Abstract
The human visual system consists of multiple topographic maps that extend from the early visual cortex (EVC) along the dorsal and ventral processing streams. Responses to illusory shapes within these maps have been demonstrated in the ventral stream areas, in particular the lateral occipital complex (LOC). Recently, the intraparietal sulcus (IPS) of the dorsal stream has been linked to the processing of illusory shapes defined by motion. It remains unclear whether the topographically organized parietal areas also respond to stationary illusory shapes, which would suggest their generic role in representing illusory content. In the current study we measured brain responses using fMRI while 30 human participants (12 male) observed flickering inducers around the fixation task. The inducers either formed an illusory diamond in the center, a triangle in the left or right hemifield, or were inverted such that no illusory figure was formed. We compared responses of parietal regions IPS0-IPS5 and SPL1 to each illusory figure with the nonillusory condition. To determine the role of attentional modulation on illusory shape responses we manipulated the difficulty of the fixation task. Our results show that all IPS areas responded to illusory shapes. The more posterior areas IPS0–IPS3 additionally displayed a preference toward contralateral shapes, while the more anterior areas IPS4 and IPS5 showed response attenuation with increased task difficulty. We suggest that the IPS can represent illusory content generated not only by moving, but also by stationary stimuli, and that there is a functional dissociation between attention-dependent anterior and spatially specific posterior topographic maps.
SIGNIFICANCE STATEMENT The traditional view of the ventral visual pathway being solely responsible for representation of objects has recently been challenged by demonstrating illusory shape representation within the dorsal visual pathway with moving bistable stimuli. Our results provide evidence for the dorsal stream contribution to representing not only moving, but also stationary illusory shapes. Our results also show a functional subdivision along the topographic maps, with spatially specific shape responses in the more posterior, and attention-dependent responses in the more anterior areas. These findings have implications for our understanding of the relationship between attention and grouping in healthy individuals and neuropsychological patients. Furthermore, IPS areas should be considered in theoretical accounts and models of how subjective content is generated in the brain.
Introduction
One of the fundamental principles of brain organization is topography. Each sensory modality contains multiple topographic maps, which are considered to be central to information processing within the respective modality (Kaas, 1997; Silver and Kastner, 2009). Functional magnetic resonance imaging of the visual system revealed the visual topographic organization far beyond the early visual cortex (EVC). Currently, >30 topographic maps have been identified across the visual processing hierarchy, extending from subcortical visually responsive areas (Schneider et al., 2004; DeSimone et al., 2015) and the EVC (Sereno et al., 1995; Engel et al., 1997), up to the parahippocampal areas along the ventral stream and the frontal eye fields along the dorsal stream (Wang et al., 2015). The exact functional role of these maps remains a topic of debate (Wandell et al., 2007).
Previous studies have shown that topographic maps in the EVC can represent not only simple sensory features (e.g., orientation, color, motion direction), but also more complex illusory content, i.e., aspects of perception than cannot be explained by the bottom-up input. For example, the EVC shows responses not only to physical, but also to illusory contours (von der Heydt et al., 1984; Larsson et al., 1999; Zhou et al., 2000; Lee and Nguyen, 2001; Maertens et al., 2008). Furthermore, it shows response enhancement within the topographic representation of the illusory surface and suppression at the location of the inducers and the background (Kok and de Lange, 2014; Kok et al., 2016; Grassi et al., 2017). This complex response pattern in the EVC is thought to be a result of feedback modulation from higher-level areas, which also respond to illusory contours (Mendola et al., 1999; Murray et al., 2002; Anken et al., 2018).
According to the two-streams hypothesis, the perception of shapes, including illusory shapes, is a function of the ventral stream (Haxby et al., 1991; Goodale and Milner, 1992). Indeed, multiple studies have demonstrated the involvement of the lateral occipital complex (LOC) in the perception of not only real, but also illusory shapes (Stanley and Rubin, 2003; Murray et al., 2004; de-Wit et al., 2009; Chen et al., 2020), making it a candidate area for providing the feedback signal to early visual areas (Fang et al., 2008; Wokke et al., 2013). However, the strict dichotomy between the ventral and the dorsal stream in the context of object perception is being continuously challenged (Freud et al., 2016). Recently, several studies reported activity in the posterior parietal cortex (PPC) and, more specifically, in the intraparietal sulcus (IPS) of the dorsal stream during perception of illusory shapes (Zaretskaya et al., 2013; Zaretskaya and Bartels, 2015; Grassi et al., 2018). These studies used bistable stimuli which can be interpreted either as an illusory global shape or as local nonillusory elements based on the same sensory input. Crucially, the illusory shapes investigated in these studies were defined by motion cues, which are expected to be processed along the dorsal stream. The PPC is known to contain multiple attention-defined topographic maps (Sereno et al., 2001; Schluppeck et al., 2005; Silver et al., 2005; Swisher et al., 2007; Konen and Kastner, 2008), which likely overlapped with previously reported activations. It is therefore unclear whether the topographic areas of the PPC also respond to stationary illusory shapes and whether these responses depend on attention, which illusory shapes are known to capture (Senkowski et al., 2005; Kimchi et al., 2016).
In this study, we were interested in whether parietal topographic areas respond to stationary illusory Kanizsa shapes (Kanizsa, 1976) and how these responses are modulated by top-down attention. We conducted an fMRI study in which we measured brain responses of human participants while they viewed Kanizsa figures and performed a central fixation task. By independently manipulating the figure location and task difficulty, we were able to dissociate the shape-related from the attention-dependent parietal activity.
