A central topic of controversy in the search for cortical mechanisms underlying perceptual awareness concerns the fundamental specialization of the visual system into a dorsal “vision-for-action/Where” stream and a ventral “vision-for-perception/What” stream. Specifically, it has been debated whether suppression of visual perception leads to differential reduction in brain activity in the 2 streams—with the dorsal stream remaining largely unaffected and the ventral stream showing a significant reduction in activity. Here, we examined this issue using the recently introduced method of continuous flash suppression (CFS), which offers a particularly sensitive measure of the link between perception and brain activity. Subjects had to detect, during CFS, images of manipulable man-made objects (tools). Our results show that despite their substantial difference in connectivity and neuroanatomical specialization, both ventral and dorsal stream areas revealed a similarly tight link to perceptual awareness, that is, strong functional magnetic resonance imaging–blood oxygenation level–dependent activity for visible tools but a significant reduction of activity in the invisible condition. Importantly, this result was found when the masks were kept identical in the visible and invisible conditions. Our data lend support to the notion that neuronal activity and perceptual awareness are tightly linked across human high-order visual cortex.
Despite considerable effort in brain research, the principles underlying the link between neuronal activity and conscious perception remain controversial (Crick and Koch 1998; Block 2007). One fundamental question is whether different cortical areas may show a differential link to conscious perception. The second question concerns the anatomical focus of neural events, which give rise to conscious perception. Some studies suggest a close relationship between local neuronal processing in sensory areas and the emergence of conscious percepts (Moutoussis and Zeki 2002), while others link perceptual awareness to activity within a distributed global network, including parietal and frontal areas (Dehaene et al. 1998, 2003). An experimental arena where both questions converge and become amenable to experimental exploration is the subdivision of the primate visual system into 2 cortical sets of areas—the ventral “vision-for-perception/What” and the dorsal “vision-for-action/Where” streams, which have been investigated extensively both in non-human primates and in humans (Ungerleider and Mishkin 1982; Milner and Goodale 1995; Mcintosh et al. 2004).
In healthy subjects, ventral stream areas have been shown to be closely linked to subjects’ perceptual states, that is, unperceived visual stimuli generally result in a drastically reduced cortical activity (Tong et al. 1998; Grill-Spector et al. 2000; Dehaene et al. 2001), albeit some residual activity does remain and can be detected using more sensitive methods, for example, multivariate pattern analysis (MVPA) (Sterzer et al. 2008). In the dorsal stream, however, preserved cortical processing has been reported in the absence of perceptual awareness (Fang and He 2005). These data appear to be in agreement with the neuropsychological observation of patients exhibiting intact visuomotor processing without seeing the grasped objects (Goodale et al. 1991), which became core evidence for the influential notion of a strong perceptual–motor dissociation in the visual system (Milner and Goodale 1995) but argue against an anatomically invariant principle underlying the link between cortical activity and perception. Concerning the question whether conscious perception depends on local versus global cortical processes, these results support the global model since they demonstrate an instance in which activity in the visual cortex is not sufficient for sensory perception. Put simply, the fact that robust activity in the dorsal stream exists while subjects fail to perceive the stimuli must imply that this dorsal activity, by itself, is not sufficient for perception. In other words, these reports demonstrate the need for an additional, likely neuroanatomically global, component in the generation of a conscious visual percept (Dehaene et al. 2006; Del Cul et al. 2007). On the other hand, reports of a close relationship between local “ignitions” of neuronal population activity and perceptual awareness (Tse et al. 2005; Nir et al. 2007; Fisch et al. 2009) appear to argue for a more local process in which intense activity anywhere in high-order visual cortex may be sufficient for visual awareness and in turn predict a similar and tight link to conscious perception in both dorsal and ventral visual streams.
To examine these fundamental issues, we have conducted an object detection experiment using the recently developed continuous flash suppression (CFS) paradigm (Tsuchiya and Koch 2005). This paradigm results in a particularly long-lasting suppression of visual stimuli and therefore is optimal for sensitive detection of blood oxygenation level–dependent (BOLD) signals (Tsuchiya et al. 2006). Target stimuli were images of tools for which a preferential activation of the dorsal stream has been shown (Chao and Martin 2000; Johnson-Frey 2004; Creem-Regehr and Lee 2005; Culham and Valyear 2006).
