• Home   /  
  • Archive by category "1"

Fsl Retinotopic Mapping Assignment


Topological (neighbor-preserving) remapping is a key principle of organization of sensory and motor areas within the mammalian brain. In primary sensory and motor cortices, these representations initially reflect the spatial layout of the receptors; for instance retinotopic maps in visual cortex topologically encode retinal locations, tonotopic maps in auditory cortex represent positions along the cochlear hair cell line, which correspond to sound frequency, and somatotopic maps in the somatosensory cortex represent locations on the body surface. Disrupting these maps has been shown to affect subsequent sensory processing and behaviour [Kaas, 1997; Sperry, 1943]. Traditionally it has been assumed that topological mapping was limited to lower-level (primary sensory and motor) cortex [Hebb, 1949]. But the entire cortex is characterized by the overwhelming predominance of local connections [Ercsey-Ravasz et al., 2013; Lund et al., 1993; Schmahmann and Pandya, 2009]. The last few decades of electrophysiological investigations in monkeys and neuroimaging research on humans has shown that topological organization extends well into higher-level (beyond primary sensory and motor) cortical areas [Felleman and Van Essen, 1991; Huang and Sereno, 2013; Sereno and Allman, 1991; Wandell et al., 2007]. In contrast to lower-level maps, however, localized activity in higher-level maps is affected as much by spatial attention as by the spatial characteristics of the stimuli [Saygin and Sereno, 2008]. In frontal cortex, there is evidence that topological maps may serve as a convenient method of allocating working memory, or maintaining pointers to specific content [Hagler and Sereno, 2006], even for tasks not overtly referencing position; for example, the exact areas that showed more robust activity during an identity two-back task (in which location was ignored), than during a location two-back task (in which identity was ignored) turned out to contain retinotopic maps. A significant role for topological maps in other complex mental operations has been suggested before [Simmons and Barsalou, 2003; Thivierge and Marcus, 2007], but direct neuroimaging evidence supporting this idea has been scant, and mostly confined to somatosensory cortex.

In the study presented here, we used fMRI to directly assess the extent of overlap between cortical regions involved in reading comprehension and those that have a topological sensory or motor map. It is generally assumed that the activation observed in temporal and frontal areas during reading falls beyond the bounds of sensory-motor maps, but this assumption has not been explicitly tested across all modalities in the same group of subjects. Recent advances in cortical surface-based mapping techniques have had less exposure in the language literature; for example, “sensorimotor” regions are often defined only by anatomical features or by basic, non-attention demanding tasks. In this study, we look at the full extent of topologically mapped sensory-motor regions that can be reliably detected using retinotopic, tonotopic and somatomotor mapping and assess where and by how much they intersect with brain regions involved in reading comprehension in the same subjects. To this end, each subject in our study participated in four separate fMRI sessions, where the sessions comprised a naturalistic reading comprehension task, retinotopic mapping, tonotopic mapping, and somatomotor mapping. A key goal of this study was to determine to what extent topological sensory-motor maps, traditionally associated with low-level sensory and motor processing are present in the cortical regions significant for a complex high-level cognitive task such as reading comprehension. Although we have used the terminologies “lower-level” and “higher-level” in its traditional sense for ease of explanation, it is important to keep an open mind about “levels”: an apparently lower-level area may be performing language-specific functions while an apparently higher-level area may be performing other functions besides language.

On the methodological front, we employed a fully surface-based group analysis as opposed to volume-based group analyses commonly used in language studies (merely displaying a 3-D averaged result on an average surface gains none of the benefits of surface-based averaging). The cerebral cortex has the topology of a 2-D sheet. Many relevant dimensions (e.g., retinotopy, somatotopy, tonotopy) vary much more rapidly tangential to the cortical surface than they do perpendicular to the cortical surface, through the several millimeters of cortical thickness. Distances measured in 3-D space between two points—but also used in standard pre-fitting 3-D smoothing—can substantially underestimate the true distance along the cortical sheet due to its folded nature [Fischl et al., 1999a, 1999b]. This artifactual within-subject blurring is then made worse by 3-D averaging of between-subject variability in the secondary crinkling of cortical folding patterns. Surface-based techniques make it possible to restrict smoothing to directions parallel to the cortical sheet, and to employ inter-subject 2-D alignment based on the patterns of sulci and gyri after secondary crinkles have been removed, which reduces both kinds of artifactual blurring and improves cross subject averaging [Fischl et al., 1999b]; this also provides a less biased estimate of overlap. Finally, siting language regions with respect to topological cortical maps provides a more precise way to compare activations across individuals and groups as well as studies. This is particularly important for refining functional localization in less well-understood regions such as frontal cortex.



20 right-handed native English speakers (nine women) participated in this study. The mean age was 28 (ranging from 19 to 58). All participants were neurologically healthy with normal or corrected to normal vision and normal hearing capacity. The experimental protocols were approved by local ethics committees and participants gave their informed written consent prior to the scanning session. The study required each participant to take part in four separate fMRI experimental sessions: Reading task, Retinotopic mapping, Auditory mapping and Somatomotor mapping. All 20 participants took part in Reading task and Retinotopic mapping experiments. 18 of the same participants took part in Auditory mapping and 17 of the same participants took part in Somatomotor mapping. The data from three participants in the Reading task and two participants in the Somatomotor mapping task did not meet the required quality criteria and were excluded from group results (discussed further in the Results section).

Experimental Stimuli and Design

Reading experiment

The reading experiment consisted of a naturalistic reading task where comprehension blocks contained a short narrative passage in English. The experiment was a random-order block design with three conditions and a central fixation screen. Each condition block was 16 sec long. During the experimental condition, a passage in English was presented one word at a time for 16 sec (64 words in total, average rate 4 words/sec). Each word was briefly presented on the screen at its natural reading position within the text. All other words appeared as grayed rectangles. The background screen was shown at 40% brightness (RGB: 102,102,102), the word was shown at 0% brightness (RGB: 0,0,0) and the rectangles were shown at 36% brightness (RGB: 92,92,92). The space between successive words/rectangles was set to 35% of the average word length. The exact display time of a particular word was made a piecewise linear function of the rendered word length. Vera.ttf, a san-serif typeface was parsed by FreeType 2.4.11 and rendered and measured in OpenGL using FTGL 2.1.3. The minimum display duration of a word was clamped to 175 msec, which was 70% of the average word duration of 250 msec. Words with widths more than 70% of the average word width were then given linearly increasing display durations with a slope chosen so the total paragraph duration equaled the desired block length (16 sec). This resulted in virtually identical duration/word-length slopes across different 64-word paragraphs. The guided reading experience felt much more subjectively natural when duration was controlled by word length than when a fixed word duration was used. In order to control for low level visual processing, there were two other conditions.

The condition 2 consisted of the “Hindi” version of the English passage (simple substitution of characters from the font, lekhani_dynamic.ttf, not a translation, so rendered word lengths remained identical) using presentation mode and fixation durations identical to the English condition. In the third condition, Dot, a 0.5 deg visual angle dot instead of a word briefly replaced each of the grayed rectangles, again with the same varying fixation durations and rectangle lengths. The baseline condition (condition 0) presented a single central fixation dot. The experiment consisted of 4 runs, where each run was comprised of 32 blocks presented in a random order. Participants, who had no familiarity with “Hindi” were instructed to do their best to comprehend the English passages in all four runs and to follow the Hindi script “words,” or dot in the other two conditions. The passages were self-contained and unrelated to each other. The level of comprehension achieved for the English passages was measured with a questionnaire afterward. The stimulus presentation technique used here, where each English word, Hindi word or dot was briefly presented in its natural reading position with all other words grayed out served several purposes. In the spirit of the classic attention study by Posner [1980], the subject's exogenous attention is automatically drawn toward each newly highlighted position, where the word or dot appears, in a manner very similar to natural reading. The subjects moved their eyes along with the highlighted word and reported a naturalistic reading experience, which was aided by a naturalistic (word-length dependent) fixation duration. Additionally this presentation mode also ensured that participants made controlled eye saccades (and extremely similar eye movements across conditions) as opposed to uncontrolled eye movements if the passages were presented in its entirety. In order to further ensure that participants stayed attentive and made similar eye-movements for all conditions, they were instructed to press a button when the color of an English word, a Hindi pseudo word, a dot, or the central fixation dot changed from black to off-white (50% brightness, RGB 128,128,128) for an average of 0.25 seconds. The responses to button press events were logged and analyzed to assess the quality of task execution. Button press events were modeled as an extra regressor and used as an additional quality control check during data analysis. The English screen of the experiment and a sample English passage are shown in Figure 1. The stimulus was programmed in C/OpenGL/X11 (stimulus program available on request). An optimized random order of the conditions within each run was generated using AFNI's [Cox, 2012] RSFgen program. Prior to the scan, participants were shown a demo of the experiment (using different English passages). They were instructed to read with an intention to comprehend during the “English” condition, to move the eyes along with the “Hindi”/“Dot” word in Hindi and Dot conditions and to fixate at the central dot during the fixation condition. They were asked to perform the target detection task that occurred randomly across all four conditions. The participants were informed that the button press events would be logged to assess how well they followed the eye movement instructions (and performed the task) and that the comprehension of English passages would be assessed using a questionnaire after the scan.