Materials and Methods
Participants
Thirty healthy volunteers (12 male), aged 19–31 years (M = 23.53 ± 2.92 SD), participated in the experiment. We performed an a priori power analysis using GPower 3.1.9.7 (Faul et al., 2007) to determine the sample size, based on a previous fMRI study involving illusory content within the PPC (Grassi et al., 2018). Based on the smallest effect size reported within the PPC (Cohen's d = 0.59), with an α level of 0.05 and 80% power, at least 25 participants would be required for a two-tailed paired t test. Thus, our sample was sufficient to test the hypotheses of this study. All but one of the participants were right-handed. All participants had normal or corrected-to-normal vision and did not take any psychotropic medications at the time. Before the experiment, the participants were prescreened for MRI contraindications and had signed a written informed consent form. The study was conducted according to the Declaration of Helsinki and the study's procedure and protocols were approved by the ethics committee of the University of Graz. The current study represents a part of a bigger project, which was preregistered before data collection (https://aspredicted.org/ke4b9.pdf).
Stimuli and task
Visual stimuli were generated using MATLAB R2019b (MathWorks, 2019) and Psychophysics Toolbox 3 (Brainard, 1997; Pelli, 1997; Kleiner et al., 2007) on a Linux computer (Ubuntu 18.04 LTS). The stimuli were presented in the MR scanner room on a 32″ γ-corrected monitor (resolution: 1920 × 1080 pixels, refresh rate: 60 Hz, maximum surface luminance: 405 cd/m2, Nordic NeuroLab), which was viewed via a mirror mounted above the head coil. The total distance from the display to the eyes of the participants was 143.5 cm. Before entering the scanner room, we ensured sufficient visual acuity of each participant with the Snellen chart. We gave the participants a thorough instruction regarding the detection task (see below) and advised them not to move their head during the entire MRI acquisition.
In each experimental block, we presented one of the four possible stimulus configurations (“no illusion,” “diamond,” “left triangle,” and “right triangle”; Fig. 1A) on the screen for a total of 12 s. Each stimulus consisted of four black (RGB (Red, Green, Blue): 0, 0, 0) pac-man inducer combinations, which were presented against a gray background (RGB: 128, 128, 128). Each inducer consisted of a circle shape with a cut out wedge and subtended 4.08° of visual angle. The inducers were arranged in a diamond shape around the central fixation circle at an eccentricity of 5.77° of visual angle in four possible configurations that yielded four experimental conditions. In the “no illusion” configuration, all inducers were oriented with the 90° cut out wedge facing outwards, thus not creating an illusory shape in the center. In the “diamond” configuration, the inducers were aligned to form a centrally positioned illusory diamond shape. In the “left triangle” configuration, top and bottom inducers had a 45° cut out wedge and were rotated such that an illusory triangle formed within the left visual hemifield. The “right triangle” configuration was a vertically flipped version of the left triangle configuration resulting in an illusory triangle within the right visual hemifield. To account for adaptation in the visual cortex, the stimuli were continuously flickering between pac-man inducers and full circles every 500 ms throughout each experimental block.
Stimulus types and experimental paradigm. A, Stimuli arranged in four possible inducer configurations: (1) no illusion, (2) diamond, (3) right triangle, and (4) left triangle. B, Sequence of events within one example experimental block. The example block represents diamond inducer configuration continuously flickering around the easy detection task. Participants fixated centrally positioned symbols and responded to changes in symbol color; t, block duration. The proportions in this figure are altered for illustration purposes.
To ensure central gaze fixation and equal distribution of attention across the stimulus conditions, participants performed a central detection task. A stream of orange (RGB: 153, 80, 0) and blue (RGB: 0, 107, 209) symbols subtending 0.6° of visual angle was presented at the center of the display. A pseudo-randomly chosen symbol was presented one at a time from a fixed set of 26 possible symbols (!”$%&/()=?{[]}\+*∼#-:;><@̂). Each symbol was presented for 12 frames (∼200 ms) on the screen. The participants were instructed to always attend to the centrally positioned stream of symbols. Participants had to perform one of the two tasks, which remained the same throughout one run. In the first task (“easy”), participants were instructed to press a button on the response box in their hand each time the symbol color changed to blue (5% of all symbols), regardless of the symbol identity. In the second task (“hard”), the participants were instructed to press the button each time the plus symbol (“+”) appeared on the screen (5% of all symbols), regardless of its color. On average, target symbols appeared around three times per block in either task. The symbol stream appearance and target frequency were thus identical in the two task conditions, and only the task demands varied. The sequence of events within one block can be seen in Figure 1B.
Each of the four stimulus configurations (block) was presented six times per run. The sequence of experimental blocks was determined for each run by generating a first-order counterbalanced sequence of conditions (Brooks, 2012). Baseline blocks (12 s each) were added after every fourth experimental block, with a total of six baseline blocks per run. During the baseline blocks, a central fixation point was presented in place of the symbol stream and nonflickering black circles were presented in place of pac-man inducers. In total, there were 30 blocks in each run. The total duration of one run was 360 s (excluding the four initial dummy scans, which were subsequently discarded). Between each run there was a short (1–2 min) break, during which the participants could rest their eyes. In total, one scanning session included eight experimental runs (four runs per task condition). The order of runs with easy and hard detection tasks was pseudorandomized and counterbalanced between the participants. Additionally, after the main experiment we collected an anatomical MRI scan for each participant. The total duration of the whole session was around 60 min.