Our results show a clear-cut reduction in BOLD activity linked to suppression of perceptual awareness in both visual streams, that is, we find strong BOLD activity for visible tools but a significant reduction of cortical activity under invisibility. These results argue for neuroanatomical invariance of perception-related cortical activation and are compatible, albeit not exclusively, with local models of perceptual awareness.
Materials and Methods
Fifteen subjects (12 female, average age: 27 years, range: 23–35 years) with normal visual acuity participated in this study. Subjects gave written informed consent and were paid for their participation. Ten subjects were right-handed. Seven subjects were right eye dominant (see Supplementary Table 1 for more information). Eye dominance was assessed by means of the hole-in-card test as a modified version of the original ABC test (Miles 1930).
Stimuli were video-projected in a dual-projector setup (Thompson et al. 2008) onto a screen behind the head coil at a viewing distance of 100 cm using 2 identical projectors (LCD EMP74; Epson) with a refresh rate of 60 Hz and 2 comparable PCs (2.4 GHz, 3 GB RAM, NVidia GeForce 8500GT) running Matlab 2008a (The Mathworks). For dichoptic stimulation, linear polarizing filters (TechSpec linear polarizing film; Edmund Optics) were mounted in front of each projector’s lens. The screen maintained the polarization of the light projected through it. Subjects viewed the screen with linear polarizers (orientation of polarization 45° and 135°) mounted in a magnetic resonance imaging (MRI)-compatible plastic frame. All subjects confirmed that there was no detectable “ghosting” (or cross talk), that is, the leaking of an image to one eye, when it is intended exclusively for the other eye. Stimuli were presented using the Cogent Toolbox (John Romaya, Laboratory of Neurobiology, Wellcome Department of Imaging Neuroscience; http://www.vislab.ucl.ac.uk/). For CFS stimulation, one PC generated the target stimuli, while the second PC presented the continuous flash stimuli. The 2 PCs communicated via the serial port with latencies of <1 ms. Care was taken that target stimuli were never presented before or after the sequence of flash stimuli (maximal delay of 1 refresh frame = 17 ms).
Target and Mask Stimuli
Twelve different gray-scale images of tools (300 × 300 pixels, 5° × 5° visual angle; Supplementary Fig. 1) served as target stimuli and were presented to the subject’s nondominant eye. A Gaussian kernel with a radius of 6 pixels was used to blur the edges of the stimuli to facilitate suppression.
Two types of gray-scale stimuli served as masks: Mondrians and sinusoidal gratings (450 × 450 pixels, 7.4° × 7.4° visual angle). Mondrians were generated by randomly positioning 1000 black, white, and medium gray squares of different size (ranging from 0.25° to 1.7°). The sinusoidal gratings had a spatial frequency of approximately 3 cycles/degree and were oriented in 0°, 45°, 90°, and 135°. Mask stimuli were flashed to the subject’s dominant eye at 10 Hz. On each flash, a new stimulus was used to avoid immediate repetitions. A red fixation cross (0.5° × 0.5°) at the center of the screen and a frame of black and white squares (7.4° × 7.4°) were displayed throughout all trials to facilitate fixation and binocular alignment.
Our study applied 2 analysis approaches based on the different mask types. In the first analysis approach, we followed the study by Fang and He (2005) who used physically different conditions to manipulate subjects’ perceptual state. In contrast to the study by Fang and He, however, where masks were completely eliminated in order to achieve the target visible condition (a procedure that is likely to generate mask-related differential activity), we only modulated the mask shape from Mondrians to grating stimuli, which should reduce mask-specific differential activations. Mondrian patterns have been used in a number of previous CFS studies (Jiang and He 2006; Yang et al. 2007; Sterzer et al. 2008, 2009) and are known to provide reliable visual suppression even at relatively high target stimulus contrasts (Tsuchiya et al. 2006). During extensive psychophysical piloting, we observed that flashing full-contrast sinusoidal gratings, despite being highly salient, results in much less perceptual suppression than Mondrians. Thus, in the first approach, we compared brain activity in the Mondrian and grating conditions (invisible and visible, respectively). Importantly, we complemented this approach with a second analysis approach. In the second analysis, we focused only on the data obtained in the Mondrian mask condition to avoid any stimulus confounds related to the masks. Mondrian trials were subdivided into those in which subjects failed to perceive the target (invisible) and trials in which they reported seeing parts of the targets (visible). Thus, in the second approach, perceptual state was modulated while the mask stimuli were kept identical.
