Seven subjects (5 females, 2 males, mean age = 24.42 years, SD = 4.82 years) participated in the study. They were graduate students at Boston University, recruited from our pool of subjects. All of them had normal or corrected-to-normal vision. All underwent a psychophysical testing session prior to the scan, to make sure that their performance was at least 70% correct for Δ τ = 0.5 sec for the symmetric configuration regardless of the initial x-offset and background motion coherence (0% or 75%) [16] . All participants signed an informed consent form before the start of the experimental sessions in accordance with the requirements on research involving human subjects, as approved by the Massachusetts General and Boston University Institutional Research Boards. All subjects fully satisfied the inclusion criteria for participating in MRI/fMRI studies and none of the exclusion criteria were met. They participated previously in other psychophysical and functional imaging tasks conducted by our research team. Those studies had no similarity with the task reported here. All subjects reassured us they could pay attention throughout an experimental task, maintain fixation, and stay still during the imaging experiments.

The stimuli were generated on an Intel-based Macintosh laptop and displayed at a resolution of 1024 × 768 pixels and a refresh rate of 75 Hz. Two of the dots, referred to as target dots, were red (51.20 cd/m²) and the rest of them were white (79.55 cd/m²), all displayed against a gray background (10.22 cd/m²). They were back-projected onto a translucent screen (27.3 cm × 36.5 cm) using a LCD projector. Subjects viewed the translucent screen through a mirror mounted on the head coil of a whole-body scanner. The distance between the eyes of the subject and the mirror was approximately 4 cm and the distance between the mirror and the screen was approximately 81 cm, therefore, the total viewing distance was about 85 cm. This setup provided a square viewing aperture subtending 17˚ × 17˚. fMRI data were acquired at Athinoula A. Martinos Center for Biomedical Imaging, Massachusetts General Hospital, using a 3T Siemens whole body scanner and a standard 8-channel head coil. Structural images were obtained as T1 weighted magnetization prepared rapid acquisition gradient echo images (MPRAGE) (128 slices with slice thickness of 1.33 mm, voxel size: 1.00 × 1.00 × 1.33, FOV = 256, TR = 2.53 sec, TE = 3.39 msec, flip angle = 90˚). Two T1-weighted images were collected for each subject. Functional images were obtained with gradient echo, echo planar (EPI) interleaved sequence (33 slices with slices oriented along the AC-PC line, slice thickness of 3 mm with 20% distance factor, FOV = 200, TR = 2.00 sec, TE = 30 msec, flip angle = 90˚) for measurement of BOLD contrast.

A field of moving white dots simulated the observer’s forward self-motion in 3D and was presented through a square viewing aperture. The dots remained stationary with respect to one-another in the simulated space. Subjects were asked to indicate which one of two red dots would pass their eye plane first, mimicking they were moving forward through the field. All of the dots subtended 2 pixels × 2 pixels (4 arcmin × 4 arcmin) throughout the simulated approach and were placed such that they maintained a density of 2 dots/deg². The screen size of all dots, including the targets, remained constant, thus eliminating all local tau cues. The motion of the dots simulated the subjects’ forward self-motion along a straight-line trajectory at a speed of 150 cm/s. In each trial, the direction of simulated self-motion was toward the center of the aperture. Dots that moved out of the volume behind the observer’s eye plane were randomly assigned to new locations such that the density of the dots remained constant (Figure 1) [1] .

The target dots were placed at different depths such that the difference between their passage times (tau difference Δtau) was set to be 0.5 sec for all trials. This value was chosen because it should be just above detection threshold, rendering the task of identifying the leading target meaningful but still allowing for errors. The initial depth of the reference target was 1200 cm. Thus, the possible TTP values from stimulus onset until both targets would have passed the observer, ranged from 7.5 s to 8.5 s. This left ample time for the 2AFC responses, which had to be made before the leading target passed the vertical eye-plane of the observer. Observers were not required to respond as quickly as possible.

The two target dots were placed such that they were always on opposite sides of the track vector. Their lateral distances from the track vector (x-offsets) were either 10 cm or 50 cm in the simulated space. This resulted in four different lateral target offset combinations: Two equidistant symmetric placements (leading target 10 cm to one side-trailing target 10 cm to the other side, or 50 cm - 50 cm). In the other two combinations, the targets were placed with asymmetric x-offsets (10 cm - 50 cm, and 50 cm - 10 cm). These 4 stimuli were paired with two coherence levels of the background dots. Remember that an entirely incoherent flow-field no longer specifies observer motion. Our previous psychophysical data [1] showed that 50% coherence of the dots begins to provide prominent global flow information relative to the condition of 0% coherence dots. This design led to eight unique stimuli, each of which was repeated 16 times within a run, using a randomized event-related design paradigm.