After the MRI acquisition, participants received a questionnaire with six questions for assessment of their subjective visual experience during the experiment. The participants rated the perceived difficulty of the tasks and the subjective strength of illusory figures on a Likert scale ranging from 1 to 6. The questionnaire included the following questions: “Please rate your subjective perception of the difficulty of the,” Q1: “symbol task”; Q2: “color task” (1, not difficult at all; 6, very difficult). “During the experiment, how strongly did you perceive the following illusory figures?” Q3: “illusory diamond during the symbol task”; Q4: “illusory diamond during the color task”; Q5: “illusory triangles during the color task”; Q6: “illusory triangles during the symbol task” (1, not at all; 6, very strongly).
MRI data acquisition
Neuroimaging data were acquired using a 3T Siemens MAGNETOM Vida scanner (Siemens Healthineers) with a 64-channel head coil. Functional MRI scans were acquired using the simultaneous multislice (SMS) accelerated echo-planar imaging (EPI) sequence using T2*-weighted blood oxygenation level-dependent (BOLD) contrast (58 axial slices, TR = 2000 ms, TE = 30 ms, FOV = 220 mm, flip angle = 82°, voxel size = 2.0 × 2.0 × 2.0 mm, SMS factor 2). To compensate for distortion correction during preprocessing, we also acquired one EPI image with opposite phase encoding direction (P ≫ A). Additionally, a high-resolution anatomical scan was acquired using the T1-weighted (T1w) MPRAGE sequence (TR = 2530 ms, TE = 3.88 ms, TI = 1200 ms, voxel size: 1.0 × 1.0 × 1.0 mm, GRAPPA factor 2).
MRI data preprocessing
Before any data preprocessing, we removed facial identifiers from all participants with pydeface 2.0.0 (Gulban et al., 2019). Afterwards, we visually inspected the quality of MRI images with MRIQC 0.16.1 (Esteban et al., 2017). Anatomical and functional MRI data were preprocessed using fMRIPrep 20.2.3 (Esteban et al., 2019), which is based on Nipype 1.6.1 (Gorgolewski et al., 2011). The main preprocessing steps are summarized below.
Anatomical data preprocessing
The T1w image was corrected for intensity nonuniformity with ANTs N4BiasFieldCorrection function and used as T1w-reference, which was then skull-stripped. CSF, white-matter (WM), and gray-matter (GM) brain tissue segmentation was performed on the brain-extracted T1w using fast function from FSL 5.0.9 (Woolrich et al., 2009). Brain surfaces were reconstructed using FreeSurfer recon-all function (Dale et al., 1999).
Functional data preprocessing
First, a reference volume and its skull-stripped version were generated using a custom methodology of fMRIPrep. For each BOLD run the following preprocessing transformations were performed: A B0-nonuniformity map was estimated based on the EPI with opposite phase-encoding direction using 3dQwarp from AFNI 20160207 (Cox, 1996). Based on the estimated susceptibility distortion, a corrected EPI reference was calculated for more accurate co-registration with the anatomic reference. The BOLD reference image was co-registered to the T1w reference image using FreeSurfer's bbregister (Greve and Fischl, 2009). Head-motion parameters with respect to the BOLD reference image were estimated before any spatiotemporal filtering using mcflirt (FSL 5.0.9). Slice-time correction was performed using 3dTshift from AFNI 20160207. These transforms were concatenated and applied in one interpolation step. The BOLD-time series were resampled to the fsaverage surface using mri_vol2surf (FreeSurfer). For the region of interest (ROI) general linear model (GLM) analyses (see below), we used unsmoothed preprocessed BOLD runs. For the whole-brain GLM analyses, each preprocessed BOLD run was smoothed along the cortical surface with a 2D Gaussian kernel of 5-mm full-width at half-maximum (FWHM) using mri_surf2surf (FreeSurfer).
Data analysis
First-level GLM analysis
All first-level GLM analyses were performed using FsFast (FreeSurfer Functional Analysis Stream). For each participant, we performed a standard GLM analysis, with each experimental condition modeled as a unique regressor convolved with a canonical hemodynamic response function (HRF). In contrast to early visual areas, areas of the superior parietal cortex are known to show a transient BOLD response pattern (Steinman et al., 1997; Yantis et al., 2002; Zaretskaya et al., 2013), which we confirmed for our paradigm with an additional finite impulse response GLM analysis. Thus, we modeled experimental conditions in the GLM as 0.1 s events at block onsets rather than full 12-s block durations. There were eight unique condition regressors in total, with the first four modeling stimulus configurations under the hard task (no illusion, diamond, left triangle, right triangle) and the next four under the easy task (no illusion, diamond, left triangle, right triangle). Additionally, run-specific effects as well as slow signal drifts were added as nuisance regressors. Illusory shape activity was quantified by calculating six contrast estimates (CES). The latter were defined as the difference of β estimates between each illusory shape condition (diamond, left triangle and right triangle) and the “no illusion” condition, separately for each task type. The resulting contrasts thus represent illusory shape-related activity after accounting for the overall effects of task, which should cancel in the task-specific subtraction. The GLM was fit to the unsmoothed data for the ROI analysis. All statistical analyses and plotting were performed in RStudio 4.1.2 (RStudio Team, 2021). Where appropriate, error bars were plotted with confidence intervals (CIs) adjusted for within-subject design using the Cousineau-Morey method (Morey, 2008).
