Main fMRI experiment: Three-dimensional flow fields (3D-FF).

In the main event-related fMRI experiment, hereafter called 3D-FF experiment, we separately investigated five different motion components: radial, circular, translational, spiral and random motion. To this aim we created a wide field visual display of moving dots with a flow structure that was able to evoke different types of egomotion. Our stimuli were “star fields” composed by high-contrast light dots on a dark background, three-dimensionally designed to give the impression to the observers to be moving in the 3D space. Both speed and size were logarithmically scaled with eccentricity (i.e., as a function of the distance from the center of the display) as when a person moves through a 3D environment. Most previous imaging studies have used displays filling only the central visual field, which do not recreate the wide-field motion typically found in a real scene. The extent of the stimulus field is a crucial parameter which determines the intensity of vection (e.g., [15], [16]). Here, with a wide field stimulation, the subject felt to be immersed in the flow patterns and experienced vivid sensations of egomotion as those experienced in a 3D environment. Star fields simulated various flow patterns, shown in Figure 1 and described as follows:

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 1. Schematic representation of the stimuli we used in the main fMRI experiment (ff-3D).

https://doi.org/10.1371/journal.pone.0060241.g001

In the translational motion condition, all dots moved horizontally, rightward or leftward, with a different speed according to the different dot size (i.e. the different apparent distance), simulating an observer translating horizontally (such as when a person looks through the window of a moving train).

In the circular motion condition, all dots moved in concentric circular paths to produce an impression of clockwise or counterclockwise rotation, consistent with rotation of the observer around the line of sight.

In the radial motion condition, all dots moved outwards or inward along the radii of an hypothetical circle to produce an impression of expansion or contraction, respectively. This flow pattern is consistent with movement of the observer forward or backward along the line of sight.

In the spiral motion condition, the dots moved as in the radial flow field condition with an added rotational component, resulting in a spiral motion. The rotational component was either clock- or counterclockwise and the global flow pattern changed over time.

Global patterns of optic flow were produced by controlling the local motion directions of the dots. In the four coherent 3D flow patterns, dot patterns inverted direction every 500 ms along either radial (in and out), circular (cw and ccw), translational (left and right) or spiral (cw/in and ccw/out) trajectories. The fast inverting time along a specific direction was chosen to be identical to that used in the flow fields stimulus used to localize area V6 [5].

In the random condition, the moving dots changed their local direction and speed at random (every 500 ms), like the random movements of a fly in a 3D box. The purpose was to provide a condition in which local motion was present in all directions and at all locations in a 3D environment, with no global flow structure (incoherent 3D motion).

Original observations in macaque showed the presence in the motion areas V6, MT and MSTd of classes of cells responding to different speeds, from very low (about 1°/s) to very high (more than 100°/s; [17]–[21]). Thus, to optimally activate these motion areas, we used a range of velocities instead of a single speed. In the attempt to activate as many speed-sensitive neurons as possible with a single stimulus we used three different average velocities of about 18°/s, 30°/s and 50°/s.

Since the speed of our stimuli were logarithmically scaled with eccentricity, to simulate real depth, in the radial and spiral motion conditions each individual dot had to have increasing radial speed from the center to the periphery. The following three speed/acceleration conditions were used: dots accelerating from 1 to 20°/sec (average speed 18°/sec), from 5 to 50°/sec (average speed 30°/sec) and from 10 to 70°/sec (average speed 50°/sec). Also, the dot size was directly proportional to the Euclidean distance from the center of the screen, and subtended 0.1 to 4°. In the circular and translational motion conditions, each individual dot was of constant size and speed, but the sizes and speeds differed accordingly to the dot distance from the center of the display.

In all conditions, each dot traveled along an appropriate trajectory for a ‘limited lifetime’ of 350 ms, after which it disappeared to be ‘reborn’ at a new random position. The appearance of new dots was controlled to maintain a constant dot density (0.04 dots/degrees² on average). The dot number, size, lifetime, and speed were the same in all motion conditions.

Control conditions were also tested in the static condition, that consisted of a stationary scene equivalent to the last single frame of the last movie (or motion condition) which thus maintained the spatial structure of the optic flow stimulus. This control condition was used to isolate the motion component. There were also null trials, i.e., periods in which only a central cross was displayed on the black background. Null trials constituted the low-level baseline for the study.

