Extrastriate activity reflects the absence of local retinal input

The physiological blind spot corresponds to the optic disc where the retina contains no light-detecting photoreceptor cells. Our perception seemingly fills in this gap in input. Here we suggest that rather than an active process, such perceptual filling-in could instead be a consequence of the integration of visual inputs at higher stages of processing discounting the local absence of retinal input. Using functional brain imaging, we resolved the retinotopic representation of the physiological blind spot in early human visual cortex and measured responses while participants perceived filling-in. Responses in early visual areas simply reflected the absence of visual input. In contrast, higher extrastriate regions responded more to stimuli in the eye containing the blind spot than the fellow eye. However, this signature was independent of filling-in. We argue that these findings agree with philosophical accounts that posit that the concept of filling-in of absent retinal input is unnecessary.


Introduction 22 23
In the optic disc of the retina of each eye, there are mostly blood vessels and axons with no light-24 detecting photoreceptor cells, producing a physiological scotoma referred to as the blind spot. In 25 visual space the blind spot is located near the horizontal meridian ~15º of visual angle from the fovea 26 up a considerable discontinuity in retinal input, we are almost always unaware of our blind spot, even 28 during monocular viewing ( Figure 1A). Our visual system seemingly uses the intact visual field to 29 interpolate into the blind spot, a process known as perceptual filling-in (Ramachandran, 1992). It is 30 generally observed when spatiotemporally coherent elements abut or surround the scotoma (Fiorani Here we used our novel retinotopic mapping method (Urale et al., 2022) to identify activity within 49 fMRI voxels encoding the blind spot location while stimulating the area within and outside of its 50 boundaries, both during viewing with the blind spot eye (BE) and the non-blind spot eye (NE). We used 51 a stimulus designed to elicit filling-in when viewing with the BE, but not the NE. Moreover, we used a 52 control stimulus visible to both eyes and another that was only visible to the NE. Participants also 53 performed a behavioural task to report whether they in fact experienced filling-in. A comparison of 54 brain activity across conditions in both eyes and across stimulus conditions thus allowed us to 55 investigate signals related to filling in: We hypothesized that visual brain regions responding more 56 strongly when participants experienced filling-in should indicate a neural correlate of this illusory 57 percept. If the red circle is still visible, adjust the viewing distance (~35-55cm) until the red circle is no longer 67 visible, and the image on the left appears as a uniform striped grating. B. Stimulus conditions. Blue 68 dotted outline shows blind spot boundary of hypothetical participant. Second and third rows show 69 percept when viewing each condition through the blind spot eye (BE) and non-blind spot eye (NE), 70 respectively. C. Layout of the screen as seen by a participant. The direction of the drifting sine-wave 71 pattern and the color of the arrow, respectively, changed over the course of each experimental run. We recruited eight participants (ages: 21-43, 3 females, all right-handed) from staff and student 77 populations. Each gave written informed consent to participate. Experimental procedures were 78 approved by the University of Auckland Human Participants Ethics Committee (UAHPEC). The first 79 until the desired perceptual effect of each condition was achieved. Once satisfied with each condition, 177 participants remained in the scanner for the filling-in experiment. 178 179

Filling-in experiment and behavioral task 180
For the filling-in experiment participants completed a series of 300-second runs. Each run consisted 181 of successive presentations of the three ring stimulus conditions plus blank intervals. Each condition 182 was presented for 15 seconds. The three ring conditions occurred in successive pseudo-randomized 183 triplets (e.g., filled, small, big) separated by a blank interval. Each condition occurred five times per 184 run. Each run was divided into epochs of 3 seconds. At the beginning/end of each epoch occurring 185 during presentation of a ring stimulus, the direction of the drifting grating changed (left or right). 186 Concurrently, the fixation arrow had a 50% chance of switching directions from left to right (or vice 187 versa). To avoid adaptation, the arrow cycled through three colors (red, green, blue) at the start of 188 each epoch. 