Definition of ROIs and ROI analysis
Seven topographic areas of the PPC (IPS0, IPS1, IPS2, IPS3, IPS4, IPS5, and SPL1) were defined using the maximum probability maps (MPM) of the surface-based probabilistic atlas by Wang et al. (2015), resampled to FreeSurfer fsaverage standard space (Benson and Winawer, 2018). Since our study is hypothesis-driven, an a priori definition of the parietal areas via probabilistic MPMs is a recommended method for inferring the individual's anatomic locations within the brain (Eickhoff et al., 2006). Furthermore, the predefined areas in a standard surface space allow for unbiased and direct comparison with data from other studies (Wang et al., 2015). Figure 2A illustrates the location of ROIs on an inflated cortical surface. The number of vertices for each parietal ROI were: IPS0 = 1055, IPS1 = 900, IPS2 = 899, IPS3 = 604, IPS4 = 148, IPS5 = 53, and SPL1 = 232 vertices in the left hemisphere and IPS0 = 835, IPS1 = 675, IPS2 = 624, IPS3 = 805, IPS4 = 160, IPS5 = 18, and SPL1 = 295 vertices in the right hemisphere. We extracted mean values over vertices within each ROI for each contrast estimate map of each participant. The per-ROI means were then used to conduct statistical analysis. Since we do not expect a lateralization of the illusory diamond response, we averaged mean shape response over both hemispheres. For the illusory triangle response, we averaged the two contralateral triangle values (i.e., the left hemisphere value for the contrast “right triangle > no illusion” and the right hemisphere value for the contrast “left triangle > no illusion”) and the two ipsilateral values (i.e., the left hemisphere value for the contrast “left triangle > no illusion” and the right hemisphere value for the contrast “right triangle > no illusion”).
The following statistical tests were conducted to investigate the illusory shape response in the parietal ROIs. First, we tested the presence of illusory shape responses in each parietal ROI by testing for non-zero contrast differences (“diamond > no illusion,” “contralateral triangle > no illusion,” and “ipsilateral triangle > no illusion”) during the easy task with a two-tailed one sample t test. Second, we tested the spatial specificity of the illusory shape response within each ROI by comparing contralateral and ipsilateral responses during the easy task with a two-tailed paired samples t test. Third, we tested whether higher task demands affected illusory diamond responses by directly comparing diamond responses for the two task difficulties with a two-tailed paired samples t test within each ROI. Finally, we tested whether higher task demands affected illusory triangle response by performing a 2 × 2 repeated-measures ANOVA within each ROI, with laterality (contralateral and ipsilateral) and task (easy and hard) as factors. All t tests and ANOVA factors were corrected for multiple comparisons using Holm–Bonferroni method (Holm, 1979) and the corrected statistics are labeled in the figures. All raw p-value counterparts of the corrected statistics are also included either directly in the text or in a table.
Additional whole-brain group analysis
To confirm previous findings on illusory shape response in the ventral and early visual areas we performed an additional whole-brain group analysis. The analysis was performed using FreeSurfer mri_glmfit command with -wls flag (weighted-least-squares), testing for non-zero differences. To test for the presence of the illusory shape response, we compared all illusory shape conditions with the no illusion condition during the easy task (C = [0 0 0 0 −1 0.333 0.333 0.333]). To test for spatial specificity of shape responses, we compared the left triangle with the right triangle condition during the easy task (C = [0 0 0 0 0 0 1 −1]). Finally, to test the impact of task on the illusory shape responses, we compared illusory shape response in easy and in hard task (C = [−0.5 0.167 0.167 0.167 0.5 −0.167 −0.167 −0.167]). Correction for multiple comparisons was performed using precomputed z-based Monte Carlo simulation with a cluster-wise threshold at p < 0.05 (Hagler et al., 2006) and a cluster-forming threshold at p < 0.001 to minimize false positive rates (see Greve and Fischl, 2018). Finally, we adjusted p-values with a Bonferroni correction for two spaces (left and right hemisphere). The surviving clusters were classified according to the Desikan–Killiany cortical parcellation (Desikan et al., 2006). Additionally, we calculated the overlap between the clusters and areas of the visual probabilistic atlas (Wang et al., 2015).
Behavioral data analysis
In our behavioral data analysis, we ensured that our manipulation of the task difficulty was successful. The difficulty of both detection tasks was verified with the average hit rate and the participants' subjective rating of the task difficulty. The hit rate in each run was calculated as a total number of correctly detected target instances (target color or target symbol, depending on the task), divided by the total number of target instances and multiplied by 100. Values from the runs with the same task type were averaged. We compared the average hit rates between the easy and the hard task and the subjective ratings of the task difficulty by performing a two-tailed paired samples Wilcoxon signed rank test with continuity correction. To determine whether differences in illusory shape response between tasks observed in fMRI can be explained by differences in subjective experience of the illusory shapes, we also compared participants' ratings of their subjective illusion perception between tasks with two Wilcoxon signed rank tests, one for the diamond and one for the triangle condition.
Our central task was designed to manipulate subjects' attentional demand and test its impact on the processing of illusory shapes. However, the processing of the central targets can be impaired by the presence of the task-irrelevant salient illusory figures in the periphery, yielding an opposite effect (Chen et al., 2019). To ensure that our task manipulation fulfilled its purpose and was equally efficient across different inducer configuration, we additionally analyzed behavioral performance separately for each stimulus configuration and each task. We performed a nonparametric 2 × 4 repeated-measures aligned ranks transformation (ART) ANOVA (Wobbrock et al., 2011) using the ARTool R package (Kay et al., 2021), with the factors of task (easy, hard) and stimulus configuration (no illusion, diamond, left triangle, right triangle).
