Involvement of the Extrageniculate System in the Perception of Optical Illusions: A Functional Magnetic Resonance Imaging Study

Research on the neural processing of optical illusions can provide clues for understanding the neural mechanisms underlying visual perception. Previous studies have shown that some visual areas contribute to the perception of optical illusions such as the Kanizsa triangle and Müller-Lyer figure; however, the neural mechanisms underlying the processing of these and other optical illusions have not been clearly identified. Using functional magnetic resonance imaging (fMRI), we determined which brain regions are active during the perception of optical illusions. For our study, we enrolled 18 participants. The illusory optical stimuli consisted of many kana letters, which are Japanese phonograms. During the shape task, participants stated aloud whether they perceived the shapes of two optical illusions as being the same or not. During the word task, participants read aloud the kana letters in the stimuli. A direct comparison between the shape and word tasks showed activation of the right inferior frontal gyrus, left medial frontal gyrus, and right pulvinar. It is well known that there are two visual pathways, the geniculate and extrageniculate systems, which belong to the higher-level and primary visual systems, respectively. The pulvinar belongs to the latter system, and the findings of the present study suggest that the extrageniculate system is involved in the cognitive processing of optical illusions.


Introduction
The neural processing of an optical illusion can provide clues for understanding the neural mechanisms underlying visual perception. To study form perception and object recognition, optical illusions such as illusory contours or geometric illusions are often used [1][2][3][4][5].
There are two types of optic pathways: the geniculate and extrageniculate pathways [6][7][8]. The geniculate visual pathway involves transmission from the retina to the lateral geniculate nucleus (LGN) and then to the visual cortex. In contrast, the extrageniculate visual pathway runs from the retina to the superior colliculus, the pulvinar, and finally the visual cortex.
Some studies have reported that the neural mechanisms underlying processing of optical illusions occur in the early stage of visual processing, especially because neural activity related to this processing has been found in V1/V2 [9][10][11][12][13][14]. For instance, Murray et al. [12] demonstrated that the retinal size of an object and the depth information in a scene are combined in V1 and size illusions are reflected in the spatial pattern of activity in V1. Regarding the role of V2, Ramsden et al. [10] reported that the abutting line grating illusory contour activates orientation domains in V2 that overlap with those activated by luminance gratings.
However, other studies have demonstrated that the processing of optical illusions also depends on higher stages of visual processing. The lateral occipital cortex (LOC), which is highly involved in object recognition, has been shown to contribute to the processing of optical illusions [2,3,[15][16][17][18]. Studies on patients with brain damage support this finding [1,19]. Although these were not studies of individuals with brain damage, Daini et al. [19] revealed an association between damage to the occipital regions and the inability to perceive illusory effects. Patients with neglect and left hemianopia showed a lack of perception of illusory effects and had large bisection errors. These findings suggest that optical illusion processing likely occurs in the occipital cortex, at a retinotopic level of representation. From these previous studies, it can be concluded that the neural processes underlying the perception of optical illusions occur in the occipital cortex, specifically in the LOC.
In actuality, both early and higher-level visual stages are necessary for processing an optical illusion; the illusory aspect depends on the interaction between different levels of visual processing [20][21][22]. Feedback processing has been suggested to occur during optical illusion processing, that is, the processing of an optical illusion may follow an inverse hierarchical path [23][24][25][26][27]. Wokke et al. [28] used transcranial magnetic stimulation to disrupt signaling in the V1/V2 and LOC at different time points while participants performed a discrimination task for an illusory figure. Their results demonstrated that both V1/V2 and the LOC are critically involved in the perception of an optical illusion. These areas seem to be related in an inverse hierarchical fashion such that the critical time window for V1/V2 follows that for LOC. It was therefore suggested that after initial perceptual processing in the early visual cortices, optical illusions are detected by a higher visual area. Then, the information is fed back to the early visual cortices for completion, with strengthening of the figure-ground segregation processes or reception of predictive signals from higher visual areas.
However, these previous reports on the processing of optical illusions have the following limitations: (1) the visual stimuli differed between the main and control tasks, (2) the optical illusion figures were almost always restricted to the Kanizsa type and Müller-Lyer figure, and (3) the neural processing of the control stimuli was not always clarified. Therefore, the major aim of the present study was to examine the neural mechanisms underlying the processing of optical illusions using various illusory figures. We hypothesized that there are common mechanisms that account for line (e.g., Müller-Lyer) and area (e.g., Kanizsa) types of optical illusion. In previous studies, the use of different stimuli for the main and control tasks may have led to differences in the activated brain regions. Therefore, as suggested by previous studies, the question of whether or not the information from higher levels is fed back to early visual cortices remains, and there could be an alternative explanation that suggests involvement of extrageniculate pathways. Therefore, we performed functional magnetic resonance imaging (fMRI) during the presentation of optical illusions using Müller-Lyer, Ponzo, Hefler, Zerbino, Ebbinghaus, Jastrow, and Delboeuf figures. These figures consist of many kana letters, which are Japanese phonograms. We used the same stimuli for both the main task and the control task to detect differences in cognitive processing.