189 During each run, participants were asked to perform tasks designed to encourage fixation to the arrow 190 and attention towards the ring stimulus (when present). They were instructed to press a button 191 (positioned at their index finger) once if they saw the fixation arrow change direction, and to press 192 twice in close succession ("double press") if when the arrow pointed in the same direction as the 193 movement inside of the ring stimulus. For example, participants were to press the button once if the 194 arrow changed from left to right while the drifting pattern was perceived as moving leftward (or if it 195 was not visible at all) or press twice if the drifting pattern moved rightward. Thus, single presses 196 represent fixation compliance (arrow subtask), and double presses imply attention encompassing 197 both the fixation arrow and the ring stimulus. Concurrently, participants were instructed to press a 198 second button (positioned at their middle finger) once if they experienced filling-in (filled task). 199 Following several runs (3-5) while viewing through the BE, participants were partially removed from 200 the scanner bore to shift the eye patch to the other eye, before completing the same number of runs 201 viewing with the NE. Instructions in were identical during runs while viewing through the BE and NE. 202 203

Magnetic resonance imaging 204
We used a Siemens MAGNETOM Skyra 3 Tesla scanner with a 32-channel head coil based at the Centre 205 of Advanced Magnetic Resonance Imaging at the University of Auckland. During all functional imaging, 206 we removed the front element of the coil to allow for an unrestricted field of view. The setup featured 207 20 effective channels covering the back and sides of the head. In Session 1, we performed 4-6 pRF 208 mapping runs of 250 T2*-weighted image volumes in each eye and an accelerated multiband sequence 209 with a TR of 1000 ms, 2.3 mm isotropic voxel resolution, and 36 transverse slices. These slices were 210 angled to be roughly parallel with the calcarine sulcus. The scan had a TE of 30 ms, a flip angle of 62°, 211 a 96 × 96 field of view, a multiband/slice acceleration factor of 3, and in-plane/parallel imaging 212 acceleration factor of 2, and rBW of 1680 Hz/Px. Following the pRF runs, the front element of the coil 213 was attached to achieve optimal signal-to-noise levels for collecting a structural scan. This scan was a 214 T1-weighted full-brain anatomical magnetization-prepared rapid acquisition with a gradient echo 215 (MPRAGE) scan with a 1 mm isotropic voxel size. The scan procedure in Session 2 was identical to 216 Session 1, except with 3-5 runs per eye (i.e., 6-10 runs total) each with 300 T2*-weighted images. 217 218

Pre-processing 219
For both sessions we realigned and co-registered functional data to the anatomical scan in SPM12 220 (Wellcome Centre for Human Neuroimaging; https://www.fil.ion.ucl.ac.uk/spm) using default 221 parameters. Following this, we used the automatic reconstruction algorithm on the structural scan in 222 FreeSurfer (Version 7.1.1: https://surfer.nmr.mgh.harvard.edu) to reconstruct inflated surface mesh 223 models of the boundary between grey and white matter and the boundary between grey matter and 224 the pia mater (Dale et al., 1999;Fischl, 2012;Fischl et al., 1999). We then projected functional data 225 onto this model by locating the voxel that sat on the mid-way point between each vertex on the grey-226 white matter boundary and the same vertex on the pial surface boundary. To remove slow drifts, we 227 applied linear detrending to the time series for each vertex and run, and then normalized the time 228 series to z-scores. We averaged runs from the BE and NE, respectively, leaving a single run for each 229 eye with 250 volumes each. For the Session 2 data, we instead concatenated the time series for all 230 runs across both eyes. 231 For Session 1 data, we limited further analyses (see pRF analysis below) to an area approximately 232 located in the occipital lobe by selecting vertices in the inflated cortex model with FreeSurfer y-233 coordinates of ≤ -35. Next, we determined a noise ceiling to identify vertices with a reliable response 234 to visual stimuli, by correlating the time series between even and odd runs in each vertex, separately 235 for the BE and NE conditions. We used this split-half correlation (r) and the Spearman-Brown prophecy 236 formula (Spearman, 1910) to calculate a noise ceiling: 237 Noise ceiling = 2 = 2 1 + 238 The noise ceiling therefore represents the maximum R 2 achievable for a given time series, and shows 239 the extent that a given vertex is visually responsive. To select only vertices with a reliable visual 240 response, subsequent pRF analyses were limited to vertices with a noise ceiling >.15 (see example 241 maps in Figure 2A). As expected, this revealed small clusters of visually evoked responses in the early 242 visual cortex because we only stimulated a small region of the peripheral visual field. 243 244 pRF analysis 245 We performed pRF analysis on Session 1 scan data for the purpose of identifying vertices encoding 246 the area inside and around the blind spot. These procedures are identical to those used in our previous 247 work ( Urale et al., 2022). Population receptive field locations (x, y) and size (σ) were modelled using 248 SamSrf Toolbox (Schwarzkopf, 2018). We treated the center of the blind spot as the center of the 249 (mapped) visual field and used moving bar stimuli to stimulate a circular area with a radius of 5.3° 250 around it. We created multiple stimulus apertures indicating the location of the bar stimulus during 251 the recording of each 1 s fMRI volume, and generated 100x100 pixel matrix where the activity in each 252 pixel corresponded to the presence of a stimulus at the corresponding location in the visual field 253 during the scan. We accounted for the lag in blood oxygenation-dependent response by convolving 254 this aperture with a canonical hemodynamic response function from previously collected data (de 255