Additional control analyses
The above behavioral data analysis confirmed the differences in hit rates between the tasks, which means that participants pressed more often during the easy compared with the hard task (because of higher hit rates). To account for the potential role of motor response in explaining our fMRI results, we performed two additional analyses. First, we performed three nonparametric 2 × 4 repeated-measures ART ANOVAs with the factors of task (easy, hard) and stimulus configuration (no illusion, diamond, left triangle, right triangle) for the average frequency of hits, frequency of misses and frequency of false presses per block. Second, we performed a modified GLM analysis with three additional nuisance regressors containing (1) target hits (modeled as events at button press onset), (2) false presses (modeled as events at button press onset), and (3) missed targets (modeled as events at target onset). Each of these additional regressors was modeled with a duration of 0.1 s. This modified GLM was used to repeat the ROI analysis as described in the main analysis section, but with motor response accounted for. We then compared the overall pattern of results found with this modified GLM with our original ROI results.
Our final control analysis was related to the potential effect of the stimulus flicker itself. The illusory stimuli in our experiment were not defined by motion, but they also were not entirely static because of flicker that we added to the inducers to prevent adaptation of the visual system to the stimulus. Previous research has shown that flickering stationary stimuli can activate area hMT (Tootell et al., 1995), and that areas hMT and MST are more activated by a flickering Kanizsa figure, compared with a flickering no illusion condition (Goebel et al., 1998). To determine whether parietal illusory shape responses in our experiment are related to the flicker, which should engage other dorsal stream areas including the motion-sensitive hMT and MST, we derived the location of these areas using procedures identical to the main analysis (Wang et al., 2015) and tested whether they respond higher to illusory shapes compared with the no illusion condition.
Data and code availability
Video examples of the stimuli, detailed MRI preprocessing steps, ROI and group-level fMRI results, behavioral results, as well as control analyses and relevant scripts are available at the OSF project repository: https://osf.io/n9bjr/. Our institution's data protection policy currently does not allow publicly sharing native space MRI data, but we are able to share the data on direct request from individual researchers.
Results
Illusory shape responses
We first examined whether parietal topographic regions responded to illusory Kanizsa shapes in the easy task, which is comparable to a typical fixation task used in fMRI. We performed a one-sample t test for a non-zero illusory diamond response (difference between the illusory diamond and no illusion) in each ROI. Significant illusory shape response was present in areas IPS0, IPS1, IPS3, IPS4 and IPS5. SPL1 and IPS2 did not respond to the illusory diamond (Fig. 2B).
Illusory shape responses in parietal topographic maps (easy task). A, Location of the parietal areas from the probabilistic surface-based atlas (Wang et al., 2015), overlayed onto the right hemisphere of the inflated fsaverage cortical surface. Inset shows enlarged anterior areas IPS4 and IPS5. B–D, Illusory shape-related response during the easy task within the PPC (B, diamond, both hemispheres; C, triangle, contralateral hemisphere; D, triangle, ipsilateral hemisphere). Error bars represent 95% CIs; +p < 0.05, uncorrected; *p < 0.05, **p < 0.01, Holm–Bonferroni corrected (7 tests).
The surface of the illusory diamond was occupying the central visual field around the fixation task. It was therefore easy to perceive and potentially captured attention. Thus, we also tested whether illusory shape responses were present in the parietal areas for the triangles which were placed peripherally relative to the central task. All six ROIs of the IPS (IPS0–5) responded to the contralateral illusory triangle during the easy task. SPL1 was not activated by the illusory triangle (Fig. 2C). Interestingly, IPS4 and IPS5 also responded to the illusory triangle presented to the ipsilateral hemisphere (Fig. 2D). Statistical details about illusory shape responses are presented in Table 1.
Illusory shape response (shape > no illusion, easy task)
Spatial specificity of the illusory shape response
Next, we tested whether illusory shape responses were spatially specific by directly comparing responses to the contralateral with responses to the ipsilateral triangle using two-tailed paired samples t tests for each ROI. Contralateral responses were stronger in IPS0 (t(29) = 6.397, p < 0.001, d = 1.17), IPS1 (t(29) = 4.090, p < 0.001, d = 0.747), IPS2 (t(29) = 3.203, p = 0.003, d = 0.585), IPS3 (t(29) = 3.461, p = 0.002, d = 0.632), and SPL1 (t(29) = 2.109, p = 0.044, d = 0.385). We found no hemispheric difference in illusory triangle shape response in IPS4 (t(29) = 1.724, p = 0.095, d = 0.315) or IPS5 (t(29) = 0.964, p = 0.343, d = 0.176). The comparison between contralateral and ipsilateral responses is shown in Figure 3.
A–G, Spatial specificity of illusory shape responses (easy task). Error bars represent 95% CIs (Morey, 2008); +p < 0.05, uncorrected; *p < 0.05, **p < 0.01, ***p < 0.001, Holm–Bonferroni corrected (7 tests).