In each trial, subjects had to detect the presence of a target stimulus in 1 of 2 successive 6 s runs of visual stimulation (Fig. 1). In the target-present run, the same tool image was presented 10 times for 300 ms, each time followed by a blank period of 300 ms. In the first 100 ms of each presentation, the contrast of the tool stimulus was gradually ramped up to facilitate perceptual suppression. In the target-absent run, a blank screen was presented to the nondominant eye, while the continuous flash sequence remained the same as in the target-present run. The order of target-present and target-absent runs was randomized. The 2 runs were separated by a blank interval of 6 s and followed after 6 s by 2 response screens of 3 s duration. The first response screen instructed subjects to make a forced-choice guess (2AFC) indicating via button press the run in which a tool image was presented. Next, subjects were asked to give a subjective visibility rating on a 3-point scale (1 = nothing detected, 2 = tool parts, 3 = whole tool). Similar rating procedures have been used in previous studies (Del Cul et al. 2007; Lamy et al. 2009) and provide a robust basis for the assessment of subjective perceptual awareness. The following trial started after a blank screen of variable duration (mean: 3.75 s, range: 1–6.5 s). Mondrian and grating masks were assigned randomly to each trial, but care was taken that they occurred with the same absolute frequency within a scan session. Each scan session consisted of 12 trials resulting in a session length of approximately 7 min. All 12 tool images were used once within a session, and their order was randomized. All subjects performed 5 scan sessions.
We conducted a localizer functional magnetic resonance imaging (fMRI) experiment for each subject to identify cortical regions that preferentially responded to tool stimuli. We used a standard block design with 5 different categories of stimuli: tool stimuli at full contrast, tool stimuli at the subject-specific contrast as determined by the staircase procedure, Mondrian masks, grating masks, and 2 other types of masks that were not analyzed further. Image size was the same as in the main CFS experiment. During each block, 9 items of a category were presented (800 ms per stimulus; 200 ms interstimulus intervals). Blocks were separated by 6 s blank periods, and each condition was repeated over 5 blocks in counterbalanced order. To maintain attention, subjects had to perform a size change detection task, that is, indicate by a button press when a stimulus was of smaller size (67% original size), which happened at variable time points once per block.
Acquisition and Processing of fMRI Data
Functional images were acquired by T2*-weighted gradient-echo echo-planar imaging (35 slices, flip angle = 75°, time repetition [TR] = 2000 ms, time echo [TE] = 30 ms, voxel size 3 × 3 × 4 mm) on a 3T MRI scanner (Tim Trio, Siemens). We recorded 5 experimental sessions of 210 volumes each and a localizer session of 225 volumes. The first 2 volumes were skipped to account for T1 saturation effects. Anatomical images were acquired using a T1-weighted MPRAGE sequence (176 slices, flip angle = 9°, TR = 2300 ms, TE = 2.98 ms, FOV 256, voxel size 1.0 × 1.0 × 1.1 mm). We used BrainVoyager QX 2.0.8 (Brain Innovation) for image pre-processing (slice scan time correction, 3D motion correction, high-pass filter with 2 cycles/experiment, normalization to Talairach stereotactic space [Talairach and Tournoux 1988], spatial smoothing with an isotropic Gaussian kernel of 6 and 8 mm full-width-half-maximum for single subject and group analyses, respectively) and estimation of statistical maps using a general linear model approach (Friston et al. 1994) with 6 rigid-body realignment parameters as nuisance covariates. In the main experiment, all CFS conditions (i.e., “tools invisible” and “tool parts visible” in target-present runs in Mondrian trials, “tools visible” in target-present runs in grating trials, target-absent runs in Mondrian trials, target-absent runs in grating trials) were modeled as epochs of 6 s duration. Blank and response screens were modeled separately as epochs of 6 s duration. In the localizer experiment, all conditions were modeled as epochs of 9 s durations.
