezin, S. E. Petersen. Science 270, 802 (1995); G. Rees, C. D. Frith, N. Lavie, ibid. 278, 1616 (1997).
39. Superior parietal activity has not been observed during the performance of difficult visual tasks that do not involve spatial shifts of attention [M. Corbetta, F. M. Miezin, S. Dobmeyer, G. L. Shulman, S. E. Petersen, J. Neurosci. 8, 2383 (1991); H. J. Heinze et al., Nature 372, 543 (1995); M. Corbetta, G. L. Shulman, F. M. Miezin, S. E. Petersen. Science 270, 802 (1995); G. Rees, C. D. Frith, N. Lavie, ibid. 278, 1616 (1997).
40. Superior parietal activity has not been observed during the performance of difficult visual tasks that do not involve spatial shifts of attention [M. Corbetta, F. M. Miezin, S. Dobmeyer, G. L. Shulman, S. E. Petersen, J. Neurosci. 8, 2383 (1991); H. J. Heinze et al., Nature 372, 543 (1995); M. Corbetta, G. L. Shulman, F. M. Miezin, S. E. Petersen. Science 270, 802 (1995); G. Rees, C. D. Frith, N. Lavie, ibid. 278, 1616 (1997).
41. For reviews on modulation of activity by attention in ventral visual pathways, see R. Desimone and J. Duncan, Annu. Rev. Neurosci. 18, 193 (1995); J. H. R. Maunsell, Science 270. 764 (1995).
42. For reviews on modulation of activity by attention in ventral visual pathways, see R. Desimone and J. Duncan, Annu. Rev. Neurosci. 18, 193 (1995); J. H. R. Maunsell, Science 270. 764 (1995).
43. S. M. Courtney, L. G. Ungerleider, K. Keil, J. M. Haxby, Nature 386, 608 (1997); S. P. O. Scalaidhe, F. A. W. Wilson, P. S. Goldman-Rakic, Science 278 (1997).
44. S. M. Courtney, L. G. Ungerleider, K. Keil, J. M. Haxby, Nature 386, 608 (1997); S. P. O. Scalaidhe, F. A. W. Wilson, P. S. Goldman-Rakic, Science 278 (1997).
45. K. M. Heilman and T. Van Den Abell, Neurology 30, 327 (1980); M.-M. Mesulam, Annals Neurol. 10, 309 (1981); M. Husain and C. Kennard, J. Neurol. 243, 652 (1996).
46. K. M. Heilman and T. Van Den Abell, Neurology 30, 327 (1980); M.-M. Mesulam, Annals Neurol. 10, 309 (1981); M. Husain and C. Kennard, J. Neurol. 243, 652 (1996).
47. K. M. Heilman and T. Van Den Abell, Neurology 30, 327 (1980); M.-M. Mesulam, Annals Neurol. 10, 309 (1981); M. Husain and C. Kennard, J. Neurol. 243, 652 (1996).
48. M. Corbetta, F. M. Miezin, G. L. Shuman, S. E. Petersen, J. Neurosci. 13, 1202 (1993); A.C. Nobre et al., Brain 120, 515 (1997).
49. M. Corbetta, F. M. Miezin, G. L. Shuman, S. E. Petersen, J. Neurosci. 13, 1202 (1993); A.C. Nobre et al., Brain 120, 515 (1997).
50. G. R. Fink et al., Nature 382, 626 (1996).
51. M. Corbetta, Proc. Natl. Acad. Sci. U.S.A. 95, 831 (1998).
52. G. W. Humphreys, C. Romani, A. Olson, M. J. Riddoch, J. Duncan, Nature 372, 357 (1994); M. Husain, K. Shapiro, J. Martin, C. Kennard, ibid. 385, 154 (1997).
53. G. W. Humphreys, C. Romani, A. Olson, M. J. Riddoch, J. Duncan, Nature 372, 357 (1994); M. Husain, K. Shapiro, J. Martin, C. Kennard, ibid. 385, 154 (1997).
54. We thank O. Josephs and C. Büchel for assistance, R. Turner for advice on fMRI, and T. Griffiths and R. Frackowiak for comments on the manuscript. This work was funded by the Wellcome Trust.