Retinotopic mapping

In the retinotopic mapping experiment, we mapped polar angle using a phase-encoded stimulus very similar to that used in previous recent work by Sereno et al. [2013]. The stimulus (Fig. 2) consisted of a continuously rotating thin wedge (18 deg wide) populated with a random-colored checkerboard with 35% luminance contrast. The checkerboard was overlaid with white dot fields moving in 500 msec periods of coherent motion that extended slightly beyond the checkerboard wedge to 21.6 deg wide (each new flow period had randomly chosen contraction/dilation and clockwise/anti-clockwise components, dots had 50% average luminance contrast), as well as two simultaneous asynchronous streams of random objects (tiffs of single objects with a transparent background, 0.5 sec duration, 0.1 sec gap) and random black letters (0.4 sec duration, 0.1 sec gap) placed at random eccentricities; both kinds of stimuli were scaled with eccentricity to fit within the confines of the 21.6 deg wide wedge, and their centers were rotated together with the wedge. The objects had an additional random radial (inward or outward, eccentricity-scaled) motion. This polar angle mapping stimulus was designed to evoke activation in the maximum number of lower-level and higher-level visual areas. The participant was presented with the periodic stimulus (64 sec per full rotation cycle, eight cycles/run). 512 second runs (four in total) alternated between clockwise and anti-clockwise rotation of wedges. Participants were required to fixate on the center dot at all times. Additionally they were instructed to monitor for occasional numbers (among the letters) and occasional upside down objects (among the right-side-up objects) to maintain a high and continuous level of peripheral attention to the entire wedge during central fixation.

Auditory mapping

In the auditory mapping experiment, we mapped tonotopically organized cortical areas using a frequency-modulated stimulus taken unchanged from the previous study by Dick et al. [2012]. The stimulus consisted of bandpass-filtered nonlinguistic vocalizations adapted from the Montreal Affective Voices [Belin et al., 2008], a series of recordings by actors (only male voices were utilized in the stimulus) producing sounds associated with a set of eight emotions. Each run consisted of eight 64 sec band-pass-filtered cycles, where the center filter frequency repeatedly logarithmically ascended from 150 to 9,600 Hz, or repeatedly descended from 9,600 to 150 Hz (no gap at the wrap-around point). The session consisted of 4 runs alternating between ascending and descending frequency sweeps. During scanning, subjects were asked to monitor the stimuli and press a button whenever they heard laughter (laughter targets were distributed non-periodically through the stimulus train). The band pass filter made this task challenging.

Somatomotor mapping

In order to reveal sensory and motor maps in primary and secondary somatosensory and motor cortical areas, participants were asked to move different body parts in response to periodic auditory cues (spoken name of the body part to be moved, rendered using Mac OS X text-to-speech with “Alex” voice) in as gentle and localized fashion as possible. Each run consisted of 8 cycles of 64 sec, and in each cycle participants successively moved 11 different body parts progressing from tongue to toe. Runs (four in total) alternated between movement cycles in each direction (from tongue to toe and toe to tongue). The stimuli were presented by an in-house C/OpenGL program and the conceptual design is similar to that described in Zeharia et al. [2012]. The participants were allowed to familiarize themselves with the auditory cues and practice the controlled movements prior to the scan.

Auditory instructions were: “say tchuh-tchuh”; “say pup-pup”; “crinkle eyebrows”; “touch thumb index”; “wave wrist”; “contract biceps”; “pull in stomach”; “squeeze buttocks”; “contract quads”; “wave ankle”; “rub big toe”.

Experimental Set-up

For visual experiments (Reading task and Retinotopic mapping), the stimuli were projected into the bore using an Eiki LC-XG300 XGA video projector onto a translucent direct-view screen at the participant's upper chest level. All polar angles of the visual field were stimulated out to an eccentricity of at least 54 degrees of visual angle (much larger than the usual 8–12 degrees of all-polar-angles eccentricity achieved when a standard screen is viewed via a mirror). This avoids artifactual periodic modulation of voxels representing visual field locations beyond the outer edge of the stimulus due to surround inhibition. A black matte shroud situated just outside the bore blocked the beam from making low-angle reflections off the top of the bore. The rear of the head coil was elevated with a wooden wedge and thinner bed cushions were used to help naturally tilt the head forward. For auditory and somato-motor mapping, the stimuli were delivered binaurally using in-house safety-enhanced Sensimetrics (Malden) S14 earbuds and cushions. During all scanning sessions, memory foam cushions (NoMoCo Inc.) were packed around the head to provide additional passive scanner acoustical noise attenuation and to stabilize head position. Responses were made via an optical-to-USB response box (LUMItouch, Photon Control, Burnaby, Canada) situated under their right hand.

Imaging Parameters

Functional images were acquired on a 1.5 T whole-body TIM Avanto System (Siemens Healthcare), at the Birkbeck/University College London Centre for NeuroImaging (BUCNI), with RF body transmit and a 32-channel receive head coil. For the first 16 fMRI sessions, images were acquired using the standard product EPI pulse sequence (24 slices, 3.2 × 3.2 × 3.8 mm, 64 × 64, flip = 90°, TE = 39 ms, TR = 2 sec), while the remaining 64 sessions used multiband EPI (40 slices, 3.2 × 3.2 × 3.2 mm, flip = 75°, TE = 54.8 ms, TR = 1 sec, accel = 4) [Moeller et al., 2011]. Individual scans had 260 volumes for standard EPI and 520 volumes for multiband EPI. To allow longitudinal relaxation to reach equilibrium, 4/8 initial volumes were discarded from each run for standard/multiband EPI. For each imaging session, a short (3 min) T1-weighted 3D MPRAGE “alignment scan” (88 partitions, voxel resolution 1 × 1 × 2 mm, flip angle = 7°, TI = 1,000 ms, TE = 4 ms, TR = 8.2 ms, mSENSE acceleration = 2x, slab-selective excitation) was acquired with the same orientation and slice block center as the functional data, for initial alignment with the high-resolution scans used to reconstruct the subject's cortical surface. Two high resolution T1-weighted MPRAGE scans (176 partitions, 1 × 1 × 1 mm, flip angle = 7°, TI = 1,000 ms, TE = 3.57 ms, TR = 8.4 ms) were acquired along with the fMRI sessions for cortical surface reconstruction (using FreeSurfer 5).

Data Analysis

Anatomical image processing

For each subject, the cortical surface was reconstructed with FreeSurfer [version 5; Dale et al., 1999] from the aligned average of the two high-resolution T1-weighted MPRAGE scans. Both mapping data and reading data employ a cross-subject surface-based analysis stream that begins by sampling responses and statistics to individual reconstructed cortices (cross-subject 3D averaging was not used at any point in the pipeline).

Analysis of phase-encoded mapping data

The first level single subject data from each run was motion corrected and registered to the last functional scan (across all runs) using AFNI's 3dvolreg [Cox, 2012] program (using heptic interpolation). The functional images were registered with the high resolution MPRAGE using a multi-stage registration pipeline. In the first stage, we generated a registration matrix to register the “alignment scan” (same block center and rotation as functional images, but only 3 min long) to the “high resolution” scan, since the contrast of these two scans was very similar. Each functional was registered to the “alignment scan” separately (these scans shared the same block center and rotation and this step was done to mainly correct any small variations between the four functional scans and the alignment scan). These registrations were done using FSL's FLIRT tool [version 5; Jenkinson and Smith, 2001; Jenkinson et al., 2002] and were then fine tuned using FreeSurfer's bbregister [Greve and Fischl, 2010]. The two registration matrices (functional -> alignment and alignment -> hi-res MPRAGE) were combined to generate the 4x4 registration matrix that aligns functional EPI to high-resolution MPRAGE (the matrix actually transforms 3D surface vertex coordinates to 3D EPI block index coordinates). The final registrations (EPI -> hi-res MPRAGE) were visually checked to ensure accuracy. The time courses from all the runs were averaged (after time-reversing the runs in the clockwise rotation for retinotopy, the upward frequency sweep for auditory, and the toe-to-tongue movement direction for the somatomotor map). The time reversed scans were time-shifted to compensate for hemodynamic delay before averaging. The averaged time courses were analyzed using linear Fourier methods [Bandettini et al., 1992; Engel et al., 1994, 1997; Hagler and Sereno, 2006; Sereno et al., 1995], which can be exactly recast as a general linear model. Voxels preferentially responding to a particular point in the stimulus cycle will show higher amplitude at the stimulus frequency than at any other 'noise' frequency, after excluding (i.e., linearly regressing out) the three lowest temporal frequencies as motion artifact. For retinotopic data, the phase of this vector at the stimulus frequency indicates the polar angle of the stimulus location. For auditory data, the phase indicates a particular point on the stimulus frequency ramp. For somatomotor data, this corresponds to the location of the moved body part. The individual and group analyses utilized a complex-valued cortical surface-based stream that was previously described [Hagler et al., 2007; Huang et al., 2012; Sereno et al., 1995] and briefly summarized below.