Cortical surfaces were reconstructed from the anatomical images using BrainVoyager QX 2.0.8. The procedure included segmentation of the white matter using a grow-region function, the smooth covering of a sphere around the segmented region, and the expansion of the reconstructed white matter into the gray matter. The sulci were smoothed using a cortical inflation procedure. The surface was cut along the Calcarine sulcus and unfolded into the flattened format. The obtained statistical maps were superimposed on the unfolded cortex, and the Talairach coordinates were determined for the center of each regions of interest (ROI). All group maps are based on random-effects models (Friston et al. 1999). The statistical group maps were corrected for multiple comparisons using false discovery rate (FDR) control (Benjamini and Hochberg 1995; Genovese et al. 2002) and projected on the inflated and flattened cortical surface of one representative subject. The BVQXtools/NeuroElf toolbox by Jochen Weber (http://neuroelf.net/) was used for batch processing of fMRI data and calculation of parameter estimates.
Definition of ROI
Based on the localizer experiment, we defined a number of ROIs from which we extracted parameter estimates at the single subject level. Tool-sensitive high-order visual areas as well as early visual cortex were identified on a subject-by-subject basis in 2 steps. First, at the group level, the contrast “Tools > Mondrians” was mapped at P < 0.05 (FDR corrected) and projected on an inflated and flattened cortical surface. Based on this map (see Fig. 3B), search spaces around the centers of 5 different peak activations in the left and right hemisphere were defined: lateral-occipital area (LO) and posterior-fusiform gyrus (pFS) belonging to the lateral-occipital complex (LOC) (Malach et al. 1995, 2002), intra-parietal sulcus (IPS), a region we tentatively defined as V3A/V7, and early visual cortex (negative peaks here because Mondrians activated more than tools). The size of the search spaces was adjusted to eventually cover either approximately 30 functional voxels (small ROIs; Supplementary Table 2) or approximately 400 functional voxels (large ROIs; Supplementary Table 3). Next, for each subject’s first-level contrast Tools > Mondrians, all voxels within this search space were selected that passed a more lenient threshold (P < 0.001, uncorrected, for small ROIs; P < 0.05, uncorrected, for large ROIs). Tool stimuli used for the definition of ROI had the same contrast as in the main experiment. Similarly, we identified putative area hMT+ (Tootell et al. 1995) based on flashed grating masks in the main experiment (contrast “grating masks > baseline”; Supplementary Table 2).
Contrast of Tool Stimuli
To adjust the contrast of the tool stimuli, we first minimized major differences between the images (e.g., in the original images the fork was darker than the mug). Next, we used a staircase procedure to determine detection thresholds individually for each subject (Levitt 1971); Mondrian masks were used as continuous flash stimuli. While the staircase procedure and the main experiment were performed under identical viewing conditions inside the scanner, there was a difference of trial structure including timing parameters (staircase experiment: 2.4 s stimulation, 0.75 s blank, 2.4 s stimulation, 0.75 s blank, temporal 2AFC; main experiment: 6 s stimulation, 6 s blank, 6 s stimulation, 6 s blank, temporal 2AFC and visibility rating). Subjects performed the staircase experiment during the MPRAGE anatomical scan (ca. 9 min); 2 to 3 staircases (ca. 20 trials each) were completed during this period. Based on extensive previous psychophysical testing, we reduced the obtained contrast by a fixed percentage to achieve reliable perceptual suppression throughout the main experiment in the Mondrian condition, while keeping the target stimuli fully visible in the grating mask condition. In 3 subjects, contrast was reduced to 75% of the determined contrast threshold; in 12 subjects, it was reduced to 50%. In most subjects, the staircase procedure yielded relatively high contrast thresholds; accordingly, the average contrast used in the main experiment was 0.44 (range 0.15–0.75). We tested whether the tool stimuli at the individual contrast levels remained fully visible in the grating mask condition, which was the case for all subjects. In a separate control experiment under the same viewing conditions inside the magnet, we monocularly superimposed parts of tool images on the Mondrian masks (“piecemeal percept”) to test whether subjects (N = 3, new subjects) would report the target stimuli. Targets were reported as visible in ∼100% of trials.