Attentional modulation of the illusory shape response
Finally, we tested whether the observed illusory shape responses were modulated by task difficulty by comparing responses in the easy task with responses in the hard task. For the illusory triangle, a 2 × 2 repeated-measures ANOVA with the factors of laterality (ipsilateral, contralateral) and task (easy, hard) revealed that shape-related response in IPS0, IPS1, IPS2, and IPS3 showed a preference for the contralateral stimulus presentation (main effect of laterality IPS0: F(1,29) = 48.628, p < 0.001, ηp2 = 0.626; IPS1: F(1,29) = 21.705, p < 0.001, ηp2 = 0.428; IPS2: F(1,29) = 10.908, p = 0.003, ηp2 = 0.273; IPS3: F(1,29) = 14.632, p < 0.001, ηp2 = 0.335; see Fig. 4A–D). Crucially, these lateralized illusory triangle responses were not attenuated by the hard task and there was no interaction between task and laterality. In contrast, IPS4 showed an attenuation of the illusory triangle response during the hard task compared with the easy task (main effect of the task F(1,29) = 7.726, p = 0.009, ηp2 = 0.21; Fig. 4E). Crucially, this attenuation occurred regardless of laterality (no main effect of laterality) and there was no interaction between task and laterality. IPS5 revealed a trend in the same direction as IPS4 (Fig. 4F), but it did not reach significance (main effect of task F(1,29) = 3.795, p = 0.061, ηp2 = 0.116). SPL1 did not show any significant effects. Figure 4G shows the graphical representation of the significant main effects for each IPS area.
Spatial specificity and task effects in the IPS. A–D, Main effect of laterality in areas IPS0–3. E, F, Main effect of task in area IPS4 and trend of main effect of task in IPS5. Error bars represent 95% CIs (Morey, 2008); ∼p = 0.061, +p < 0.05, uncorrected; *p < 0.05, **p < 0.01, ***p < 0.001, Holm–Bonferroni corrected (21 effects). G, Visualization of the location of the main effects of laterality and task within the PPC on the right inflated cortical surface. The circle size is proportional to the respective effect size.
To determine whether higher attentional demands reduced the illusory diamond response, we compared the responses across different task difficulties with a two-tailed paired samples t test for each ROI. However, we found no effect of task on the illusory diamond response in any of the seven parietal ROIs (IPS0: t(29) = 0.961, p = 0.345, d = 0.175; IPS1: t(29) = 0.656, p = 0.517, d = 0.120; IPS2: t(29) = 0.544, p = 0.590, d = 0.099; IPS3: t(29) = 1.050, p = 0.303, d = 0.192; IPS4: t(29) = 1.286, p = 0.209, d = 0.235; IPS5: t(29) = 0.237, p = 0.814, d = 0.043; SPL1: t(29) = 0.242, p = 0.810, d = 0.044).
Behavioral results
Our behavioral results confirmed that our manipulation of task difficulty was successful. A Wilcoxon signed rank test showed a significant difference in hit rates between the two detection tasks (Z = 4.772, p < 0.001, r = 0.873), with the easy task runs having higher hit rates than the hard task runs (Fig. 5A). Corresponding effects were found in another Wilcoxon test for the subjective difficulty ratings as well, with the easy task being perceived as easier compared with the hard task (Z = −3.818, p < 0.001, r = 0.716; Fig. 5B). Another two separate Wilcoxon signed rank tests with continuity correction comparing subjective illusion strength revealed no difference in illusion strength between the tasks (diamond: Z = 0.133, p = 0.894, r = 0.076; triangle: Z = −0.619, p = 0.536, r = 0.111). Finally, we tested whether hit rates in each stimulus configuration and task variant differ from each other. A 2 × 4 ART ANOVA on the hit rate showed that the surrounding stimulus configuration did not change the difference in hit rates between the easy and the hard task. There was a significant main effect of the task F(1,232) = 139.682, p < 0.001, ηp2 = 0.376, but no effect of the stimulus configuration (p = 0.818) and no interaction (p = 0.723; Fig. 5C).
Objective and subjective measurements of task difficulty. A, Total mean hit rate in easy and hard detection tasks. B, Mean ratings of subjective task difficulty, rated on a six-point Likert scale (1, not difficult at all; 6, very difficult). C, Hit rate per stimulus configuration. dia, diamond; inv, no illusion; LT, left triangle; RT, right triangle. Error bars represent 95% CIs (Morey, 2008); ***p < 0.001.
Whole-brain results
To check whether our paradigm produces an expected pattern of brain activity outside of IPS, we additionally performed a whole-brain analysis that paralleled our ROI analysis. We found illusory shape-related responses during the easy task in both hemispheres in the EVC (V1–V3) and in areas hV4, V3b and LO1, confirming findings of multiple previous studies. Furthermore, 18 vertices within the left LO2 area also responded to the illusory shapes. Only one of the clusters that survived correction for multiple comparisons was located in the parietal cortex (cluster size: 205 vertices, peak MNI coordinates: [21.7 −57.9 56.6]). The cluster overlaps with the parietal area IPS3 (129 vertices, 62.9% of the cluster vertex size) and IPS4 (67 vertices, 32.68% of the cluster vertex size). Whole-brain group results with significant clusters for the shape response are shown in Figure 6A. For the sake of comparison with previous findings, we also compared the location of this parietal cluster with two parietal clusters related to illusory shape perception reported in Zaretskaya et al. (2013). 205 vertices of the “SPL” cluster from their study (16.69% of the total “SPL” vertex size) overlaps with our parietal cluster, and there is no overlap with the “aIPS” cluster from the same study.
Whole-brain group-level results (corrected for multiple comparisons). A, All illusory shapes > no illusion configuration, easy task. B, Left illusory triangle > right illusory triangle, easy task. Clusters of significant responses are shown on the inflated cortical surface of the fsaverage template. LH, left hemisphere; RH, right hemisphere.