Haas et al., 2014). 256
We then regressed the time series of each vertex on the time series of each pixel in this matrix. The 257 result was 100×100 of β coefficients for the observed time series from each vertex. The mosaic of 258 coefficients for each vertex therefore represented its responsiveness to areas across the stimulated 259 visual field, i.e., its pRF profile. The maximum of the pRF profile was taken as the peak response pixel. 260 Further analyses were limited to vertices for which the squared correlation between this pixel and 261 fMRI response was greater than 0.1. Additionally, the size and spatial location of the profile were 262 determined by fitting a symmetric two-dimensional Gaussian profile, with x and y coordinates, 263 standard deviation, and amplitude (βpRF) as free parameters. Analyses were further restricted to 264 vertices where goodness-of-fit of the Gaussian profile R 2 >0.5. This left vertices that were reliably 265 responsive to the bar stimulus, and which were determined as having a sufficiently clear pRF profile. Additionally, six motion regressors (x, y, z, pitch, roll, yaw) and a global constant for each run were 274 included in the GLM. For all conditions, signal was calculated by summing the β weights for the 275 presentation of the various ring stimuli and contrasting conditions against one another. We also 276 calculated the overall response to the big ring condition irrespective of the eye-of-origin. We then 277 smoothed these activation maps across the spherical surface mesh (full width at half maximum of 3 278 mm) for the purpose of selecting regions of interest. 279 280

Regions of interest 281
We used two separate procedures to determine regions of interest (ROIs). To analyze the retinotopic 282 representation of the blind spot and its surround, we used results of the pRF analysis from Session 1; 283 for each participant, we projected pRFs from the NE condition limited to those with a good fit pRFs (R 2 284 >.5) and manually selected the vertex falling at the center of the cluster in V1, and the ones in dorsal 285 and ventral V2 and V3, respectively. A geodesic region growing procedure defining a boundary 12 286 steps from this center location across the grey-white matter surface mesh resulted in a roughly circular 287 patch around each cluster. Because the blind spot intersects the horizontal meridian, corresponding 288 to the border between V2 and V3, we only defined a combined ROI in those areas. These pRF-derived 289 ROIs are denoted with the "pRF" subscript, i.e., V1pRF, V2+3pRF. Moreover, we used the NE retinotopic 290 map together with a convex hull of the blind spot localizer data to specify which pRFs fell within and 291 outside the blind spot. Note that in V2+3pRF for participant N02, no pRFs were identified inside the 292 blind spot and this participant was therefore excluded from analysis for this subregion of interest. 293 Following the retinotopic analysis, we conducted a coarser analysis of visually responsive brain regions 294 irrespective of retinotopic specificity. ROIs were defined by first identifying areas responsive to the big 295 ring stimuli, regardless of the eye-of-origin. We used smoothed maps of the overall response to the 296 big stimulus for either eye in conjunction with brain areas defined by an atlas (Sereno et Table 1). We then combined several of these individual atlas ROIs into larger clusters 304 (Supplementary Table 1). Visual areas that showed consistent visual activation across all participants 305 were selected for subsequent analyses. A total of five areas were selected: V1, V2, V3, V3A/B, V4, and 306 MT. While responses were present in other areas, these were either inconsistent across participants 307 or unlikely to signal visually evoked responses (Davey et al., 2016;Raichle, 2015). Within each of the 308 selected visual areas in the atlas, a ROI was generated by selecting a contiguous cluster of vertices 309 around the peak response using a region growing approach containing coefficients greater than the  Supplementary Information). Note, while some participants did not report filling-in during all such 326 trials, the small number of trials without this percept precluded us from analyzing these trials 327 separately. 