A fast Fourier transform was first performed on the average time courses of each voxel. An F-statistic value was obtained for each voxel by comparing the power at the stimulus frequency (eight cycles per scan) to the average power at the remaining frequencies after excluding the second and third harmonics of the stimulus frequency and one frequency above and below the first three harmonics. For individual subjects' activations illustrated below, the F-statistic was thresholded at P < 0.001 (corresponding to F(2,232)=7.12 for subject 1 who was scanned on standard EPI with 256 time points and F(2,488)=7 for subject 2 and subject 3 who were scanned on multiband EPI with 512 time points). Surface-based cluster size exclusion [Hagler et al., 2006] was used to correct for multiple comparisons with cortex surface clusters smaller than 30 mm2 excluded, achieving a corrected P-value of 0.01.

Group analysis of phase-encoded mapping data was performed using the methodology developed by Hagler et al. [2006] in which the real and imaginary components of the signal at the stimulus frequency were averaged across the subjects, preserving any phase information consistent across subjects (this is a vector average, which properly treats wrap around in the circular phase variable). This was performed by projecting each participant's complex-valued phase-encoded map to the FreeSurfer spherical atlas (using FreeSurfer mri_surf2surf), only performing ten steps of surface-based smoothing (∼3mm FWHM in 2D) before vector averaging across subjects at each vertex in the common surface coordinate system. Second-level surface-based cluster size exclusion [Hagler et al., 2006] was used to correct for multiple comparisons, with vertex level F-statistics thresholded at P < 0.01/P < 0.05 and cortical surface clusters smaller than 40 mm2/92 mm2 excluded, achieving a corrected p-value of 0.05. We used the fsaverage “inflated_avg” surface for display (made by averaging inflated surface coordinates) instead of the fsaverage “inflated” surface (made by averaging folded surface coordinates and then inflating the average) because inflated_avg represents original average surface area better. This is because folding variations (sulcal crinkles) are removed before surface-averaging, making inflated_avg, more appropriate for displaying surface-averaged data (mesh defects in the north and south icosahedral poles of FreeSurfer5's inflated_avg were corrected before using it for display. Corrected and flattened inflated_avg surfaces are available here: http://www.cogsci.ucsd.edu/~sereno/.tmp/dist/csurf/fsaverage-adds.tgz).

For Somatomotor mapping data, in addition to the above analysis, we reanalyzed the data utilizing the subject-ICA based noise reduction approach using ICA-AROMA [Pruim et al., 2015] to reduce the motion artefacts and activations due to B0 deformations. Each subject's raw data was pre-processed in FSL. Pre-processing involved motion correction with MCFLIRT, intensity normalization and spatial smoothing using a Gaussian kernel of 5 mm full width at half maximum. We then used ICA-AROMA to identify residual motion-related artefacts. The identified motion components were removed from the original unsmoothed raw-data (as smoothing the raw-data disrupts the map structure) by means of a linear regression using the FSL program fsl_regfilt. The resultant data set was used as input for the phase-encoded analysis pipeline described above. ICA-AROMA was successful in reducing the noise related activations and group results utilizing this technique are depicted in the somatomotor mapping results described below. The original analysis (without noise correction using ICA-AROMA) is also included in the Supporting Information Figures for comparison.

Analysis of Reading comprehension data

For the reading experiment, single subject fMRI data was motion corrected and skull stripped using FSL tools (MCFLIRT and BET). First level fMRI analysis was carried out by applying the General Linear Model (GLM) within FEAT using FILM prewhitening (FSL, version 5) with motion outliers (detected by fsl_motion_outliers) being added as confound regressors if there was more than 1 mm motion (as identified by MCFLIRT). A small number of scans with excessive motion above the threshold of 1 mm were excluded from analysis. High-pass temporal filtering of the data and the model was set to 100 seconds based on the power spectra of the design matrices (estimated by cutoffcalc; part of FSL). Three main explanatory variables were modeled and controlled: Reading English text, viewing Hindi text and viewing dot “text” (stimulus duration of each block: 16 seconds). Button press responses to target color change events were modeled as the fourth condition of interest (stimulus duration modeled as 1 second). The stimulus waveforms were convolved with FSL's double gamma “canonical” hemodynamic response function (HRF) [Glover, 1999] to generate the main regressors. In order to capture slight deviations from the model, temporal derivatives of all explanatory variables convolved with double gamma “canonical” HRF were included. The registration from functional to anatomical (6 DOF) and standard space (12 DOF) was first done using FSL's FLIRT and further optimized using boundary based registration (bbregister; FreeSurfer) similar to the procedure for the phase-encoded mapping data. A fixed effects analysis was performed across runs from an individual subject (usually four runs unless a run was excluded due to excessive motion/lack of attention as evidenced by poor performance on qualitative assessment/lack of response to the targets) to get group FEAT (GFEAT) results of first-level contrast of parameter estimates (COPEs) and their variance estimates (VARCOPEs) in the standard space. Across-subject group analysis was then carried out on the cortical surface using FreeSurfer tools. The GFEAT results of each subject were first sampled to individual cortical surfaces and then resampled to the spherical common average reconstructed surface (fsaverage). Surface-based spatial smoothing of 3 mm FWHM was applied on the icosahedral sphere. A mixed effects GLM group analysis was performed on the average surface using the mri_glmfit program from FreeSurfer. Surface-based cluster size exclusion utilized in phase-encoded mapping data was used to correct for multiple comparisons with vertex level T-statistics thresholded at P < 0.01, and cortical surface clusters smaller than 40 mm2 excluded, achieving a corrected P-value of 0.05. Finally corrected significance values (P < 0.05) of reading activation were displayed on the average surface.

The single subject raw data was not spatially smoothed (both for phase-encoded analysis and reading analysis). As with the phase-encoded data, a 10 step (∼3mm FWHM) surface level smoothing was applied for final illustration of the results. Hence, 3D Gaussian random field based cluster correction provided by FSL was not appropriate for multiple comparison correction of the reading data. We have instead used the surface-based cluster correction similar to that employed for phase-encoded data analysis. The GFEAT results were sampled to their respective anatomical surface, thresholded at P < 0.001 (Z = 3.09) and corrected for multiple comparisons with cortex surface clusters smaller than 30 mm2 excluded, achieving a corrected P-value of 0.01.

The target (font color change) presentation timings and the button press events were logged during the experiment and analyzed to assess the performance of the task. A target was considered as detected if there was a response (a key press event) within 1 second after the color change event had ended. For each participant, the number of targets detected and the mean response time in each condition were calculated. The cross-subject mean response time and target detection rates were assessed for significant differences across conditions.

Overlap analysis

All overlaps were calculated using “original vertex-wise area” in FreeSurfer. Original vertex-wise area in FreeSurfer is defined as the sum of 1/3 the area of each adjacent triangular face on the FreeSurfer “white” surface (refined gray/white matter boundary estimate). That single-vertex sum is not exactly constant across vertices because of slight non-uniformities in the final relaxed state of the surface tessellation. However, the sum of vertex-wise areas over a connected region of vertices exactly represents the summed original area of the enclosed triangles (plus the 1/3 fraction of triangles associated with the boundary vertices; along a straight edge of vertices, this last contribution corresponds to half of the area of the triangles just beyond the edge). The minimum areal increment that can be measured is roughly the average original vertex-wise area, which is ∼0.6 sq mm.


We first discuss the results obtained for each experiment in the study individually, followed by the overlap analysis results of reading activation with topological visual, auditory and somatomotor maps. A separate section is included to discuss several new and previously unreported sensory maps observed during the mapping experiments.

For the result figures corresponding to individual experiment results, we illustrate the amplitude of the vertex wise response for both reading experiment and phase-encoded maps (ignoring phase) for a stairstep of t-values. This is followed by figures illustrating single modality phase hue maps for sensory-motor maps where hue is used to indicate the map coordinates.