Materials and Methods Participants
Eighteen right-handed volunteers (9 men, 9 women; mean age, 21.9 ± 1.2 years) participated in this study. All participants had normal or corrected-to-normal vision. The participants provided written informed consent before the experiment in accordance with the Declaration of Helsinki. Our study was approved by the ethics committee of Mie University.

Stimuli
The stimuli used in the present experiment were illusory figures consisting of many kana letters, which are Japanese phonograms (e.g., Fig 1A-1D). Each stimulus was adopted from two books of optical illusions [29,30]. We changed the line of the stimuli to Japanese phonograms. We used the following types of illusions: 2 types of Müller-Lyer illusions with the Japanese phonogram "e" or "hi"; the Ponzo illusion with "na"; 2 types of Hefler illusions with "sa" or "ki"; the Zerbino illusion with "he" as an optical illusion of lineal length; 3 types of Ebbinghaus illusions with "ka," "mi," or "su"; 2 types of Jastrow illusions with "to" or "re"; and the Delboeuf illusion with "hu" as an optical illusion for area. Each stimulus was displayed in two different ways by flipping it vertically or horizontally. Thus, we used 12 types of stimuli and there were 24 total stimuli. Stimuli were displayed using magnetic resonance (MR)-compatible goggles (CinemaVision, Resonance Technology Inc., CA) at a total resolution of 240,000 pixels and covered a visual angle of 30°horizontally and 22.5°vertically. The actual size of the stimulus was 630 × 473 pixels. Black stimuli were presented in the center of the screen on a white background. Stimuli were controlled using E-Prime software (Psychology Software Tools, Inc., PA) on a personal computer. There was no modulation of stimulus size across trials.

Tasks
Our experimental design consisted of two experimental tasks, namely a shape and a word task. During the shape task, a stimulus was presented and participants were instructed to judge whether they perceived the length/area constituted by the Japanese letters of two optical illusions as being the same or not, and to say their answer aloud. During the word task, the participants saw the stimulus and read aloud the kana letters present. Participants were instructed to respond as quickly as possible on all trials at the response instruction. During the baseline periods, a fixation cross was presented. Participants practiced the tasks both inside and outside of the scanner until they could execute them confidently with specific training material before the experiment.

Session
A mixed design containing features of both blocked and event-related approaches was used. Each condition (word/shape task) lasted for 39 s, including the instruction slides, followed by a baseline period (20 s). When contrasts were calculated against the baseline, the baseline was thus implicitly defined as a rest block of 20 s (mean signal during non-modeled periods). During each condition, 3 trials/stimuli were presented in random order every 13 s; each stimulus appeared for 4 s, with an interstimulus interval of 9 s (Fig 1E). Each stimulus was presented with equal frequency. This experiment was composed of 2 sessions per participant, and each condition was presented 8 times in random order per session.

fMRI data analysis
Preprocessing and data analysis were performed using SPM8 software (Wellcome Department of Imaging Neuroscience, London, UK). The first five functional images were discarded to account for magnetic saturation. In total, 319 images were acquired per session and per subject. The functional images were temporally corrected for acquisition time differences with regard to the middle slice, realigned to the first image to correct for movement-related effects, coregistered to the anatomical image, normalized to the Montreal Neurological Institute (MNI) brain template, and spatially smoothed with an isotropic Gaussian kernel (full width at half maximum = 8 mm). We conducted voxel-wise statistical analyses based on the general linear model. For the statistical model, an event-related design was modeled using the canonical hemodynamic response function and temporal derivative, and low-frequency drifts were removed using a high-pass filter (128 s). The onsets were defined as the onset time of the stimulus presented. For each participant, we computed contrasts for "shape task > baseline," "shape task > word task," "word task > baseline," and "word task > shape task." A random-effects model was used for the group analysis. We used two different correction levels: voxel-level correction was employed for baseline contrasts and cluster-level correction for differential contrast. We assessed the statistical significance at a single voxel threshold of p < 0.05, family-wise error [FWE]-corrected (voxel-level corrected) or cluster threshold of p < 0.05, FWE-corrected with a voxel threshold of p < 0.001, uncorrected (cluster-level corrected), and activations that involved a contiguous cluster of at least 10 voxels were reported. MNI coordinates indicating the peak activation were converted to Talairach coordinates [31] using a non-linear transformation of the MNI brain image to the Talairach brain image (http://imaging.mrc-cbu.cam.ac. uk/imaging/MniTalairach). The active cortical areas were found using Talairach Client [32].