As can be seen in Figure 2A, Mondrian mask stimuli suppressed visual perception of tools in most trials (whole: 5.1% ± 1.9, parts: 17.9% ± 3.7, nothing: 77.0% ± 4.4; mean frequency of visibility ratings ± standard error of the mean). Presenting identical target stimuli but masking them with sinusoidal gratings produced the opposite result, now the tools were reported as fully visible in most trials (whole: 97.5% ± 0.9, parts: 1.4% ± 0.8, nothing: 1.1% ± 0.4). Thus, as reported previously (Fang and He 2005), manipulating the mask stimuli was effective in changing subjects’ perceptual state, which was the basis for the first analysis. Importantly, 13 out of 15 subjects also reported perceiving tool parts in Mondrian mask trials (17.9% of trials). This allowed the analysis of perceptual effects independently from changes in the physical stimulation, that is, the second analysis approach. Only 8 subjects reported perceiving whole tools in Mondrian mask trials (5.1% of trials). These data were not analyzed further. In trials with fully or partly visible tool stimuli, subjects’ 2AFC performance was >90%, irrespective of condition. Figure 2B shows that average 2AFC performance dropped to 62% in the invisible Mondrian trials but was significantly above chance level (t14 = 0 3.07, P = 0.008; t-test against 50%).
Tool-Specific fMRI-BOLD Activity Maps
We found robust and highly consistent preferential activation for the contrast Tools > Mondrians in the localizer experiment (Fig. 3). Tool-sensitive regions comprised the LO and pFS belonging to the lateral-occipital complex (Malach et al. 1995, 2002), as well as IPS and a region we tentatively defined as V3A/V7. Note that the contrast “Tools > Gratings” resulted in highly similar maps and that early visual cortex showed significantly stronger activation for mask stimuli than for tools. Thus, essentially, the entire constellation of high-order visual areas both belonging to the ventral as well as the dorsal stream showed a clear preferential activation to the tool images (Fang and He 2005).
Arguably such strong selectivity should reduce the “contamination” from activity evoked by the masks when the stimuli are presented simultaneously. If this is the case, the preferential tool activation should be maintained even when the tool images are dichoptically combined with the grating mask stimuli in the CFS paradigm. Importantly, in this condition, the tools remain fully visible (Fig. 2A). The corresponding fMRI results of the CFS experiment are shown in a statistical parametric map for the contrast “Tools visible > Gratings” (Fig. 4A). As can be seen, there was a significant activation primarily in ventral and dorsal stream areas largely corresponding to the regions, which were previously defined in the localizer experiment (Fig. 3). Next, we examined the impact of changing the mask stimulus from gratings to Mondrian stimuli. In this manipulation, the tool stimuli remained physically identical to the CFS grating mask condition, while their visibility was reduced (Fig. 2A). Our results show that under this “invisible” condition, the preferential tool activation was drastically reduced and not significant even at more lenient statistical thresholds (Fig. 4A, small inset). Could this difference in BOLD activation be attributed to a large differential activation to the Mondrian masks versus the grating patterns in high-order visual areas? A direct comparison of these 2 target-absent conditions revealed that this was not the case (Fig. 4B). As can be seen, the differential activation in high-order areas for the Mondrian versus grating masks was not significant. In contrast, low-order visual areas indeed showed a significant preferential activation to the Mondrian over the grating masks.