328 We then tested the preferential response of neuronal populations in early visual cortex representing 329 the area in and around the blind spot during viewing through the BE versus the NE (Figure 2A). Figure  330 2B shows activity measured for individual V1 population receptive fields (pRFs) of one participant, split 331 by eye and stimulus condition. There were fewer responsive pRFs in any of the BE versus NE 332 conditions. Importantly, pRFs whose centers fell inside the blind spot did not respond at robust levels 333 during BE viewing at all. To quantify this across all participants, we performed a group analysis of 334 stimulus conditions (big, filled, and small). For each pRF, we calculated a difference score indicating 335 stronger activation to BE than NE stimulation (BE-NE; βΔ). We further divided pRFs into subregions 336 depending on whether they were within or outside the blind spot. Then we averaged these values 337 across all pRFs within each subregion ( Figure 2C,D). Responses in both V1pRF and V2+3pRF were stronger 338 in the NE than the BE for the filled and small conditions. In V1pRF, responses differed significantly with 339 location relative to the blind spot boundary (repeated measures analysis of variance: F(1,7)=8.95, 340 p=.02), and between the three stimulus conditions (F(2,14)=4.54, p=.03). The difference in responses 341 to the three conditions also depended on location (interaction: F(2,14)=3.81, p=0.048). We observed 342 similar results for V2+3pRF (location effect: F(1,6)=12.75, p=0.012; condition: F(2,12)=4.6, p=.033; but 343 no interaction: F(2,12)=0.63, p=0.55). 344 Crucially, pRFs inside the blind spot in V1pRF responded significantly less to BE than NE stimulation in 345 the filled and small conditions (one-sample t-tests vs. 0; small: t(7)=-4.77, p=0.002; filled: t(7)=-3.11, 346 p=0.017; α=0.0292 corrected for false-discovery rate), but not the big condition (t(7)=-0.78, p=0.459). 347 In V2+3pRF this difference was only significant for the small condition (t(6)=-3.56, p=0.012) but not the experiment was that these effects should be robustly identifiable at the level of individual participants. 353 We therefore defined the minimum criterion for consistent effects that at least 6 of the 8 participants 354 must show the same significant difference as the group average (one-sample t-test vs zero, p<.05,  (F(4,28)=5.13, p=0.003), and there was an interaction between these effects (F(8,56)=2.28, p=0.034). 373 Because one participant potentially failed to comply with task instructions, we repeated this analysis 374 after removing their data, but found the overall pattern remained qualitatively the same 375 (Supplementary Figure S2 and Supplementary Information). We also repeated our consistency analysis 376 for these data. For all regions of interest, we found weaker responses to the small stimulus in the BE 377 in at least 6 participants. In V1GLM, the filled stimulus also met this criterion. Meanwhile, the big 378 stimulus produced consistently stronger BE responses in V4GLM for all participants, and in MTGLM both 379 the big and filled stimuli produced stronger responses in at least 6 participants. In contrast to early regions, we found stronger visually evoked responses through the blind spot eye 416 (BE) in several extrastriate regions beyond V2/V3. We posited such a response pattern as a signature 417 of perceptual filling-in because it indicates the presence of a stimulus that was not there. However, 418 our findings do not support the interpretation that these higher extrastriate responses are a neural 419 correlate of filling-in. As hypothesized, a neural correlate should only appear for the filled condition, 420 where the stimulus abutted the blind spot border and participants experienced a filled grating. The 421 big condition was equally visible through both eyes and participants rarely reported filling-in (most 422 likely erroneous button presses); however, we measured the same signature of stronger BE responses 423 also for this condition, and in fact this was even greater than for the filled condition. What then could 424 cause such pronounced differences in responses between the eyes? 425 One explanation is that the gap inside the ring stimulus modulates higher visual areas differently, 426 depending on the eye. Filling-in could occur due to the absence of signals from retinotopic locations 427 mapped to inside the blind spot. Combined with a signal outside its borders (as in the big and filled 428 conditions) there is less modulation compared to NE viewing. Receptive fields in higher-level regions 429 are large (Dumoulin & Wandell, 2008) and encompass the blind spot as well as its surrounding space. 430 Greater activity in these regions could indicate that rather than an active filling-in process, higher 431 extrastriate neurons simply discount the missing input, because the monocular neurons in the 432 corresponding part of V1 do not respond. The process might also factor in the luminance difference provide an error signal to extrastriate cortex to ignore the lack of patterned input. 