Effective connectivity analysis
In the analysis, we focused on the function of the pulvinar and visual cortex, which are part of the geniculate and extrageniculate optic pathways to investigate whether the extrageniculate system is involved in the cognitive processing of optical illusions. The geniculate visual pathway involves transmission from the retina to the lateral geniculate nucleus (LGN) and from there to the visual cortex. In contrast, the extrageniculate visual pathway starts from the retina and includes the superior colliculus, the pulvinar, and finally the visual cortex. First, volumes of interest (VOIs) were selected, based on activity and anatomical constraints [6][7][8]. V1 was defined as an area showing greater activation for shape and word tasks compared to the baseline. The pulvinar was defined as an area showing greater activation for the shape task compared to the word task. VOI selection was based on the T-contrasts adjusted with the F-contrasts (p < 0.01 uncorrected). VOIs were spherical, with a radius of 8 mm around the peak activation. The variance explained by the first eigenvariate of the blood oxygenation level dependent signals was higher than 69%.
The effective connectivity was tested by DCM-10, implemented in SPM8 toolbox (Wellcome Department of Imaging Neuroscience, London, UK). Models of the DCM were defined with endogenous connections, representing coupling between brain regions (matrix A), modulatory connections (matrix B), and driving input (matrix C). In matrix A, we defined the connections between V1 and the pulvinar. In matrix B, we defined the modulatory connections from V1 to the pulvinar or from the pulvinar to V1. Images of illusory figures consisting of many kana letters, i.e., Japanese phonograms, served as the driving input (matrix C).
Model estimation was conducted to maximize the negative free-energy estimates of the models (F) for a given dataset [33]. Furthermore, Bayesian model selection (BMS) was conducted for random (RFX) effect designs. In Model 1, visual information was input to the pulvinar, and the shape task, as a modulatory input, had an effect on the forward connections from the pulvinar to V1. Model 1 represents blindsight, which is the ability to respond to visual stimuli without consciously seeing them and is exhibited by people with cortical blindness resulting from lesions in the primary visual cortex. In Model 2, visual information was input to V1, and the shape task, as a modulatory input, had an effect on the forward connections from V1 to the pulvinar. In Model 3, visual information was input to V1 and the pulvinar, and the shape task, as a modulatory input, had an effect on the forward connections from the pulvinar to V1. In Model 4, visual information was input to V1 and the pulvinar, and the shape task, as a modulatory input, had an effect on the forward connections from V1 to the pulvinar (Fig  2A). The four models were compared by a random-design BMS to identify the model with the highest exceedance probability.

Behavioral data
Fourteen sessions were excluded because the rate at which a stimulus was identified as an optical illusion was below the level attributed to chance (50%: true or false). Thus, we used 22 sessions in the final analysis. The identification rates for the optical illusions were 77% and 73% for the shape task with lines and the shape task with areas, respectively. The rates of identification for word stimuli were 98% and 98% for the word task with lines and the word task with areas, respectively. We performed a task (shape or word) × stimulus type (line or area) mixed-factor analysis of variance (ANOVA) on the stimulus identification rates for optical illusions or words. There was no significant main effect of stimulus type [F (1, 21) = 0.851, p = 0.37] or interaction between task and stimulus type [F (1, 21) = 0.296, p = 0.59]. However, there was a significant main effect of task [F (1, 21) = 71.05, p < 0.001]. Because reading single kana letters is an easy task, the behavioral results indicate that the identification rate for word stimuli was higher than that for optical illusions (Table 1).