In order to obtain a more sensitive measure for the BOLD activation levels under the different conditions, we performed a ROI analysis. Tool-sensitive regions were identified in the localizer experiment by mapping the contrast Tools > Mondrians individually for each subject. We defined ROIs with a small (between 27 and 36 functional voxels across regions) and a large focus (between 247 and 653 voxels). The results for the small ROIs are presented in Figure 5 (see Supplementary Fig. 2 for large ROIs). We applied 2 different analysis approaches: comparing invisible and visible conditions in trials with Mondrian and grating masks (Fig. 5A) and comparing invisible and visible conditions in trials with identical Mondrian mask stimuli (Fig. 5B). Since no significant differences were found between the left and right hemisphere, the ROI data were collapsed across hemispheres (2 × 5 repeated measures analysis of variance [ANOVA] on parameter estimates in visible tool trials; main effect “laterality”: F1,14 < 1; interaction “laterality × ROI”: F4,56 = 1.19, P = 0.327; Greenhouse–Geisser corrected).
ROI Analysis (Grating and Mondrian Trials)
As can be seen in Figure 5A, the activity difference between the 2 mask types was confined to early visual cortex (t14 = 2.26, P = 0.040, two-sided paired t-test) and not significant in all high-order areas (all Ps > .200), thus confirming the fMRI mapping results presented above. Figure 5A shows strong differential activation in most ROIs: visible tools in grating trials were associated with larger activation than invisible tools in Mondrian trials. This effect was evident in high-order visual areas including ventral (LO: t14 = 10.09, P < 0.001; pFS: t14 = 9.52, P < 0.001) and dorsal (V3A/V7: t14 = 3.34, P = 0.005) stream. IPS showed a trend towards significance (t14 = 1.92, P = 0.076). Early visual cortex continued to reflect mask type and responded significantly stronger to invisible tool stimuli in Mondrian trials than to visible stimuli in grating trials (t14 = 3.13, P = 0.007). In agreement with Figure 4A, none of the ROIs showed a difference between invisible tools (in Mondrian trials) and the activity to Mondrian masks alone (early visual: t14 < 1; LO: t14 = −1.34, P = 0.202; pFS: t14 < 1; V3A/V7: t14 = −1.11, P = 0.287; IPS: t14 < 1). However, using MVPA of parameter estimates on a trial-by-trial basis (Cox and Savoy 2003; Kamitani and Tong 2005; Haynes and Rees 2006), we could reliably separate multivoxel response patterns to invisible tools from response patterns to Mondrian masks in area LO (see Supplementary Fig. 3).
ROI Analysis (Mondrian Trials with Identical Stimulation)
In order to overcome the stimulus confound caused by the 2 different mask types, we applied the second analysis approach in which visible and invisible conditions in identical Mondrian mask trials were compared. As can be seen from Figure 2A, subjects reported perceiving tool parts in 17.9% of Mondrian mask trials (13 out of 15 subjects reported tool parts). The difference between subjects’ objective 2AFC performance in “tool parts visible” trials and in trials in which tools were fully visible was only marginally significant (90% versus 98%, t12 = 2.01, P = 0.067, two-sided paired t-test), therefore allowing the comparison of a “tools visible” condition with a “tools invisible” condition, both of them occurring in Mondrian trials and thus under identical physical stimulation (Fig. 5B). In early visual cortex, the difference between visible and invisible trials showed a trend toward significance (t12 = 2.02, P = 0.067). All high-order visual areas, however, still showed robust visibility effects. In addition, we examined the activity profile in a putative hMT + ROI (see Materials and Methods); the results showed a weak visibility effect that was marginally significant (t12 = 2.11, P = 0.057).
To directly test for the effect of visibility and, at the same time, for a possible difference between the dorsal and ventral visual streams, we grouped the areas into a ventral (LO, pFS) and dorsal ROI (V3A/V7, IPS) and submitted the resulting parameter estimates to a 2 × 2 repeated-measures ANOVA with factors “visibility” and “visual stream.” The ANOVA yielded a significant main effect of “visibility” (F1,12 = 13.05, P = 0.004), but only a marginally significant “stream” effect (F1,12 = 3.91, P = 0.072). Since the visibility effect (i.e., the reduction in BOLD activity associated with the loss of perceptual awareness of the tool stimuli) was highly similar in the 2 streams, the “visibility × visual stream” interaction turned out to be not significant (F1,12 < 1). For a more detailed analysis beyond the stream segregation, we repeated the ANOVA with all ROI separately as factor “ROI” (2 × 4 repeated-measures ANOVA with Greenhouse–Geisser correction). The main effect of “ROI” turned out to be highly significant (F3,36 = 11.24, P < 0.001). Pairwise post-hoc comparisons revealed that BOLD activity in LO was larger than in all other high-order visual areas (all Ps < 0.005, two-sided paired t-tests). The “visibility × ROI” interaction, however, was not significant (F3,36 < 1).