Next, we tested for spatial specificity of the illusory shape-related response by contrasting illusory shape responses for the left triangle with illusory shape responses for the right triangle during the easy task. This time, we found positive clusters in the contralateral (right) hemisphere only, in the EVC (V1–V3), hV4, V3a, V3b, and LO1. Whole-brain group results with significant clusters for the spatially specific illusory shape responses are shown in Figure 6B.
Finally, we tested for attention modulation of the illusory shape response by contrasting all illusory shape responses in the easy task with illusory shape responses in the hard task. There were no surviving clusters for this analysis. All significant clusters from the illusory shape and spatial specificity analyses are shown in Table 2. Additionally, surface area overlaps of the significant clusters and the areas of the probabilistic visual atlas for illusory shape and spatially specific responses can be found in Table 3.
Significant clusters of illusory shape and spatially specific responses in the whole brain after correction for multiple comparisons
Whole-brain results overlap (positive clusters only) with visual topographic areas
Control analyses
Since the target frequency was the same in the easy and the hard task, but the hit rate was higher in the former, we tested whether our fMRI results can be explained by the overall differences in motor activity. We compared the average hit frequency, the average miss frequency and the average false alarm frequency per block, considering each stimulus configuration separately. We observed a significant main effect of task in a 2 × 4 ART ANOVA on the mean frequency of hits (F(1,232) = 106.707, p < 0.001, ηp2 = 0.315), mean frequency of misses (F(1,232) = 163.105, p < 0.001, ηp2 = 0.413) and mean frequencies of false presses F(1,232) = 47.449, p < 0.001, ηp2 = 0.17). Crucially, there was no main effect of the stimulus configuration or interaction in either of the analyses (all p-values ≥ 0.434). This result shows that the participants had on average more hits in the easy task and more misses and false presses in the hard task, but these responses were not influenced by the stimulus configuration. Since in our analysis we always subtract the no illusion condition from each illusory figure condition within the same task, the motor response differences across task should cancel out. It is therefore unlikely that differences in motor responses can explain our observed fMRI effects.
To rule out the effect of motor response on our fMRI results more directly, we performed a control analysis modeling hits, misses, and false alarms in the GLM as nuisance regressors. The results of this analysis showed that there are no substantial differences in the pattern of responses from our main analyses (data not shown). The only p-value that changed from significant to nonsignificant was the response of IPS5 in the ipsilateral triangle condition (changing from 0.034 to 0.053). We therefore conclude that the possible confounding effects of these events on the results were small to insignificant.
Finally, we also tested whether our stimulus, which contained a continuous flicker, engaged other classical dorsal stream regions such as hMT and MST. This analysis showed that neither hMT (t(29) = −0.271, p = 0.788, d = −0.05) nor MST (t(29) = 0.987, p = 0.332, d = 0.180) responded to the illusory diamond. Likewise, there was no contralateral triangle representation in these areas [hMT (t(29) = −1.365, p = 0.081, d = −2.492; MST (t(29) = 0.075, p = 0.94, d = 0.014)]. Overall, the illusory shape responses in the IPS in our study cannot be explained by an unspecific engagement of the dorsal stream as a whole.
Discussion
The results of our study show that IPS areas respond to illusory shapes even when they are not defined by motion. They also indicate that different parietal areas respond to illusory shapes in a different manner. The posterior areas show a spatially specific and attention modulation-independent response. On the other hand, the anterior areas show an attention-dependent response, which is not specific to illusory shape location. Our results thus also show a dissociation between the location-specific and attention-dependent activation during illusory shape perception along the posterior-anterior gradient within the IPS.
Stationary illusory shape responses in the parietal cortex
Several previous fMRI studies already reported illusory shape responses within the superior parietal cortex and the IPS, but those studies relied on motion-defined bistable stimuli which either did or did not produce an impression of a moving illusory shape (Zaretskaya et al., 2013; Grassi et al., 2016, 2018). A bistable paradigm has the advantage of fully matching the illusory and nonillusory conditions. However, it requires participants to track their own subjective experience of the stimulus, which inevitably couples illusory perception and attention. Because of this coupling, previous studies could test neither the topographic specificity of the shape responses, nor their dependence on attention. In the current study, highly controlled and task-irrelevant Kanizsa stimuli allowed us to overcome these constraints, dissociating the illusory figure from the allocation of attentional resources. We demonstrate that illusory content of the Kanizsa shapes is represented within the posterior maps IPS0–3, even when the shapes are task-irrelevant and attention is engaged elsewhere. This finding implies that parietal topographic areas play a general role in representing illusory content, regardless of the exact perceptual mechanism, be it motion or inducer alignment, that led to its generation.
An outstanding question remains about the exact illusory aspect of the Kanizsa shape driving the parietal activity. Subjective impression induced by a Kanizsa figure includes not only illusory contours, but also illusory surface, which, in turn, is perceived as brighter and closer to the observer (Kogo et al., 2014). In the current study, we focused on illusion-related activity as a whole, without differentiating between single illusory aspects. Future studies are needed to determine the unique contribution of each illusory feature to the IPS activity.
Spatial specificity of shape responses
Our study also revealed that the response of the posterior IPS regions (IPS0–3) is spatially specific. These areas respond stronger to the illusory triangles presented to the contralateral compared with the ipsilateral half of the visual field. Since each IPS map represents the contralateral visual field half, such specificity implies a coarse topographic specificity of illusory shape responses. The topography of the illusory shape responses and its independence from attention in these areas resembles the responses of early visual areas V1 and V2, which show a differentiated pattern of illusory surface enhancement and inducer/background suppression (Kok and de Lange, 2014; Grassi et al., 2017; but see de-Wit et al., 2012). Because of the relatively small surface area and much larger receptive field sizes of the IPS maps compared with V1/V2, it is challenging to show such fine-scale response differences with conventional fMRI. Future studies using high-resolution fMRI at ultrahigh field have the potential to test whether there is a distinction between illusory surface enhancement and inducer suppression (Zaretskaya, 2021).