Overall, we investigated five motion conditions (radial, circular, translational, spiral and random motion) together with two control conditions (static and null trials) for a total of seven experimental conditions. Trials were 3 s periods showing one of the seven conditions while subjects had to fixate a central red cross of 0.4×0.4°. The intertrial interval was set to zero so that the movies were displayed one after the other to avoid that the brisk switch on-off of the movies could bias the neural response in the regions of interest. The experiment was constituted by 15 trials for each of the seven conditions (5 trials for each of the three speeds) for a total of 105 trials. Except for null trials (presented as a sequence of three consecutive null trials every 18 trials), trials were arranged in a pseudo-randomized order which was different for each scan but fixed across subjects. The three speeds were randomly mixed between trials and collapsed together in the analysis.

Localizer scans.

In a second set of fMRI experiments, we mapped the three motion areas V6, MT and MST+ using two different kinds of localizer visual stimuli. Each functional localizer comprised eight alternations of two conditions presented in blocks of 32 s each (16s ON vs. 16s OFF).

Retinotopic mapping.

In a third experiment, we mapped the retinotopic organization of the cortical visual areas in ten out of 13 subjects, using phase-encoded stimuli, as described elsewhere [5], [23], [24], [25], [26]. Shortly, the stimuli consisted of high-contrast flickering colored checks in either a ray- or a ring-shaped configuration (polar angle and eccentricity, respectively). These stimuli spared a central 0.75° circular zone of the visual field to avoid ambiguities caused by fixation instability. Periodic stimuli moved slowly and continuously, and checks reversed between bright and dark at a rate of 8 Hz (64 s/cycle, 8 cycles/scan). The average luminance of the stimuli was 105 cd/m2.

Imaging parameters.

For the 3D-FF experiment and localizer scans, MR 30 axial slices were 4 mm thick (with a 0 mm gap, interleaved excitation order), with an in-plane resolution of 3×3 mm, oriented approximately to the AC-PC line. From the superior convexity, sampling included almost all the cerebral cortex, excluding only the ventral portion of the cerebellum.

Each participant underwent six consecutive scans for the 3D-FF experiment and four scans for the localizer. Each scan took either 324 s (3D-FF) or 256 s (localizer scans) with 162 or 128 single-shot EPI images per slice, respectively. Retinotopic mapping was acquired in a separate day using the same apparatus, setup, and coil as for the main experiment. Images were acquired as in the main experiment, but in this case MR slices (4 mm thick, with an in-plane resolution of 3×3 mm) were oriented approximately parallel to the calcarine fissure and covering only the posterior part of the brain. Part of the retinotopic data were acquired with slices having a smaller thickness of 2.5 mm and oriented approximately parallel to the calcarine fissure (thus covering all the brain). This voxel size strikes a compromise between sufficient signal-to-noise and the ability to assign activations to the proper sides of the sulci and gyri. To increase signal to noise, data were averaged over three runs for each stimulus type (eccentricity and polar angle). Thus, in total, each participant underwent six scans for the retinotopic mapping. Each scan took 512 s, with 256 single-shot EPI images per 32 contiguous slices. Other standard imaging parameters were in common between experiments (TR = 2 s, TE  = 30 ms, TA  = 66.6 ms, flip angle  = 70°, 64×64 matrix, bandwidth = 2298 Hz/pixel). Overall, a total of 190 scans were carried out on the 13 subjects (78 scans for the FF3d experiment, 52 scans for the functional localizer and 60 scans to map retinotopic visual areas).

In each scan, the first 8 seconds of the acquisition were discarded from data analysis in order to achieve a steady state, and the experimental tasks started at the beginning of the fifth volume. The cortical surface of each subject was reconstructed from two to three structural scans (T1-weighted Magnetization Prepared Rapid Gradient Echo, MPRAGE, sequence, TR  = 2.00 s, TE  = 4.38 ms, flip angle  = 8°, 1×1 mm in-plane resolution, matrix 256×256, 176 contiguous 1 mm thick sagittal slices, bandwidth  = 130 Hz/pixel). The last scan of each functional session was an alignment scan (also MPRAGE, 1x1x1 mm) acquired in the plane of functional scans. The alignment scan was used to establish an initial registration of the functional data with the surface. Additional affine transformations that included a small amount of shear were then applied to the functional scans for each subject using blink comparison with the structural images to achieve an exact overlay of the functional data onto each cortical surface.