fMRI data
Shape task vs. baseline. During the shape task, the trials in which stimuli were identified as optical illusions were associated with bilateral activation in the inferior and superior frontal gyrus, precentral gyrus, cingulate gyrus, thalamus (dorsomedial nucleus [dm]), ventral lateral nucleus [vl]), and the pulvinar [p]), cuneus, and occipital area relative to baseline (voxel-level corrected) (Fig 3A, Table 2). During the trials in which the stimuli were not identified as optical illusions, bilateral activation occurred in the inferior and superior frontal gyrus, precentral Extrageniculate System Involvement in Perception gyrus, cuneus, and occipital area relative to baseline (voxel-level corrected), but there was no bilateral activation in the thalamus, including the pulvinar (Fig 3B, Table 3).  Furthermore, for trials in which stimuli were identified as optical illusions, we compared the data for the shape tasks with lines and areas to the baseline data. There was no difference among the types of stimuli.
Word task vs. baseline. For this analysis, we used the trials in which the stimuli were identified as words. Compared to the baseline, the word task induced bilateral activation in the inferior and superior frontal gyrus, cingulate gyrus, thalamus (dm, vl, and p), and occipital area (voxel-level corrected). In addition, activation was observed in the inferior parietal lobule including Brodmann area (BA) 40 (voxel-level corrected) in both tasks compared to the baseline.
Furthermore, for trials in which stimuli were identified as words, we compared the data for the word tasks with lines and areas to the baseline data. There was no difference among the types of stimuli.
Shape task vs. word task and word task vs. shape task. We used the trials in which the stimuli were identified as optical illusions in the shape task or as words in the word task for these analyses. Therefore, the number of experimental trials was different in both tasks. The shape vs. word task comparison (Fig 4, Table 4) revealed activation in the right inferior frontal gyrus and left medial frontal gyrus, and in the right thalamus including the pulvinar (cluster-level corrected). In contrast, the word vs. shape task comparison did not reveal significant activation.
Effective connectivity. BMS was used for deciding which model best explained the measured responses. The values for the expected posterior probability were 0.08, 0.09, 0.7, and 0.12 in Models 1, 2, 3, and 4, respectively (Fig 2B). Model 3 out-performed the other three models, with an exceedance probability of 0.97 compared to 0.00, 0.01, and 0.02 in Models 1, 2, and 4, respectively (Fig 2C). The higher performance of Model 3 suggests that visual information was indeed input to V1 and the pulvinar, and that the shape task, as a modulatory input, had an effect on the forward connections from the pulvinar to V1.