Dissociation of attention-related and illusion-related response
Another major finding of our study is that the attentional effects were confined to the anterior map IPS4, with a trend in IPS5. These two maps showed a complete absence of responses to peripheral illusory triangles during the more difficult task which forced the allocation of more attentional resources to the fovea. Anterior IPS responses may thus reflect engagement of attentional resources regardless of what engages them. During the easy task illusory shapes may attract attention (Rauschenberger and Yantis, 2001; Senkowski et al., 2005; Kimchi et al., 2016), which leads to higher IPS4 activity for the illusory shape configuration. During the hard task, attention is engaged by a demanding task in the fovea, which leads to equal IPS4 activity for illusory and nonillusory surrounding configuration. The unspecific signaling of attentional engagement could also explain the lack of response lateralization in these areas.
The relationship between attention and grouping is a matter of a long-standing debate (Driver and Baylis, 1998; Treisman and Kanwisher, 1998), fueled in part by findings in patients with parietal extinction. In such patients, a shape-forming inducer configuration that spans both hemifields facilitates the identification of the inducer in the unattended hemifield, releasing it from extinction (Mattingley et al., 1997; Gögler et al., 2016; Conci et al., 2018; Nowack et al., 2021). In this context, parietal cortex is viewed as an area that accomplishes attentional selection by marking and grouping coherent inducers, which are thought to be represented at the lower levels of the visual hierarchy (Nowack et al., 2021). Our study shows that the posterior maps of the IPS themselves contain a representation of illusory shapes, even if they are task-irrelevant and even under high attentional load. In the context of patient findings this may mean that Kanizsa shape integration happens automatically and preattentively, with the corresponding neural representation propagating up to the levels of the posterior IPS areas, but that the conscious access to and the ability to report on these representations, as it is required in patient studies, is ultimately enabled by attention.
Contrary to the illusory triangles, we found no attentional modulation for the illusory diamond. One potential explanation is that in the case of illusory diamond the locus of attention and the location of the illusory figure coincide. Attentional resources dedicated to the central task could thus also be used to process the foveal illusory diamond surface (Bakar et al., 2008). On the other hand, a contour of the illusory triangle does pass through the central detection task. It could thus be argued that the locus of attention coincides with the illusory surface for triangles as well. An alternative explanation for differences in attentional modulation between diamond and triangles could be the difference in size or saliency of the two shapes. Further studies are needed to tackle the reason for this difference in fMRI effects.
Illusory shape response and predictive coding
Illusory shape responses in the EVC are frequently interpreted in the context of the predictive coding theory (Rao and Ballard, 1999; Kok and de Lange, 2015; Friston, 2018). According to this theory, the response in the topographic representation of the illusory surface increases because the feedback signal contains predictions that do not match with the bottom-up sensory input, resulting in a prediction error. At the same time, the response in the topographic representation of the inducers decreases because there is a match between the prediction and the sensory input (Kok and de Lange, 2014; Grassi et al., 2017). Our findings of the illusory shape-related responses far beyond the EVC are consistent with the hierarchical nature of predictive inference engaging multiple levels of the visual hierarchy (Muckli and Petro, 2013; Kok and de Lange, 2015). The IPS could be enhancing the illusory surface representation on a coarser level and sending back shape information to early visual areas, which are capable of finer and more detailed representation of the subjective content (Roelfsema and de Lange, 2016).
Although the top-down feedback signals are assumed to explain the EVC response profile, it is not known in which areas such feedback signals could originate. One candidate region for the feedback source is the LOC (Chen et al., 2021), which is well known to respond to real (Kourtzi and Kanwisher, 2001) and illusory shapes (Fang et al., 2008; Chen et al., 2020). Our whole-brain results confirm the presence of illusory shape response in both early visual areas and the LOC (LO1). Importantly, our results point to a potential third player in this context, the IPS. Previous bistable perception studies hypothesized that the IPS could be an alternative candidate for sending the feedback signal to V1, especially because it was active even for stimuli that did not evoke the LOC response (Grassi et al., 2018). Interestingly, studies of bistable perception also found a dissociation within the PPC, with the more anterior areas presumably carrying the prediction signals (Kanai et al., 2011; Zaretskaya et al., 2013), and the more posterior areas conveying the prediction error (Kanai et al., 2011). Our results do not support the notion that anterior IPS maps provide the source for the prediction signal, because they showed no illusory shape response during the hard task, while the posterior IPS and the EVC did. Nevertheless, they highlight the functional dissociation within the IPS along the posterior-anterior gradient, which is worth further investigation.
In conclusion, our study shows that the topographic areas along the IPS represent illusory content, with responses in the posterior IPS being spatially specific and the responses in the anterior IPS depending on attentional resources. It remains to be seen whether and how responses within the IPS interact with the well-documented shape responses found in the ventral and early visual areas.
Footnotes
This work was supported by a BioTechMed-Graz Young Researcher Group grant. We thank Thomas Zussner for his help during MR data acquisition.
The authors declare no competing financial interests.
- Correspondence should be addressed to Ana Arsenovic at ana.arsenovic{at}uni-graz.at or Natalia Zaretskaya at natalia.zaretskaya{at}uni-graz.at