The overlap figures use transparent overlays to indicate reading activations over single modality phase hue maps, finally leading up to a summary outline figure containing all four kinds of data. The language contrast used for overlap analysis is English vs. Hindi. Brain activation observed for English vs. Hindi contrast partially overlapped with retinotopic, tonotopic and somatomotor maps. For clarity, in overlap figures, only positive activation after thresholding and cluster correction is shown. The overlap results for different modalities are illustrated for several individual subjects and then for the group as a whole. For the cross-subject average, the sensory-motor maps are illustrated for two separate vertex thresholds; P < 0.05 (lower threshold) and P < 0.01 (higher threshold), corrected for multiple comparisons using cluster thresholding at P < 0.05. Because the phase-encoded analysis effectively spreads the same amount of imaging data over a larger number of different effective conditions (e.g., different polar angles, sound frequencies, body parts) than the Reading experiment does, the regression analysis carried out on reading data will have more power. Hence the Reading data illustrated here uses a vertex threshold of P < 0.01 in all cross-average images. For the individual subjects, results are illustrated at a higher vertex threshold of P < 0.001, corrected to P < 0.01 for both reading and mapping data due to the higher amount of noise present in single subject data.

The individual subject data are illustrated here to show that in general the pattern of activity was similar to the cross-subject average. The majority of subjects showed reading activation patterns similar to subject-1 and subject-2, while subject-3 had more restrained reading activation. All three of the individual subjects illustrated here had superlative comprehension task performance (vivid post-scan explanations correctly citing most of the text concepts presented to them during the task); all were in their early 20's. Subject-1 was scanned using EPI sequence while subjects 2 and 3 were scanned using multiband sequence.

Among the 20 subjects who took part in reading experiment, data from 3 subjects were excluded from the analysis owing to unsatisfactory performance in the target detection task and/or assessment of poor comprehension following the session and/or excessive movement (>1 mm). The activation for the target detection regressor (button press regressor) was used as an extra quality check to decide whether the subject performed the task as per the instructions during each run. All our included subjects had comparable performance across conditions in the target detection task and motor activation for the target regressor. Among our 17 good subjects, three subjects were scanned using a single-slice EPI protocol and the remaining 14 subjects were scanned using multiband EPI, which had more complete brain coverage as well as higher signal-to-noise than the single-slice EPI data. However, the EPI protocol did not cover the entire brain, especially the anterior-ventral most portions of the temporal/frontal lobe (more so since participant's head was tilted forward for visual experiments as explained in experimental set-up). We did a further analysis of the reading data excluding the three single-slice EPI subjects. This revealed reading activation that continued up to the temporal pole as well as more anterior frontal activation. Hence for the group results of the Reading task, we have used data from the 14 multi-band subjects (N.B.: this region was more fully covered in the EPI auditory mapping experiments because of more standard head positioning). The original group result from the full 17 subjects is included as a Supporting Information Figure. Re-analysis of mapping data did not reveal any further coverage, and group results for mapping data include data from EPI subjects. Among the 17 (of the same) subjects who took part in somatomotor experiment, data from two subjects were excluded owing to high stimulus correlated head motion during the task. No exclusions were made for retinotopic and tonotopic experiments.

Target Detection Response

Figure 3 shows the target detection response results based on the performance of the included subjects. The average response time for the group when the target occurred in English, Hindi, Dot and Fixation conditions are depicted in Figure 3A. On average, participants took 0.47 seconds ± 0.01 (SEM) to respond to the target when it occurred in English/Hindi conditions and 0.42 seconds ± 0.01 (SEM) and 0.44 seconds ± 0.02 (SEM) when they occurred in Dot/Off conditions. A Wilcoxon matched-pairs signed-ranks test indicated no significant differences between “English” (median= 0.45 seconds) and “Hindi” (median = 0.46 seconds) (Z = −1.681, P = 0.098). The differences between “English” and “Dot” (median = 0.42 seconds) and “English” and “Off” (median = 0.40 seconds) conditions were found to be significant (Z = −2.96, P = 0.002 for English-Dot and Z = −2.02, P = 0.043 for English-Off).

Figure 3B depicts the average target detection success rate across all participants for different conditions. Across all runs, there were 20 targets for the English condition, 18 for the Hindi condition, 23 for the Dot and 7 for Off condition. All participants except one (see below) had comparable detection performance irrespective of the condition in which the target occurred. One participant did not respond to all seven targets that fell on the fixation, though she consistently performed in all other conditions. All other quality measures were satisfactory for this subject and her data suggested that she did fixate during Off condition (the fovea had no activation in English vs. Fixation). We have therefore conservatively considered her data as valid. On average, the mean success rate for detecting targets when they occurred in English, Hindi, Dot and Off were 91.6%, 89.1%, 87.5% and 86.6% respectively. The average for Off condition is less mainly because of one outlier, the above mentioned subject. Excluding her data, the average success rate for Off condition was 92%. A Wilcoxon matched-pairs signed-ranks test was carried out to assess statistical significance of the average success rate. The median success rate for English, Hindi, Dot and Off were 90%, 89%, 87% and 100% respectively. There were no significant differences between English and any of the other conditions (English-Hindi: Z = −1.733, P = 0.087; English-Dot: Z = −1.949, P = 0.051; English-Off: Z = −0.94, P = 0.94).

Reading Activation

Figure 4 and the top of Figure 5 illustrate the average cross-subject activation for a stairstep of t-values (all beyond a minimum threshold of P < 0.05, uncorrected) for each condition (English, Hindi and Dot) relative to fixation (Fig. 4A–C), and those for two main contrasts, Hindi vs. Dot (Fig. 4D), and English vs. Hindi (Fig. 5A). The reading activation illustrated in the group results come from 14 subjects who were scanned using the whole brain multiband sequence. Supporting Information Figure S1 shows the cross-subject results of all 17 subjects including the three subjects who were scanned using the single-slice EPI sequence. The main contrast used to assess reading comprehension is English vs. Hindi. In the later figures depicting overlap with sensory-motor maps, transparent white regions outlined in black depict the regions that showed significantly higher activation when reading English compared to Hindi. In all subjects, activation for English > Hindi was more widespread and pronounced in left hemisphere than in right hemisphere, as expected. The right hemisphere regions activated were a mirror image subset of the left hemisphere counterparts.

The cross-subject reading activation for English vs. Hindi (Fig. 5A) is spread around three main connected regions in the left hemisphere. The first region was anterior to the occipital pole, extending laterally into the occipital cortex with two “offshoots.” The first offshoot stretched across the intraparietal sulcus, covering regions in the inferior and superior parietal lobules. The second offshoot stretched along inferior temporal gyrus and occipito-temporal sulcus and extended ventrally onto the fusiform gyrus joining the medial activation that extended from calcarine fissure to the fusiform gyrus. Less extensive but prominent activation was observed in medial regions in the cuneus, precuneus and at the isthmus of the cingulate gyrus. For English, Hindi, and Dot (versus Fixation), roughly the same regions in the occipital and parieral lobe were activated. However, the activation differences were most significant in those regions for English vs. Hindi. The activation along occipito-temporal sulcus and fusiform gyrus was also mainly observed in the English condition.

The second main region (English vs. Hindi) is the superior temporal cortex, covering regions along the superior temporal gyrus and sulcus extending into the middle temporal gyrus and supramarginal gyrus. Except for a small region in posterior STS, the Hindi and Dot conditions do not have any significant temporal lobe activation.

The third region included two distinct frontal regions—one near the precentral sulcus and another more extensive region near the inferior frontal sulcus in the pars opercularis and pars triangularis region. The right hemisphere activation profile is similar but covering a much smaller total extent of cortical area. The activation near the precentral sulcus is present for Hindi and Dot conditions as well, with no significant differences in the Hindi vs. Dot contrast (Fig. 4D). The activation observed for English in the inferior frontal region near pars opercularis is absent in both Hindi and Dot conditions.

The reading activations of subject-1 and subject-2 were strikingly similar to the cross-subject profile, with activated regions corresponding to the three main regions just described in the left hemisphere. Individual subject-3 (see below) showed the greatest variation from the average, with more extensive activation observed in superior temporal sulcus and middle temporal gyrus, but then less extensive activation elsewhere in the cortex.