Anatomical image processing.

FreeSurfer was used for surface reconstruction [30], [31]. High resolution structural images obtained from each subject were manually registered and averaged. After reconstructing each hemisphere, we completely flattened the inflated occipital lobe after first cutting it off posterior to the Sylvian fissure, and making an additional cut along the Calcarine fissure. Stereotaxic coordinates were calculated through an automatic nonlinear stereotaxic normalization procedure [32], performed using the SPM8 software platform (Wellcome Department of Cognitive Neurology, London, UK), implemented in MATLAB (The MathWorks Inc., Natick, MA, USA).

Functional image processing: main experiment and localizer scans.

Images from the main experiment and functional localizers were preprocessed and analyzed using SPM8 (Wellcome Department of Cognitive Neurology, London, UK). Functional time series from each subject were first temporally corrected for slice timing, using the middle slice acquired in time as a reference, and then spatially corrected for head movement, using a least-squares approach and six-parameter rigid body spatial transformations. They were then spatially normalized using an automatic nonlinear stereotaxic normalization procedure (final voxel size: 3×3×3 mm) and spatially smoothed with a three-dimensional Gaussian filter (6 mm full-width-half-maximum). The template image for spatial normalization was based on average data provided by the Montreal Neurological Institute [33] and conforms to a standard coordinate referencing system [34]. The time series of functional MR images was first analyzed separately for each participant. The effects of the experimental paradigm were estimated on a voxel-by-voxel basis, according to the general linear model. The onset of each trial constituted a neural event, which was modeled through a canonical hemodynamic response function, chosen to represent the relationship between neuronal activation and blood flow changes. Separate regressors were included for each trial type (radial, translational, circular, spiral, random, static and fixation), yielding parameter estimates for the average hemodynamic response evoked by each trial type. The model also included a temporal high-pass filter, to remove low-frequency confounds with a period above 128 s. Serial correlations in the fMRI time series were estimated with a restricted maximum likelihood (ReML) algorithm using an autoregressive AR(1) model during parameter estimation, assuming the same correlation structure for each voxel, within each run. The ReML estimates were then used to whiten the data. These subject-specific models were used to compute a set of contrast images per subject, each representing the estimated amplitude of the hemodynamic response in one trial type relative to the fixation baseline. Contrast images from all subjects were entered into a within-subjects ANOVA with non-sphericity correction, where subject was considered as a random effect, thus allowing to draw inferences related to the whole population our participants were extracted from.

We used the model described above to search the whole brain for regions differentiating any of the five motion types (translational, circular, radial, spiral and random) from the static condition (static frames). The resulting statistical parametric map of the F statistics was thresholded at the voxel level and by cluster size. Correction for multiple comparisons was performed using false discovery rate (p<0.05; extent threshold  = 10 voxels). The resulting regions include all voxels showing a reliable BOLD response during motion vs. static frames, irrespective of the kind and amount of motion and of the sign (positive or negative) of the evoked BOLD response. In-house software (BrainShow, written in Matlab) was used to visualize the resulting regions onto a population-average, landmark- and surface-based (PALS) atlas [35], and to assign anatomical labels to activated areas at the level of Brodmann areas and cortical gyri. Brodmann areas were derived from the Talairach Daemon public database [36], while cortical gyri were derived from a macroscopical anatomical parcellation of the MNI single-subject brain [37]. BrainShow has been used in previous studies from our and other groups (e.g. [5], [38], [39], [40], [41]) and is freely available on request for academic usage (E-mail: gaspare.galati@uniroma1.it ).