438 The stronger differential responses we observed for the big condition likely reflects the greater 439 stimulus energy in the big condition during BE-viewing, combined with the lack of inhibitory input 440 from inside the blind spot. The filled condition during BE viewing is subject to the same absence of 441 input, but the overall stimulus energy is lower; consequently, the net difference between BE and NE 442 is smaller. perceptual experience is of a continuous vertical white bar, second-order contours, rather than 448 completing the horizontal grating texture. B. Fixating with the right eye: The central donut will appear 449 perceptually completed as a filled grey disc, rather than a donut. 450 Our suggestion that higher regions discount the missing input bears similarities to earlier explanations 452 of filling-in (Dennett, 1992(Dennett, , 2017Durgin, 1998;Durgin et al., 1995) that in the absence of information, 453 higher visual areas simply label the missing area as "more of the same". Many cognitive scientists have 454 rejected that idea in favor of filling-in as an active process; however, we argue that previous findings 455 are in fact consistent with this explanation (Churchland & Ramachandran, 1996) unlikely that these signals merely reflect differences in the inducing stimulus configuration. Therefore, 480 the result implicates shape processing in producing the filled-in percept. However, given the higher-481 order nature of these shape stimuli, and the fact that similar results were observed across the visual 482 hierarchy, suggests the V1 results could have involved feedback from higher regions. Moreover, this 483 study used multivariate decoding techniques to evaluate the information about the color content of 484 the filled percept. In contrast, we tested the presence of a signal that there is any stimulus visible 485 within the blind spot at all. An equivalent decoding analysis for our research could entail decoding the 486 direction of motion of our grating stimulus. It is possible that the V1 representation of the blind spot 487 contains a signature of motion direction. Our experiment was not designed with such an analysis in 488 mind; ideally one would present motion directions in longer trials to obtain robust signals. 489 Compelling evidence for the involvement of higher areas in perceptual filling-in of the blind spot is 490 that adapting to a drifting grating annulus surrounding the blind spot, which creates a filled-in percept 491 of a complete grating, produces a motion aftereffect at the blind spot location in the fellow eye 492 (Murakami, 1995). Such inter-ocular transfer suggests the recruitment of binocular neurons, in higher 493 extrastriate areas with large receptive fields covering both the blind spot and its surrounding space 494 (e.g., MT). In fact, similar motion aftereffects can be induced at non-adapted locations that do not 495 involve the blind spot (Snowden & Milne, 1997;Weisstein et al., 1977). It is therefore more 496 parsimonious that filling-in occurs because these neurons merely discount the absence of signals from 497 inside the blind spot and thus interpret the stimulus as a complete grating. This then leads to 498 adaptation effects when stimulating the blind spot location. 499 Could the pattern of responses we observed be due to confounds in our design instead? First, we 500 always stimulated the BE before the NE. This was necessary for delineating the blind spot border 501 behaviorally and thus determine our stimulus dimensions, and for minimizing how often participants 502 must be placed in the scanner bore. However, this could have introduced order effects, e.g., due to 503 fatigue (Wylie et al., 2020). The NE eye was also deprived of vision for ~45 minutes in the first half of 504 each session. However, it is unlikely these effects are due to short-term monocular deprivation 505 because this boosts the response of the deprived relative to the non-deprived eye (Binda et al., 2018). 506 Second, while the two eye conditions might have differed in terms of cognitive load or attentional 507 allocation -in practice participants only needed to complete the filling-in task during BE viewing -508 they were monitoring for filling-in when viewing with either eye. It is implausible such attention would 509 only affect higher-level regions but cause opposite effects in V1 and V2+3. While V4 and MT have been 510 instance, exogenous cuing affects V1 similarly to other downstream visual areas (Dugué et al., 2020). 513 If responses in higher areas were driven by elevated visual attention in the BE condition, similar 514 responses should also have been observed in pRFs outside the blind spot in V1pRF and V2+3pRF. Instead, 515 we found no differences or even subtly reduced responses there. Behavioral performance also 516 indicated that attentional allocation and cognitive demand were comparable in both eyes 517 (Supplementary Information). Crucially, the differential response in higher regions was greatest in the 518 big condition, further inconsistent with an effect stemming from altered cognitive demands related to 519 the filling-in task. 520 Third, researchers have noted a response bias in some animals of greater signals in the hemisphere 521 contralateral to the stimulated eye (Schwarzkopf et al., 2007). However, such findings are specifically 522 about early visual areas, and therefore cannot account for selective activity that excludes V1-V3. It 523 also seems unlikely that this contralateral bias would manifest more strongly in higher cortical regions. 