Retinotopic Maps

The retinotopic maps (amplitude in Fig. 5B, phase in Fig. 6A; see also phase underlays in Figs. 7-9) branch out anteriorly from the occipital pole into several “streams” (numbered 1–5 in black in Fig. 6A). In all phase maps, lower visual field is green, horizontal meridian is blue, and upper visual field is red (all contralateral). One stream extends through area MT into the superior temporal sulcus and reaches the posterior lateral sulcus (leaving a few disconnected regions in between, which join up as threshold is slightly lowered). A second stream stretches along the intraparietal sulcus and arrives at the superior part of the postcentral sulcus. A third stream spreads across the parieto-occipital sulcus (POS) into the medial posterior parietal cortex and precuneus, ending at the cingulate sulcus visual area [Huang and Sereno, 2013]. A fourth stream runs across the POS into retrosplenial cortex at the isthmus of cingulate gyrus (iCG) and continues to the edge of cortex just under the splenium of the corpus callosum. Finally, a fifth stream follows the collateral sulcus and fusiform gyrus into the ventral occipitotemporal lobe. A further disconnected set of retinotopic maps, including the frontal eye fields (FEF), frontal poly-sensory zone [Huang et al., 2012], and dorsolateral prefrontal cortex (DLPFC) are found in the frontal cortex. These maps are similar to those reported previously in Huang and Sereno [2013] using a similar stimulus. We also found a previously unreported visual map in the anterior cingulate region, referred to in Figure 6A as anterior cingulate visual area (ACv), which corresponds to the human dorsomedial eye-fields (discussed further under New maps).


Objects are often partially occluded by other objects and still perceived as coherent entities. It is still an unanswered question how the visual system assigns different image parts to figure and background. A well-known instance of boundary completion is taking place in the Kanizsa figure, in which a number of inducing elements, discs with a particular opening, are geometrically aligned such that an occluding surface is perceived on top of solid circles (Kanizsa, 1976 ; Figure 1 ). Due to proper geometrical alignment of the local inducers subjective boundaries are perceived as a collinear extension of the inducers. The contours are called subjective or illusory contours (IC) because there is no corresponding change in the physical stimulus intensity (e.g., luminance or texture) at the location of the apparent boundary. Hence subjective shapes provide us with the unique opportunity to study the cortical mechanisms that generate a contour percept in the absence of the distal stimulus.

Figure 1. Experimental stimuli and trial. (A) Kanizsa stimuli consisted of four inducers (note that stimulus depictions in the manuscript are contrast reversed), with openings (alpha) varying between ±15 around 90°. Openings larger than 90° resulted in convex shapes (two left-hand figures) and those smaller than 90° in concave shapes (two right hand figures). The ease of curvature discrimination depended on the magnitude of alpha. Large values resulted in pronounced curvature (outermost stimuli), whereas curvature was hardly detectable for small alpha, (innermost stimuli). (B) Participants performed the task inside the scanner. They were required to fixate the central cross during the entire experiment. Inducers (aligned or misaligned) were presented for 100 ms followed by a 250 ms presentation of a mask stimulus after an intervening blank period of 50 ms. Participants were required to discriminate between convex and concave shapes by pressing the right or left response button, respectively, within the remaining inter trial interval of 4084 ms. Auditory feedback (1000 Hz tone of 300 ms duration) followed each correct response.

Electrophysiological studies in non-human primates have shown that neurons in early visual cortex (V1/V2) respond to subjective contours (Grosof et al., 1993 ; Lee and Nguyen, 2001 ; Peterhans and von der Heydt, 1989 ; Ramsden et al., 2001 ; Sheth et al., 1996 ; Sugita, 1999 ). Some of the functional imaging studies with human subjects reported IC-sensitivity predominantly in higher visual areas (Hirsch et al., 1995 ; Mendola et al., 1999 ; Murray et al., 2002 ; Ritzl et al., 2003 ; Stanley and Rubin, 2003 ) like the Lateral Occipital Complex (LOC), which has been shown to respond preferentially to increasingly complex, object-like stimuli (see Grill-Spector and Malach, 2004 ; Grill-Spector et al., 2001 for review). Other studies reported IC-related activity also in earlier extrastriate areas (Ffytche and Zeki, 1996 ; Hirsch et al., 1995 ; Larsson et al., 1999 ; Ritzl et al., 2003 ) or even in primary visual cortex (Maertens and Pollmann, 2005 ; Montaser-Kouhsari et al., 2007 ; Seghier et al., 2000 ). Using functional magnetic resonance imaging (fMRI), Montaser-Kouhsari et al. (2007) observed neural adaptation to illusory contours in abutting gratings in most of the visual areas. Seghier et al. (2000) reported fMRI BOLD changes in response to moving Kanizsa-type subjective contours. We compared pre- vs. post-learning BOLD responses to Kanizsa-type subjective contours in a perceptual learning task (Maertens and Pollmann, 2005 ) and observed significant learning related changes in V1. Furthermore, we showed that in the absence of the analogous V1 representation, which can be simulated under monocular viewing conditions in the ‘blind spot’ region in V1, curvature discrimination with subjective figures is severely impaired (Maertens and Pollmann, 2007 ).

In the present experiment, we wanted to study whether responses to subjective contours in V1 do indeed follow the retinotopic representational pattern of real contours. In our earlier learning experiment (Maertens and Pollmann, 2005 ) we compared BOLD responses to subjective contours before and after training. In the current experiment we introduce a control configuration which is almost identical to the Kanizsa shape except that it does not contain a subjective shape and subjective boundaries because the inducers’ mouths are rotated outwards. We randomly intermixed Kanizsa and control stimuli in order to probe spatially specific for subjective contour evoked responses. We adopted a curvature discrimination paradigm which – in order to yield accurate performance – requires a spatially precise representation of the subjective contour (Ringach and Shapley, 1996 ).

We reasoned that in order to detect slight differences in subjective contour curvature, the subjective contour should be represented by neural populations that provide the appropriate spatial resolution (e.g., small receptive field size, retinotopic organization, orientation selectivity). The reasoning follows the psycho-anatomic matching logic (Julesz, 1971, cf. Hochstein and Ahissar, 2002 ), that different degrees of performance accuracy may be indicative of different cortical processing levels. Since primary visual cortex has the most precise retinotopy and the smallest average receptive field sizes, we reasoned that the disambiguation of slight differences in subjective contour curvature would heavily rely on a neural representation of the subjective contour in V1.

Going beyond previous work, we investigated the retinotopic specificity of subjective contour-related BOLD changes in V1. In contrast to Montaser-Kouhsari et al. (2007) we used collinear Kanizsa-type subjective contours to ensure that BOLD responses to inducers and subjective stimulus parts are spatially separable. We performed a localizer scan in order to determine the putative locations of subjective contours and inducers within primary visual cortex. We hypothesized that if subjective contours are represented retinotopically within V1, then Kanizsa stimuli should elicit stronger responses than control stimuli at the putative contour locations. As a control we compared BOLD responses to Kanizsa and control stimuli at the inducer locations, because they should be identical. We also varied the center-to-center distance between inducers of constant size, as it has been shown that subjective ratings of contour clarity (Kellman and Shipley, 1992 ) as well as performance-based measures increase with increasing contour support (Ringach and Shapley, 1996 ). We therefore predicted that if the perception of contour clarity depends on the representation of the subjective contour in V1, then the BOLD response will become stronger with increasing contour support.

In order to obtain sufficient spatial specificity of the functional MRI signal we applied a spin-echo echo-planar-imaging-sequence (SE-EPI) at a magnetic field strength of 3 Tesla. In spin-echo compared to the more common gradient-echo (GE) EPI experiments, the spin dephasing due to T2* effects is refocused, so that at the time of signal readout (time-to-echo, TE) the only loss in transverse magnetization is due to T2. Spin echo imaging is therefore insensitive to susceptibility artifacts, which are caused by magnetic field inhomogeneities (Jezzard et al., 2001 ). Of even greater importance for the present study was that the SE-BOLD signal is predominantly sensitive to extravascular water surrounding capillaries, in addition to being sensitive to intravascular water spins in vessels of all sizes. Using flow-compensating diffusion-weighting at 3T the former can be effectively reduced leaving exclusively signal contributions from the capillaries (Jochimsen et al., 2004 ; Norris et al., 2002 ). In this way, we attained the high spatial specificity necessary to demonstrate spatially specific activation changes in response to illusory contours within the primary visual cortex. The high spatial specificity of SE-EPI comes at the cost of a reduced overall functional sensitivity. We counteracted this sensitivity loss by the use of a phased array coil, which has an improved sensitivity compared with a volume (head) coil.

Materials and Methods


Fourteen observers were paid for their participation in one 60 minutes scanning session. Half of them were female. Their mean age was 25 years (SD = 2.5). All participants were right-handed and had normal or corrected-to-normal vision. Participants gave their informed written consent according to the guidelines of the Max-Planck-Institute.