After identifying the regions differentiating motion from static frames, we searched for modulation of BOLD responses in these regions as a function of the motion type. This step was performed on regionally averaged data as follows. Regions were defined as clusters of significantly activated adjacent voxels at most 8 mm away from each local maximum of the group statistical map. For each subject and region, we computed an estimate of the averaged amplitude of the hemodynamic response in each experimental condition, by entering a spatial average (across all voxels in the region) of the pre-processed time series into the individual general linear models. Such regional hemodynamic response estimates were then analyzed through a repeated-measures analysis of variance (p<0.05). The post-hoc Duncan’s test was performed when appropriate (p<0.05). Note that, although the analysis used to define the regions and the selective analysis conducted on the regionally averaged data are based on the same dataset, they are inherently independent. The first step tests for the presence of any motion-related neural response regardless of the kind of moving stimulus, while the second step tests for modulations induced by the kind of moving stimulus, thus avoiding the risk of “double dipping” [42].

We also performed the same regional analysis on the visual regions V6, MT and MST+, as individually defined on the basis of the independent localizer scans as described in previous studies. For localizer scans, stimulation blocks were modelled as box-car functions spanning the whole block duration and convolved with a standard hemodynamic response function. We used individually adjusted uncorrected thresholds to define these regions (p<0.001 or higher). To define the V6 ROI we used the same procedure described in our previous paper [5]. To define the MT/MST+ ROI we followed the same procedure described several times in previous papers (e.g., [4], [11], [12], [13], [22]) that can be summarized as follows. As contralateral stimuli drive the entire MT+, and ipsilateral stimuli drive only MST+, the activation regions obtained for the left and right stimuli overlapped substantially. MST+ was defined as all contiguous voxels that were significantly active during ipsilateral motion stimulation. MT was defined as all contiguous active voxels that were active during contralateral but not ipsilateral stimulation, with one proviso. As previous research [11], [12], [13] has shown that the centre of MST+ is located anteriorly with respect to the centre of MT, any MT voxels situated further anterior than the median value of MST+ ROI on the horizontal (axial) plane were removed from the MT ROI as done also in other previous studies (e.g., [22]).

Motion Coherence Coefficient.

To quantify the differential sensitivity of each region to the motion coherence as evoked by our visual stimulation, we extracted from each ROI (by averaging across all voxels in a ROI) the effect sizes for coherent and incoherent (i.e. random) motion conditions. We averaged together the results across the four coherent conditions to get a single value to be assigned to the coherent motion. We then computed the ratio of the means MC (Motion Coherence) and MI (Motion Incoherence) for each ROI and averaged the results across hemispheres. A MC/MI coefficient of 1 means that MC and MI have the same effect in a given ROI, a value smaller than 1 means that MI has a greater effect than MC whereas a value bigger than 1 means that MC has a greater effect than MI (for a similar approach see [6]).

Analysis of retinotopic data.

Retinotopic data were analyzed using FreeSurfer [30], [31] based on standard procedures described in details in many previous studies (e.g., [5], [24], [25], [26], [43]). Briefly, P values were estimated on a voxel-by-voxel basis by constructing an F ratio between ‘‘signal’’ (response amplitude at stimulus frequency) and ‘‘noise’’ (amplitude at other frequencies excluding second and third harmonics) with degrees of freedom equal to the number of time points. The phase of the signal at the stimulus frequency was used to map retinotopic coordinates (polar angle or eccentricity). The boundaries of the retinotopic visual areas were defined in each participant on the basis of the field-signs calculated from the maps of polar angle and eccentricity [24].

Area V6.

Results showed a region selectively activated by motion stimuli in the dorsalmost part of the parieto-occipital sulcus POs (Fig. 3). The location of this region well corresponds to the location of human area V6 recognized on the basis of a wide-field retinotopic analysis (Fig. 5) [26]. The area found here is indeed located on the dorsal margin of the POs in correspondence of its posterior bank, and has MNI coordinates (x = ±13, y = –81, and z = +44) well compatible with those of the retinotopic V6 [26].

To check whether this region actually corresponds to the medial motion area V6, we independently defined area V6 according to the functional localizer (see methods) in all scanned subjects, i.e., 26/26 hemispheres (Fig. 4A). The map found with the localizer (Fig. 4A) has the following MNI coordinates: x = ±9, y = –82, and z = +36 and its position closely resemble that obtained with the contrast M-S (Fig. 3). As described in our previous paper [5], when using this functional localizer the map of area V6 nicely overlaps with that obtained with wide retinotopic stimulation. This can be clearly appreciated in Figure 5, where area V6 (green) as defined by functional localizer shows a good overlap with the area V6 (white outline) as retinotopically defined in this subject.