524 Moreover, it is unclear if such a bias even exists in humans. It could be related to the degree of 525 binocularity and the proportion of crossover fibers in the optic chiasm, which would predict it to be 526 much less pronounced than in other species. 527 Finally, we also consider it unlikely that unstable eye fixation could explain our findings. Our 528 experimental setup precluded eye tracking; however, participants performed a relatively demanding 529 task at fixation throughout the main experiment, judging the direction of a small arrow target. 530 Moreover, any break of fixation should have caused blurring of the stimulus-evoked responses in 531 retinotopic cortex. In turn this might cause responses of pRFs estimated as falling into the blind spot, 532 but we observed high retinotopic specificity of responses in the early visual cortex. In fact, the 533 perceptual experience of a filled grating would also have been disrupted by any breaks in fixation. 534 While to our knowledge ours is the first brain imaging study to investigate the neural correlates of 535 perceptual filling-in of the blind spot in human observers directly, a handful neuroimaging 536 experiments have targeted the blind spot. When stimulating a large hemifield, there is no V1 signature 537 of filling-in in fMRI responses, only a differential response indicating the blind spot representation 538 (Tootell et al., 1998). Similarly, when the NE dominates perception during binocular rivalry, responses 539 in the corresponding part of V1 are enhanced relative to when the BE is dominant; the V1 response 540 therefore only reflects the retinal input of the currently perceived stimulus, not a filled-in percept 541 (Tong & Engel, 2001). The cortical distance between activations evoked by stimuli on opposite sides 542 of the blind spot also does not differ between blind spot and control eyes in either V1 or V2/3 (Awater 543 et al., 2005). This shows there is no passive remapping of the blind spot where visual field maps on 544 either side are "sewn-up" in this eye. That study, however, did not evoke filling-in. In general, these 545 earlier fMRI studies lacked the spatial specificity possible today (Dumoulin & Wandell, 2008;Engel et 546 al., 1997). This is crucial because the representation of the blind spot in human visual cortex is small, 547 only subtending an area of about 50 mm 2 (Adams et al., 2007). Moreover, no previous study has 548 investigated regions beyond V2. We now reveal the brain activity measured during the actual percept 549 of filling-in in unprecedented detail. Instead of any neural correlate of filling-in, we propose that 550 neurons in higher extrastriate areas simply discount the absence of retinal inputs when integrating 551 local signals. As Dennett argued (Dennett, 1992), it is the purpose of the visual system to find out what 552 is out there in the world, rather than "filling-in" what is not there. Our finding is consistent with just 553 such a mechanism whereby brain responses only reflect the available sensory evidence. Future 554 research should employ experiments specifically designed to test this hypothesis further. Neural 555 network models of perceptual processing could also prove instrumental for putting its predictions to 556 the test. hypothesis as well as direct comparisons between null and alternative hypotheses (Wagenmakers et 801 al., 2018). Adopting Jeffreys's criteria for interpreting Bayes factors (Jeffreys, 1961), we found 802 moderate evidence for no effect of eye (BF10= .279), stimulus condition (BF10 =.241), and the 803 interaction between eye and stimulus condition (BF10 =.018) in the arrow task. 804 805

Results after excluding participant N07 806
After collating the behavioral results, we noted that participant N07 reported difficulty maintaining 807 attention and complying with task instructions at several points during the scan, which was also 808 reflected by their behavioral data. Hence, to investigate if their fMRI data affected the group analyses,  Table 1. Selected visual areas as defined by Sereno et al. (2022) brain atlas. The left column shows areas pre-selected based on an initial identification of responsive areas, and the right column shows responsive clusters used as ROIs for analysis in the present study. Shaded regions across both columns denote the correspondence between pre-selected areas and labelled responsive areas consistently across participants. Note that V2 and V3 ROIs from the atlas were combined into the V2+3 cluster.