Discussion
We performed an fMRI study on healthy participants who viewed Müller-Lyer, Ponzo, Hefler, Zerbino, Ebbinghaus, Jastrow, and Delboeuf figures consisting of many Japanese letters. We used the same stimuli for the shape and word tasks. Our results demonstrated a large overlap in activation patterns when participants performed the shape and word tasks; the overlapping regions included the inferior and superior frontal gyrus, precentral gyrus, medial frontal gyrus, thalamus, and occipital area in both hemispheres. This pattern of commonly activated brain regions may constitute a shared neural network for processing both optical illusions and words; activation of the occipital area is involved in visual perception (e.g., [34]), that of the medial and inferior frontal gyrus is involved in resolving conflicting information (e.g., [35,36]), and that of the precentral gyrus is involved in giving responses aloud (e.g., [37]). Our results showed that line and area figures may have several types of neural processing in common for each task; the shape task with lines and areas vs. baseline comparison revealed activation in the inferior and superior frontal gyrus, precentral gyrus, insula, thalamus (dm, vl, and p), cuneus, and occipital area, whereas the word task with lines and areas vs. baseline comparison revealed activation in the inferior and superior frontal gyrus, cingulate gyrus, thalamus (dm, vl, and p), occipital area, and inferior parietal lobule. These results provide a new perspective on the neural mechanism of optical illusions.
The shape vs. word task comparison revealed activation in the dorsomedial nucleus, ventral lateral nucleus, and the pulvinar of the thalamus. The dorsomedial and ventral lateral nuclei of the thalamus are involved in activation of the prefrontal cortex and motor area, respectively [38][39][40]. The findings of the present study suggest that the pulvinar is involved in the cognitive processing of optical illusions. The optic pathway consists of both the geniculate and extrageniculate pathways, and the pulvinar is closely related to the extrageniculate system. The geniculate visual pathway follows a path from the retina to the lateral geniculate nucleus (LGN) and from there to the visual cortex. In contrast, the extrageniculate visual pathway consists of a pathway from the retina to the superior colliculus, the pulvinar, and finally the visual cortex. The pulvinar is considered the major source of visual processing during blindsight (e.g., [41,42]), which is the ability to respond to visual stimuli without consciously seeing the stimulus and is exhibited by people with cortical blindness resulting from lesions in the primary visual cortex (Model 1 of Effective Connectivity Analysis).
Some variation in pulvinar activation across tasks was observed; it appeared strongest in the shape tasks, but was weaker in the word task. Only trials in which the stimuli were identified as optical illusions may demonstrate pulvinar activation. Thus, our results suggest that the extrageniculate system is associated with optical illusion processing. From the results of the effective connectivity analysis, the information from an optical illusion is projected to the visual cortex from both the LGN and pulvinar. We therefore may be able to detect optical illusions when visual information follows not only the geniculate but also the extrageniculate visual pathway.
Previous studies have suggested the involvement of feedback processing during the perception of an optical illusion; that is, an optical illusion may follow an inverse hierarchical path [23,24,27]. After extraction of global configural cues in the LOC, shape information may be sent back to the low-level areas in V1/V2. However, the feedback can also be explained by visual information input from the pulvinar. Thus, the time-lagged activation of V1/V2 after activation of the LOC demonstrated in previous studies may not reflect an inverse hierarchical path, but may rather indicate visual information input from the extrageniculate visual pathway. The suggestion that the Table 4. Shape vs. word task using trials in which stimuli were identified as optical illusions in the shape task or as words in the word task. extrageniculate visual pathway is involved in the perception of optical illusions is consistent with previous results indicating that mammals, birds, and insects can detect optical illusions [10,43,44]. Our results also demonstrated LOC activation during the shape task. Previous studies have suggested that the occipital cortex is the anatomical basis of optical illusions. In particular, the observation of LOC activation is consistent with the findings of previous studies [2,3,[15][16][17][18]. The LOC is thought to be associated with the representation of objects, object fragments, and figure-ground segregation [45,46]. Harris et al. [47] suggested that the LOC integrates local elements involved in perceiving an optical illusion. Vallar et al. [1] suggested that the anatomical basis of the Müller-Lyer illusion lies in regions dedicated to the processing of illusory contours. The present study not only confirms these previous findings but also extends them by showing LOC activation in a mixture of responses to various optical illusions other than Kanizsa-type and Müller-Lyer figures.
In addition, we found activation of the frontal lobes during the shape task. The activation of the frontal lobes was stronger in the right hemisphere than in the left hemisphere. This is in agreement with electroencephalography (EEG) and magnetoencephalography (MEG) data reported by Qiu et al. [48] and Weidner et al. [3], respectively, which suggested that this component reflects high-level cognitive control related to judging information from the illusion. This suggestion agrees with data reported by Vossel et al. [49], who found that the right posterior inferior frontal gyrus is activated by stimulus changes in an oddball task. Therefore, we suggest that activation of the right inferior frontal gyrus was involved in judging whether the two stimuli were identical during our experiment.
The present study has the following limitations. First, the time course of perceiving an optical illusion was not clarified. In contrast to hemodynamic imaging methods, EEG and MEG provide a finer temporal analysis of the time course of the processes involved in the perception of an optical illusion, although their spatial resolution is much less reliable than that of positron emission tomography or fMRI. Second, we used the same stimuli consisting of many kana letters for the shape and word tasks. Therefore, our optical illusion stimuli were different from the original figure images. The relatively low identification rate for the optical illusion stimuli might have been caused by this difference. However, in our analysis, we used only trials in which stimuli were identified as optical illusions; therefore, the low identification rate may not have influenced our results. Third, differences in task difficulty between the shape and word tasks might have affected the results. For example, the results might represent a stronger decision conflict in the shape task relative to the word task. Therefore, there may have been less reliable information available for the shape task, which may have thus resulted in "noisier" data. However, in our analysis, during the shape task, the trials in which stimuli were identified as optical illusions were associated with the pulvinar. During the trials in which the stimuli were not identified as optical illusions, there was no bilateral activation in the thalamus, including the pulvinar. These results may indicate that the extrageniculate system is involved in the cognitive processing of optical illusions without any relationship to the difficulty of the shape and word tasks. In the future, some adjustments of the difficulty between the shape and word tasks will be required to eliminate the decision conflict factor.