The results show that area V6 (as defined by the general contrast M-S) responds more to any type of motion than to static stimuli (p<0.05), and more to coherent than to incoherent (random) stimuli (p<0.05, see green columns plot in Fig. 3). Among the various types of coherent motion, area V6 responds preferentially to the translational motion (p<0.05). V6 is not differentially modulated by circular, radial and spiral motion. Plots in Figure 4B show that area V6 defined by the localizer has a functional profile very similar to that observed in Figure 3 (i.e., it has a great sensitivity to translational motion, p<0.05, while it responds quite strongly but indifferently to the other types of coherent motion). One-way repeated measures ANOVA (F_(5,70)  = 2.708 p<0.05), and post hoc analysis (p<0.05) gave the same results in statistical terms in the two sets of data. This is particularly important for the reliability of the functional profile of area V6, as we used two independent set of data to draw the V6 ROI. In the V6 ROI as defined by the functional localizer (i.e., coherent vs incoherent motion) the voxels taken into account were belonging to V6 by definition, and data could be biased due to the fact that only those voxels with a preference for the coherent motion were included in the ROI. On the contrary, V6 ROI as defined by the contrast M-S does not suffer from this confound and can be considered ‘unbiased’. The strong preference for the translational coherent motion observed in both the biased and unbiased V6 ROIs highly increases the reliability of the results and again suggest that the V6 as defined by the M-S contrast is the same area as that defined by the localizer.

Areas MT+ and its functional subdivisions MT and MST+.

Results from the contrast M-S revealed significant bilateral activations in the temporal cortex (Fig. 3). The mean MNI coordinates of this region (x = ±45, y = –71, z = +4; Table 1) are in good agreement with those of the classic motion sensitive region MT+ described in earlier studies using both PET (x = ±42, y = –69, and z = 0; [47]) and fMRI (x = ±45, y =  −76, and z = +3; [1]).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Mean MNI coordinates of ROIs identified in the present study.

https://doi.org/10.1371/journal.pone.0060241.t001

It is now generally acknowledged that the large motion-sensitive region MT+ is a complex of several areas (e.g., [5], [23]), including areas as MT and MST, which have different functional profiles and could be differently involved in egomotion perception. Previous authors have shown that unlike MT, MST shows a degree of preference for at least a single flow stimulus (e.g., [4]). To check the preference of these two areas to different types of flow stimuli, we mapped MT and MST+ ROIs using an independent functional localizer following standard procedures as described in the Methods and in many previous papers (e.g., [4], [11], [12], [13]). MT was successfully defined in 22/26 hemispheres. Although ipsilateral responses were relatively weak compared with contralateral responses, a subregion of ipsilateral activity was clearly identifiable in 20/26 hemispheres and was marked as MST+. The two regions resulting from this analysis are rendered in Figure 4C. The mean MNI coordinates of MT region are x = ±45, y = –80, and z = −2, those of MST+ region are x = ±47 y = −69 z = 8 (see Table 1). Note that the mean MNI coordinates of MT region closely correspond to those provided in Kolster et al. [23] for the neighboring area pV4t (RH: x = +47, y = –81, and z = −2). Thus, what is labeled MT here could be a mixture of MT and neighboring pV4t according to Kolster et al [23]).

Figure 5 shows that the MT and MST+ ROIs occupy an anatomical position in between the Inferior Temporal sulcus (ITs) and the Middle temporal Sulcus (MTs), which is in line with the description of these two regions provided in previous fMRI studies where the two motion areas have been distinguished [11], [12], [13]. Specifically, area MT (red) was anterior to and distinct from the retinotopically defined areas V1, V2, VP, V4v, V4/V8 whose locations, indicated in the flat map, were previously identified in this subject. Area MST+ (dark blue) was always anterior and often dorsal to MT, although there was some degree of variability across subjects. A constant evidence was that MST+ typically abutted MT, as previously reported by Huk et al. [13].