Next, we analyzed fMRI activity in invisible Mondrian trials with correct and incorrect 2AFC responses (Fig. 5B). Early visual cortex was not affected by 2AFC response (t14 < 1). Data from high-order visual areas were submitted to 2 × 4 repeated-measures ANOVA with factors “ROI” and “response” (correct/incorrect). The main effect of “response” (F1,14 = 2.24, P = 0.157) as well as the “ROI × response” interaction (F3,42 = 1.27, P = 0.296) turned out to be not significant. Exploratory paired t-tests revealed a trend toward significance in pFS (t14 = 2.10, P = 0.054).
Finally, we aimed at estimating the influence of correct 2AFC performance in invisible trials. To that end, we split subjects into 2 groups based on their 2AFC performance in invisible trials, resulting in a “low 2AFC” and a “high 2AFC” performers group (Supplementary Fig. 4). The factor “group” was then added to the ANOVA of the visibility effect and also to the ANOVA of the response effect (correct/incorrect responses), resulting in 2 × 2 × 4 repeated-measures ANOVAs. Importantly, we found that the visibility effect was not modulated by subjects’ 2AFC performance (interaction “group × visibility”: F1,5 < 1). On the other hand, there was a larger differential BOLD response in invisible correct versus incorrect trials for the “high 2AFC” group, resulting in a marginally significant “group × response” interaction (F1,6 = 4.29, P = 0.084).
Visibility Effects in Physically Identical Trials
Although these results show that a significant visibility effect can be observed within the “tools + Mondrians” condition, comparing the tool exemplars in the “tools invisible” and “tools parts visible” conditions revealed a slight difference. While all 12 exemplars of tool stimuli were fully suppressed and thus invisible more or less equally often, certain tool exemplars were more likely to be reported as tool parts. Could it be that this difference in exemplar composition caused the observed activation difference between the “tools invisible” and “tool parts visible” conditions? To examine this possibility, we split all “tools visible” trials into 2 subsets, individually for each subject: subset “parts” contained all trials in which tool stimuli were presented that were sometimes reported as “tools parts visible” in the Mondrian condition, subset “non-parts” contained all tool stimuli which were never reported as parts and remained fully invisible in the Mondrian condition. We found no significant activity difference between the 2 subsets (Supplementary Fig. 5).
The main outcome of our study is that both dorsal and ventral stream cortical regions that were selectively activated by tool images showed a significant and similar link between BOLD activity and perceptual awareness. Importantly, the visibility effect was significant when identical physical presentation parameters were used (Fig. 5B). Furthermore, tool stimuli that were more likely to be perceived (by breaking suppression) were not associated with stronger BOLD activity levels compared to tool stimuli that never broke suppression under identical masking conditions (Supplementary Fig. 5). Thus, it is highly unlikely that our results could be explained by physical differences between the visible and invisible conditions during the CFS experiment. Areas in which we observed the association of visibility and BOLD activation included ventral stream object selective areas, such as pFS in the ventral branch of the LOC (Grill-Spector et al. 1999), LO as an intermediate area in the dorsal branch of the LOC (Hasson et al. 2002), as well as dorsal stream regions, such as V3A/V7 and IPS (Tootell et al. 1996; Grill-Spector and Malach 2004).
Could our results be due to top–down attentional effects? For example, it could be argued that subjects were more aroused upon detecting the task-relevant tool targets and that this global arousal was the cause of the enhanced activations we observed during perceptual awareness (Corbetta and Shulman 2002). One result of our study clearly speaks against this scenario, namely that we did not—under identical stimulation conditions—observe a positive visibility effect for tools in early visual cortex.