Stimuli and Design

Stimuli were presented at a resolution of 800 × 600 pixels on a 16" back-projection screen, mounted in the bore of the magnet behind the participant’s head, using a liquid crystal display (LCD) projector. Participants viewed the screen by wearing mirrored glasses. Four white inducers were presented tachistoscopically on a black background. Inducers varied in the magnitude of their openings in order to create convex (‘fat’) and concave (‘thin’) shapes. We used openings (α) which were either larger or smaller than 90° (75 < α < 105), and which had to be classified as ‘fat’ or ‘thin’, accordingly (Figure 1 A). Inducers of constant size were presented at two different peripheral eccentricities (see below). Their diameter was 3.5 cm corresponding to 2.4° visual angle. Four white pinwheel masks, consisting of four wedges of 45° each, were presented at the inducer positions to limit their effective viewing time (Figure 1 B).

Stimulus delivery and response registration were controlled by Presentation® software (Version 9.51, http://nbs.neuro-bs.com). A trial started with a 500 ms fixation period, followed by stimulus presentation for 116 ms. After a blank period of 50 ms a mask was presented for 250 ms (Figure 1 B). No time limit was imposed on participants’ response. They indicated whether they had perceived a concave or convex shape by pressing either the left or the right button of a two-button response pad, with their index or middle finger, respectively. Trial duration was fixed at five seconds. Observers were instructed to maintain fixation during the scans, and due to the rapid presentation voluntary eye movements are highly unlikely. Kanizsa and control stimuli were constructed from inducers that either were aligned to form an illusory square, e.g., their openings were facing inwards, or they were misaligned, e.g., their openings were facing to the margins of the screen. Stimuli did also vary with respect to the distance between the inducers. The center-to-center distance between the inducers was 7.4°, in the near, or 11.2° visual angle, in the far condition, yielding support ratios of 0.32 and 0.21, respectively. The support ratio (SR) describes the ratio between the luminance-defined part of the illusory contour and its total side length.


All participants performed a training block consisting of twenty trials with the experimental stimuli outside the scanner. The training block was repeated until at least 15 out of 20 trials were answered correctly. During scanning, participants’ curvature discrimination thresholds were measured, using a weighted up-down method for two response alternatives (Kaernbach, 1991 ). The starting inducer angle (alpha) was randomly set to the maximum deviation of either + or –15°. Each wrong response entailed an increment of 6° to make the deviation from 90° more pronounced. After the first three reversals the step size was decreased to 3°. Each correct response was followed by a reduction in the angular deviation from 90° by an amount of 2°. Again, after the first three reversals this step size was reduced to 1°. The staircase converged on a 75% correct response level. Two functional scanning blocks were performed each involving a quadruple staircase procedure (one for each combination of inducer distance and orientation), that was terminated after 25 trials of each condition. In addition, ten baseline 8 trials (null-events) were interspersed within the resulting 100 experimental trials, in which only the fixation cross was presented for five seconds without any response requirements.

fMRI Methods

Magnetic resonance (MR) imaging was performed on a 3T Siemens Magnetom Trio scanner using an eight-channel phased-array head coil. First, a T1-weighted anatomical scan was recorded with a modified driven equilibrium Fourier transform (MDEFT) sequence (Norris, 2000 ). Twelve axial slices were recorded which were aligned in parallel to the calcarine sulcus, to cover the visual cortex. The sequence parameters were as follows: TR = 1300 ms, TI = 650 ms, TE = 7.4 ms, slice-thickness = 2 mm, slice-gap = 0.4 mm, FOV = 128 mm × 128 mm with an in-plane resolution of 2 mm × 2 mm. Oversampling in the phase-encoding direction was applied to achieve the desired in-plane resolution (‘zoom-EPI’) and to remove any fold-in signal from outside the FOV. The functional part of the session consisted of three spin-echo EPI scans (TR = 2 seconds, TE = 85 ms, bandwidth = 1346 Hz/Px, matrix 64 × 64, phaseoversampling).

The functional slices were aligned as in the anatomical scan using the same FOV, slice-thickness and slice-gap. Two experimental functional scans with 310 repetitions were performed using the experimental paradigm as described above, as well as one functional localizer scan containing 318 repetitions (see below).

Statistical analysis was carried out using FEAT (FMRI Expert Analysis Tool) Version 5.63, part of FSL (FMRIB’s Software Library, www.fmrib.ox.ac.uk/fsl). Movement artifacts were corrected using MCFLIRT (Jenkinson et al., 2002 ). Differences in slice acquisition time were corrected using Fourier-space time-series phase-shifting. Potential baseline signal drifts were removed applying a high pass temporal filter (Gaussian-weighted least-squares straight line fitting, with sigma = 50.0 seconds). In the spatial domain data were smoothed using a Gaussian kernel with a full width at half maximum (FWHM) of 4 mm. All volumes were mean-based intensity normalized by the same factor. Time-series statistical analysis was carried out using FILM (Woolrich et al., 2001 ). For the localizer scan z (Gaussianized T/F) statistic images were thresholded using clusters determined by z > 2.3 and a corrected cluster significance threshold of p = 0.05 (Worsley et al., 1992 ).

In order to determine regions of interest, the functional data from the localizer scan were individually registered onto their corresponding high resolution anatomical data (1 × 1 × 1 mm) using FLIRT (Jenkinson and Smith, 2001 ): (i) the functional slices were geometrically aligned with the 2D MDEFT images using a rigid body transformation with 3 parameters, (ii) the 2D MDEFT slices were aligned with the 3D anatomy using a rigid, linear transformation with three translational and three rotational parameters, (iii) the resulting transformation matrices were concatenated using ConvertXfM (FLIRT), and finally the output matrix was applied to the functional data.

Stimulus Localizer, Retinotopy and Flattening

The purpose of the localizer was (a) to map the cortical sites responsive to illusory contours and inducers at the two stimulus eccentricities employed in the main experiment, and (b) to determine the border between visual areas V1 and V2. All stimuli were black-and-white checkerboard patterns presented on a black background that reversed contrast at a rate of 8 Hz. The inducer localizer was composed of four circles that had a diameter of 2.4° visual angle (equivalent to that of the inducers) and their centers were aligned with the center positions of the inducers in the main experiment for both, the near and far conditions. In the contour localizer four checkered bars (0.6° wide and 5° or 8.5° long), aligned as to form a square with missing edges, were presented at 3.7° or 5.5° from fixation in order to stimulate the putative illusory contour representations in the near and far condition, respectively, complementary to the inducer positions. In addition, we mapped the horizontal (HM) and vertical meridians (VM) using alternating ‘hourglass’ and ‘bow tie’ shaped checkerboard patterns that were perspectively scaled to account for cortical magnification. Circles and bars, as well as the meridian mapping stimuli were presented for eight seconds each, followed by an eight seconds fixation baseline. One full cycle thus lasted 96 seconds and might have consisted of the following sequence: near circles – far contours – horizontal meridian – vertical meridian – near contours – far circles. A run was composed of 6 1/2 cycles whereby data from the first 1/2 cycle were discarded to avoid magnetic saturation effects, and a break of 30 seconds was introduced between the third and the fourth cycle. Periods of stimulation were contrasted with fixation periods to demarcate occipital visual area V1 (Engel et al., 1997 ; Sereno et al., 1995 ), and to localize the stimulus positions (Figure 2 ). Brain inflation was performed following standard procedures as implemented in the Computerized Anatomical Reconstruction and Editing Toolkit (Caret, Van Essen et al., 2001 ; http://brainmap.wustl.edu/caret).

Figure 2. Examples of individual z-maps projected onto inflated anatomical surfaces. The figures depict a medial view upon the inflated calcarine sulcus within the right and left hemisphere of subject #1 (upper row) and #2 (lower row), respectively. The left column shows an overlay of the two z-maps capturing the contrast between horizontal (blue) and vertical (white) meridian stimulations vs. baseline (5.0 < z < 7.0). The right column shows the z-maps capturing the contrasts between contours (red-yellow) and inducers (yellow-white) vs. baseline (3.0 < z < 5.0). In addition, white outlines are drawn around the inducer ROIs and red outlines are drawn around the corresponding contour ROIs. Black lines mark the horizontal (HM) and vertical (VM) meridian mappings with the VM meridian demarcating the dorsal and ventral V1-V2 border. One can see that there are additional regions responsive to the contours which coincide with the vertical meridian representation. These are the responses to the upper and lower horizontal contours, which crossed the vertical meridian.