Purple plots in Figure 3 show that region MT+ (as defined by the general contrast M-S) responds more to any type of motion than to static stimuli (p<0.001) and a bit more to coherent than to incoherent (random) stimuli (p<0.05). Like V6, among the various type of coherent motion MT+ responds preferentially to the translational motion (p<0.05) and is not differentially modulated by circular, radial, and spiral motion. Contrary to V6, however, the response to random motion is high in MT+, being only slightly weaker than that observed for coherent motion (n.s.). This explains why MT+, in contrast to V6, results insensitive to optical flow stimulation when coherent minus incoherent stimulation was used as a paradigm [5].

Red plots in Figure 4D show that the localizer-defined area MT responds more to any kind of motion than to static stimuli (p<0.0001). Area MT distinguishes between random and at least three types of coherent motion, translational (p<0.001), radial (p<0.001), and spiral (p<0.05) with no preference among the three types of coherent motion. Also the fourth type of coherent motion we tested (circular motion) is more effective than random in activating MT, but the difference does not reach statistical significance.

Blu plots in Figure 4D show that the localizer-defined area MST+ responds more to any kind of motion than to static stimuli (p<0.0001), and more to coherent than to incoherent (random) stimuli (p<0.005). MST+ responds more to the translational (p<0.001)and radial motion (p<0.5), which do not differ each other, with the circular motion being the less favorite condition.

To test for functional differences between MT and MST+ ROIs, the two regions were submitted to a 2 by 6 repeated-measures ANOVA with factors Region (MT and MST+) and stimulus type (radial, translational, circular, spiral, random, static), where the five motion conditions and the control static condition were considered as six levels of a single variable. This analysis yielded a significant Region by stimulus type interaction (F_5,95 = 5.81; p<0.0001), indicating a different sensitivity of MT and MST+ for the different type of stimuli. Post-hoc comparisons revealed that MT responds more strongly to random motion than MST+ (p<0.05), while the two areas did not differ each other respect to the other components.

Area V3A.

The contrast M-S showed significant activity in the posterior segment of the IPs (Fig. 3). The mean coordinates of this region are x = ±20 y = −87 z = +29 (Table 1). This location corresponds to the position of the retinotopic dorsal visual area V3A, especially in the X and Y directions (e.g., [25]). It does not extend either medially or antero-laterally toward the typical positions of retinotopic dorsal areas V3 and V7, respectively. The activation is located posteriorly and laterally to that of the V6 ROI, bordering its posterior part. Retinotopic mapping confirms that this region is indeed V3A (Fig. 5). In fact, in Figure 5 the motion area V3A (violet) is located on the posterior segment of the IPs (pIPs), a location remarkably coincident with the position of retinotopic area V3A, whose anatomical landmark is indeed the pIPs [25], [48].

Violet plots in Figure 3 show that area V3A responds more to any kind of motion than to static stimuli (p<0.001), but it is not particularly sensitive to the coherence of motion. It has a great response to random motion (which is not statistically different with respect to the coherent motion) and responds equally well to different coherent movements. In other words, V3A responds to motion independently to the presence/absence of coherence in it.

Area CSv.

The M-S contrast showed significant bilateral activity in the depth of the medial posterior cingulate sulcus (Fig. 3). Figure 5 shows more specifically that the CSv area (light blu) is located in between the posterior ascending portion of the cingulate sulcus (also called the marginal ramus of the cingulate sulcus) and the corpus callosum. This location, as well as the mean coordinates of this region (x = ±15 y = −33 z = +39; Table 1), corresponds well to the original definition of human CSv provided by Wall and Smith [4].

Cyan plots in Figure 3 show that area CSv is weakly, if any, activated by coherent motion (p<0.05), while its activity is consistently suppressed below the baseline by random motion (p<0.005) and static stimulations (p<0.05). This means that CSv does not discriminate between the various types of coherent motion, which indeed evoked BOLD signals that do not statistically differ each other.

Area IPSmot.

The contrast M-S showed significant activity along the horizontal segment of the IPs (Fig. 3). The mean MNI coordinates of this region (x = ±30 y = −60 z = 45; Table 1) are in line with those of area VIP as described in previous fMRI studies [4], [6].

Orange plots in Figure 3 show that area IPSmot responds more to translational motion (p<0.005) than to every other kind of visual stimulations we used. The area shows also a weaker but significant response to spiral motion with respect to the random condition (p<0.05), followed by circular, radial random and static conditions, which do not statistically differ each other.