Our finding of a tight link between BOLD activation and target visibility was based on subjective reports of visibility. It could be argued that such reports may be an inaccurate measure of the perceptual state of the observers. For example, it could be the case that a subset of subjects adopted a particularly conservative subjective criterion and therefore tended to report that they failed to see the target even when some target percept broke through the mask. Such high criterion could explain the fact that a number of subjects showed a 2AFC performance above chance level even in “invisible” trials. However, our analysis which was aimed at examining the impact of such criterion shifts on the BOLD visibility effect failed to reveal a significant difference between “low 2AFC” and “high 2AFC” performers (see Supplementary Fig. 4). We also found that subjects could reliably detect parts of tool images when they were monocularly superimposed on the Mondrian masks. It should be noted that failures to report a perceptual event would simply weaken the BOLD visibility effect and hence should go against the observed findings. Thus, we can safely conclude that although differences in subjective criterion were indeed likely to occur in our subject population, they did not have a significant impact on the main outcome of the study, that is, the finding of significant enhancement in BOLD responses associated with target visibility.
When using different mask stimuli (gratings versus Mondrians) to generate changes in target visibility, our results appear to differ from previous brain imaging studies showing preserved activation to invisible tools in dorsal stream areas during CFS (Fang and He 2005). There are, however, substantial differences between the studies’ designs, including the presentation of masks in the visible condition and the procedure of behavioral report. In the experiments of Fang and He, subjects were asked to report if they perceived any shape or object after functional scans of 260 s duration in which target stimuli were rendered invisible and the main task was to detect an occasional size change of the fixation point. In contrast to this “offline” assessment of perceptual awareness, we used a trial-by-trial (“online”) temporal 2AFC detection task with the aim to better exclude residual visibility. Trial-by-trial detection is likely to be a more conservative test for visibility since subjects can use any low level feature difference between target and masks to detect the target’s presence. As it has been shown previously (Avidan et al. 2002; Lerner et al. 2002), high-order visual areas such as the LOC are extremely sensitive even to poorly visible low contrast images or object parts.
It is important to emphasize that our results do not demonstrate a complete abolition of neuronal activity when stimuli become invisible. Indeed, using a multivariate pattern classifier approach (Kamitani and Tong 2005; Haynes and Rees 2006), we could show that residual activity was still present in area LO under conditions of target invisibility, similar to findings from a recent CFS study (Sterzer et al. 2008). Thus, it is indeed possible that subliminal processing, such as priming effects, may utilize this low level activity to affect behavior. Such low level activity in dorsal stream areas may explain the recent demonstration of category-specific priming effects reported for tool images during CFS (Almeida et al. 2008, 2010). However, it is important to clarify that these results are specifically related to subconscious processing of invisible targets, while our study was aimed at examining the change in activity when moving from a visible to a nonvisible condition.
Summing up, our results show a consistent relationship between perceptual awareness and BOLD activation in both ventral “vision-for-perception/What” stream and dorsal “vision-for-action/Where” stream areas (Ungerleider and Mishkin 1982; Milner and Goodale 1995) of the human brain. Crossing the perceptual threshold from invisible to visible perception was linked to a significant increase in BOLD activation in all target-selective high-order visual areas. Thus, our data are in line with recent psychophysical, neuropsychological, and neuroimaging findings challenging a strong perceptual-motor dissociation in the visual system (Franz and Gegenfurtner 2008; Konen and Kastner 2008; Cardoso-Leite and Gorea 2010; Schenk and Mcintosh 2010) and support the notion that the link between neuronal activity and visual awareness is stream-invariant throughout high-order human visual cortex.
G.H. was supported by a Minerva fellowship (Max Planck Society). R.M. was supported by I.S.F. Bikura, and Mark Scher’s estate grants.
The authors thank E. Okon for technical assistance. G.H. would like to thank Martin Hebart and Lisandro Kaunitz for helpful discussions on the paradigm. Conflict of Interest: None declared.