We visualized z-maps from the localizer scan overlaid on the inflated (instead of totally flattened) cortical surface in order to retain some topographic information. At first, we displayed the thresholded z-maps (z > 3.0) for horizontal and vertical meridians overlaid on the inflated anatomy of each subject (Figure 2 ). The primary visual cortex was defined as the cortical region enclosing the horizontal meridian representation along the fundus of the calcarine sulcus and restricted by the closest dorsal and ventral vertical meridian representations. We used this map in order to determine two regions of interest (ROI) within the primary visual cortex: contour representations and inducer representations. In order to characterize these ROIs we selected the single voxel that was maximally responding to either the contour or to the inducer localizer in both hemispheres of individual observers. We had to exclude five out of 14 participants due to a lack of response strength in the localizer scan. That means in these participants it was impossible to identify our ROI, e.g., either the meridians or the stimulus locations, and hence they could not be included in the analysis.


Psychophysical Performance

Discrimination thresholds were computed by averaging the alpha-values (angular deviation from 90°) over the final 10 trials of each condition in each of the two blocks. These thresholds indicate the 75% accuracy level (see Methods). Figure 3 displays the mean curvature discrimination thresholds as a function of inducer orientation (aligned vs. misaligned) and inter-inducer distance (near vs. far) averaged over the selected nine participants. A 2 × 2 repeated measures ANOVA was calculated for the thresholds with the factors inducer orientation (aligned, misaligned) and inducer distance (near, far). A significant main effect was observed for the inducer orientation F(8) = 5.76, p = 0.04) as performance thresholds were much lower in the aligned than in the misaligned condition (Figure 3 ), whereas no difference was observed between the near and far stimuli.

Figure 3. Mean thresholds. Shown are 75% performance thresholds in units of angular deviation from 90° averaged across nine participants as a function of inducer distance (x-axis) and alignment (differently colored bars). Error bars indicate the standard error of the mean for each condition.

Functional Imaging: IC-Related BOLD Signals in the Primary Visual Cortex

We localized our regions of interest in each individual subject (see Materials and methods section). The mean contour and inducer ROIs averaged across the remaining nine participants are depicted in the upper row diagrams of Figures 4 and 5 . We then extracted the spatially and temporally smoothed peristimulus fMRI response evoked by near and far aligned and misaligned stimuli from the voxels of interest using PyNIfTi (http://niftilib.sourceforge.net). Event-related signals spanning a time window of 12 seconds were averaged time-locked to stimulus onset. We included correct and incorrect trials alike. A percent signal change measure was calculated using the following formula: psc = [S(i) – S(1)] × 100/S(1) with S(i) referring to the BOLD signal intensity at time step i after stimulus onset and S(1) referring to the BOLD signal intensity at the first time step. Means and standard errors of the mean were calculated across nine participants at the inducer and contour ROIs in each hemisphere.

Figure 4. Mean ROI locations and peristimulus plots at ROIs in the far condition. Upper part: Shown are ROI locations of inducers (blue) and contours (red) in the left (LH) and right hemisphere (RH) averaged across nine participants. The plots are analogous to a coronal view with the x-axis representing the left-right and the y-axis representing the ventral-dorsal axis. Error bars show the standard error of the mean, the corresponding MNI coordinates in y-direction are written down close to the data points. Lower part: Plots depict the evoked BOLD changes locked to stimulus onset and spanning a time window of 12 seconds. The upper two graphs show the peristimulus plots at the contour ROIs and the lower two graphs those at the inducer ROIs. Left and right columns show the peristimulus plots in the left and right hemisphere. Error bars represent the standard error of the mean.

The diagrams in Figures 4 and 5 reveal that the averaged BOLD responses follow the predicted interaction pattern between the stimulus condition and the cortical region of interest: Aligned stimuli evoked stronger BOLD responses than misaligned stimuli at the cortical IC representation, i.e., in striate cortex near the fundus of the calcarine sulcus. In contrast, BOLD activation changes evoked by aligned and misaligned inducers did not markedly differ at the striate locations responsive to inducers. These observations were confirmed by 2 × 2 × 3 repeated measures ANOVAs with the factors ROI (contour vs. inducer) inducer orientation (aligned vs. misaligned) and time step (3, 4, 5) that were performed separately for the near and far condition. Both ANOVAs revealed a main effect for inducer alignment [F(1,8) = 16.46, p = 0.004 and F(1,8) = 24.71, p = 0.001 in the near and far conditions, respectively] indicating that aligned stimuli elicited stronger BOLD changes than misaligned stimuli, and a significant interaction between ROI location and inducer alignment [F(1,8) = 22.28, p = 0.002 and F(1,8) = 27.23, p = 0.001 in the near and far conditions, respectively]. This interaction reflected that BOLD responses to aligned and misaligned stimuli were significantly different from each other at the contour ROI [t_near(8) = 5.10, p < .001, t_far(8) = 5.17, p < .001], but not the inducer ROI.


We compared fMRI BOLD responses in primary visual cortex to subjective contours and control stimuli in areas defined by their responsiveness to physically defined contours and inducers. We had previously shown that retinotopic regions in V1 that respond to real contours are also activated in response to subjective contours after some amount of training (Maertens and Pollmann, 2005 ). Here, we observed an increased BOLD response to subjective contour but not to control stimuli (Figures 4 and 5 ) in primary visual cortex at locations that were defined by their response to analogous real contours. Primary visual cortex was functionally defined as the area along the calcarine sulcus that extended to the first dorsal and ventral vertical meridian representations (Figure 2 ). Activation changes were individually checked to originate from striate cortex by comparing the stimulus ROIs with the meridian mappings. In contrast to the activation pattern at the contour ROI, activation changes evoked by aligned and misaligned inducers did not elicit differential responses at the striate locations responding maximally to the inducers.

The response amplitude in the current study as measured by the percent signal change may appear comparably small, e.g., the maximum response to aligned stimuli at the inducer location does not exceed 14%, whereas in our earlier experiment the maximum response change was about 20%. This may be due to the known reduced functional sensitivity of SE compared with GR-EPI. However, in the absence of large amplitude changes we still obtained a sufficiently high signal to noise ratio to yield significant results. Somewhat unexpectedly, the BOLD signal change in response to the inducers was comparatively small. We suspect that this was due to a temporal overlap between BOLD responses in subsequent trials, which resulted from a larger than expected spatial overlap between inducers in the near and far condition. Even though the activation peaks were separable (Figures 4 and 5 ) there was a considerable overlap between responses to near and far inducers. It should be noted that this overlap only affected the inducers, which were present in every trial, but not the activation changes in response to subjective contours, i.e., the comparison between trials with inward- and outward-bound inducers.

There were also no performance differences between the near and far conditions. Even though it has been reported that the perceived strength of subjective contours depends on the support ratio (Kellman and Shipley, 1992 ; Ringach and Shapley, 1996 ), we did not find any difference between illusory figures with support ratios of .32 and .21 with the parameters used here. Since we did not observe an effect of the support ratio on the behavioral level, there was also no point in checking for differences in BOLD responses to high and low support ratio conditions. Thus, it remains an issue of further study to determine whether there is a quantitative relationship between subjective contour clarity and BOLD responses in V1 or higher visual areas.

Our findings are in apparent contradiction to some of the previous human imaging studies (Hirsch et al., 1995 ; Mendola et al., 1999 ;Murray et al., 2002 , 2004 ; Ritzl et al., 2003 ; Stanley and Rubin, 2003 ) that did not find indications for subjective contour processing in V1. However, in these studies participants were required to either passively view the stimulus display (Hirsch et al., 1995 ; Mendola et al., 1999 ; Stanley and Rubin, 2003 ) or to detect the presence or absence of a global stimulus shape (Murray et al., 2002 , 2004 ; Ritzl et al., 2003 ). Our task required participants to specifically focus on the subjective contour compared to the entire subjective figure. It has been shown that the neural processes underlying subjective contour- and figure perception are dissociable (Stanley and Rubin, 2003 ) and might involve different neural populations. Whereas neurons at higher levels in the processing hierarchy might respond more to the figural aspects of the Kanizsa stimulus, neurons in early visual cortex provide the higher spatial precision that may be required to resolve differences in the exact path of the subjective boundary. So, paradigms that focus on different aspects of the stimulus are likely to yield different results regarding the underlying neural mechanisms. In addition, it has been suggested that in the process of scene segmentation surfaces or ‘salient regions’ are identified before their exact boundaries (Stanley and Rubin, 2003 ). In other words the default mode of processing is crude and limited to figural information (Hochstein and Ahissar, 2002 ). Hence, without the particular requirement to discriminate fine spatial differences, subjective contours were unlikely to evoke sufficiently strong activity in V1 to be detected with fMRI. With more sensitive measures it might be possible to observe these responses even in the absence of a particular task (Murray et al., 2006 ).

It is unlikely that our results are attributable to differential effects of attentional allocation (e.g., Noesselt et al., 2002 ). Since Kanizsa and control stimuli were randomly interleaved, subjects could not know in advance whether attention should be directed to the subjective contour or to the inducers. Now if, regardless of the actual condition, participants were attending toward the subjective contour we should have seen the effect of attention in the subjective contour and in the control condition. However, we found stronger responses only in the subjective contour condition. We can’t of course exclude the possibility that subjective contours are drawing attention, however then attention would only be the consequence of a subjective contour percept and might act to amplify an already evoked subjective contour response.

It remains to be explained how – in the absence of a physical stimulus to the retina and the lateral geniculate nucleus – the subjective contour responses in V1 have been generated. We suggest, as others have done before (Maertens and Pollmann, 2005 ; Murray et al., 2004 ; Lee and Nguyen, 2001 ), that the responses to subjective contours in V1 result from two sources: On the one hand, there are probably lateral interactions between V1 neurons that were excited by the inducers. The known range of lateral connections within V1 is however considerably smaller than the range across which subjective contours have been interpolated in the current experiment (e.g., Angelucci and Bullier, 2003 ). Therefore, it is plausible to assume that feedback from extrastriate neurons, which pool information across larger regions of space, also plays an important role in the process of boundary interpolation (Lee and Nguyen, 2001 ; Maertens and Pollmann, 2005 ; Murray et al., 2004 ; Stanley and Rubin, 2003 ). In several studies BOLD responses to subjective figure stimuli were observed in regions in the lateral occipital cortex (Hirsch et al., 1995 ; Mendola et al., 1999 ; Murray et al., 2002 ; Ritzl et al., 2003 ; Stanley and Rubin, 2003 ), which are known to preferentially respond to object-like stimuli (e.g., Malach et al., 1995 ). However, activity in these extrastriate regions alone would not be sufficient in itself, because as stated in detail above, it is the functional properties of V1 neurons that enable a crisp subjective contour percept. Hence one could think about a scenario in which neurons in higher visual areas, which respond to the subjective figure, send feedback to retinotopically specific lower visual areas, in which contour responses are being fine-tuned by lateral interactions.

Conflict of Interest Statement

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.


The work was supported by a Gertrud Reemtsma fellowship to Marianne Maertens.


Angelucci, A., and Bullier, J. (2003). Reaching beyond the classical receptive field of V1 neurons: horizontal or feedback axons? J. Physiol. Paris 97, 141–154.

Pubmed Abstract | Pubmed Full Text | CrossRef Full Text

Engel, S. A., Glover, G. H., and Wandell, B. A. (1997). Retinotopic organization in human visual cortex and the spatial precision of functional MRI. Cereb. Cortex 7, 181–192.

Pubmed Abstract | Pubmed Full Text | CrossRef Full Text

Ffytche, D. H., and Zeki, S. (1996). Brain activity related to the perception of illusory contours. Neuroimage 3, 104–108.

Pubmed Abstract | Pubmed Full Text | CrossRef Full Text

Grill-Spector, K., Kourtzi, Z., and Kanwisher, N. (2001). The lateral occipital complex and its role in object recognition. Vision Res. 41, 1409–1422.

Pubmed Abstract | Pubmed Full Text | CrossRef Full Text

Grill-Spector, K., and Malach, R. (2004). The human visual cortex. Annu. Rev. Neurosci. 27, 649–677.

Pubmed Abstract | Pubmed Full Text | CrossRef Full Text

Grosof, D., Shapley, R. M., and Hawken, M. J. (1993). Macaque V1 neurons can signal ‘illusory’ contours. Nature 365, 550–552.

Pubmed Abstract | Pubmed Full Text | CrossRef Full Text

Hirsch, J., DeLaPaz, R. L., Relkin, N. R., Victor, J., Kim, K., Li, T., Borden, P., Rubin, N., and Shapley, R. (1995). Illusory contours activate specific regions in human visual cortex: evidence from functional magnetic resonance imaging. Proc. Natl. Acad. Sci. U.S.A. 92, 6469–6473.

Pubmed Abstract | Pubmed Full Text | CrossRef Full Text

Hochstein, S., and Ahissar, M. (2002). View from the top: hierarchies and reverse hierarchies in the visual system. Neuron 36, 791–804.

Pubmed Abstract | Pubmed Full Text | CrossRef Full Text

Jenkinson, M., Bannister, P., Brady, M., and Smith, S. (2002). Improved optimisation for the robust and accurate linear registration and motion correction of brain images. Neuroimage 17, 825–841.

Pubmed Abstract | Pubmed Full Text | CrossRef Full Text

Jenkinson, M., and Smith, S. A. (2001). A global optimisation method for robust affine registration of brain images. Med. Image Anal. 5, 143–156.

Pubmed Abstract | Pubmed Full Text | CrossRef Full Text

Jezzard, P., Matthews, P. M., and Smith S. M. (2001). Functional MRI: An Introduction to Methods. Oxford, Oxford University Press.

Pubmed Abstract | Pubmed Full Text

Jochimsen, T. H., Norris, D. G., Mildner, T., and Moller, H. E. (2004). Quantifying the intra- and extravascular contributions to spin-echo fMRI at 3T. Magn. Reson. Med. 52, 724–732.

Pubmed Abstract | Pubmed Full Text | CrossRef Full Text

Kaernbach, C. (1991). Simple adaptive testing with the weighted up-down method. Percept. Psychophys. 49, 227–229.

Pubmed Abstract | Pubmed Full Text

Kanizsa, G. (1976). Subjective contours. Sci. Am. 234, 48–52.

Pubmed Abstract | Pubmed Full Text

Kellman, P. J., and Shipley, T. F. (1992). Strength of visual interpolation depends on the ratio of physically specified to total edge length. Percept. Psychophys. 52, 97–106.

Pubmed Abstract | Pubmed Full Text

Larsson, J., Amunts, K., Gulyas, B., Malikovic, A., Zilles, K., and Roland, P. E. (1999). Neuronal correlates of real and illusory contour perception: functional anatomy with PET. Eur. J. Neurosci. 11, 4024–4036.

Pubmed Abstract | Pubmed Full Text | CrossRef Full Text

Lee, T. S., and Nguyen, M. (2001). Dynamics of subjective contour formation in the early visual cortex. Proc. Natl. Acad. Sci. U.S.A. 98, 1907–1911.

Pubmed Abstract | Pubmed Full Text | CrossRef Full Text

Maertens, M., and Pollmann, S. (2005). FMRI reveals a common neural substrate for illusory and real contours in V1 after perceptual learning. J. Cogn. Neurosci. 17, 1553–1564.

Pubmed Abstract | Pubmed Full Text | CrossRef Full Text

Maertens, M., and Pollmann S. (2007). Illusory contours do not pass through the ‘blind spot’. J. Cogn. Neurosci. 19, 91–101.

Pubmed Abstract | Pubmed Full Text | CrossRef Full Text

Malach, R., Reppas, J., Benson, R., Kwong, K., Jiang, H., Kennedy, W., Ledden, P., Brady, T., Rosen, B., and Tootell, R. (1995). Object-related activity revealed by functional magnetic resonance imaging in human occipital cortex. Proc. Natl. Acad. Sci. U.S.A. 92, 8135–8139.

Pubmed Abstract | Pubmed Full Text | CrossRef Full Text

Mendola, J. D., Dale, A. M., Fischl, B., Liu, A. K., and Tootell, R. B. (1999). The representation of illusory and real contours in human cortical visual areas revealed by functional magnetic resonance imaging. J. Neurosci. 19, 8560–8572.

Pubmed Abstract | Pubmed Full Text

Montaser-Kouhsari, L., Landy, M., Heeger, D. J., and Larsson, J. (2007). Orientation-selective adaptation to illusory contours in human visual cortex. J. Neurosci. 27, 2186–2195.

Pubmed Abstract | Pubmed Full Text | CrossRef Full Text

Murray, M. M., Foxe, D. M., Javitt, D. C., and Foxe, J. J. (2004). Setting boundaries: brain dynamics of modal and amodal illusory shape completion in humans. J. Neurosci. 24, 6898–6903.

Pubmed Abstract | Pubmed Full Text | CrossRef Full Text

Murray, M. M., Imber, M. L., Javitt, D. C., and Foxe, J. J. (2006). Boundary completion is automatic and dissociable from shape discrimination. J. Neurosci. 26, 12043–12054.

Pubmed Abstract | Pubmed Full Text | CrossRef Full Text

Murray, M. M., Wylie, G. R., Higgins, B. A., Javitt, D. C., Schroeder, C. E., and Foxe, J. J. (2002). The spatiotemporal dynamics of illusory contour processing: combined high-density electrical mapping, source analysis, and functional magnetic resonance imaging. J. Neurosci. 22, 5055–5073.

One thought on “Fsl Retinotopic Mapping Assignment

Leave a comment

L'indirizzo email non verrà pubblicato. I campi obbligatori sono contrassegnati *