The visual stimulus was occluded after 3 seconds. The next trial would not be presented until a decision had been made. The timing and order were randomized using optseq2 (http://www.freesurfer.net/optseq/). Inter-stimulus Intervals (ISIs) between trials varied from 1 - 7 s. Frames with static dots were presented within the ISIs, serving as a baseline condition. During the whole scanning period, subjects were required to fixate a small central cross (40 × 40 arcmin). Stimuli were presented binocularly in a two-alternative forced choice (2AFC) paradigm without feedback. The subjects’ task was to determine which of the two targets would arrive at their eye plane first. Subjects entered their responses by pressing a designated key on a magnet-compatible button box.

A separate block design employing a MT localizer task was performed by all subjects in two runs. The human middle temporal complex (hMT) has been shown to be highly involved in motion processing, including optic flow. Accordingly, area hMT was functionally localized by utilizing moving and static dot patterns [17], so that we could locate the exact position of hMT for further analysis [18] . Other anatomical regions were defined with normalized functional images, using the Automated Anatomical Labeling (AAL) atlas.

Imaging data analysis was performed using the Statistical Parametric Mapping software package (SPM12, Wellcome Department of Cognitive Neurology, London, UK) and utilizing MATLAB (The MathWorks, Natick, Massachusetts, USA). The preprocessing steps were as follows: 1) format conversion by converting data from original DICOM files to Nifti files, 2) slice timing eliminated the time shift for all voxels of functional images, 3) realignment by motion correction, 4) co-registration of the anatomical image with the mean functional image, 5) spatial normalization normalized all images to a standard space (Montreal Neurological Institute, MNI) with a voxel size of 3 × 3 × 3 mm, and 6) smoothing: all functional data were smoothed with a FWHM (Full Width at Half Maximum) kernel of 4 mm.

For each subject, the onset and duration of each condition was modeled by a general linear model (GLM). The motion parameters from realignment were also used as multiple regressors in generating the design matrix. Trials with correct responses and incorrect responses were separated as different conditions. With the contrast images of each subject, a group level randomeffect analysis was performed for each condition. The resulting t-value maps were set as uncorrected for multiple comparisons, p < 0.05. Clusters with less than 10 contiguous voxels were excluded.

Based on group-level activation maps in normalized space, we defined several functional regions of interest (ROI) for each single subject. Every functional ROI was defined as a sphere, with its center at the respective local maximum of the activation cluster and with a 5-mm radius. Subsequently, we calculated the percent BOLD signal change for each functional ROI using Marsbar [19] .

Bilateral hMT areas were defined using localizer tasks [18] . We set the minimum overlapped proportion on individual activation maps as 0.5, based on the group-constrained subject-specific (GSS) method [20] .

Response accuracy for each condition was first calculated per subject and then averaged across subjects (Figure 2). As expected, subjects could do the task and performed well above chance when the two target dots were symmetric around the track vector of the simulated motion. This was the case for both 0% coherence and 75% coherence conditions. Thus, subjects exploited the simple image-base cue of eccentricity to guide their answer choices. In contrast, when the two target dots were placed asymmetrically, such that the correlation of target eccentricity and proximity to the observer was severely reduced, the response accuracy was around chance level. Thus, the behavioral performance showed that in the absence of local expansion cues, TTP judgments were based on retinal eccentricities.

Interestingly, when the flow field contained additional information about the track vector, performance improved. When the target dots were symmetric, subjects performed better under the 75% coherence condition compared to 0% coherence. Thus, global motion information provided by background dots

enhanced subjects’ performance only if the targets were spaced symmetrically around the track vector. However, when the target dots were asymmetric, there were no significant differences between 0% coherence and 75% coherence (p > 0.05 in paired t-test). These results replicate previous behavioral results we collected with a similar experimental design [1] .

We contrasted activation during the trials against the activation within the static dots presentation (baseline) to obtain significance activation maps. The analysis was performed separately for 0% and 75% coherence levels in the optic flow field Table 1 & Table 2. The runs with the same coherence value were grouped together and the contrasts were done separately for each experimental condition. Based on behavioral performance, we separated correct responses and incorrect responses for each condition. Therefore, the experimental conditions were the cases where initial target x-offsets were symmetric and 10 cm from the center of the aperture (10 vs 10), symmetric and 50 cm from the center of the aperture (50 vs 50), and the leading target’s initial x-offset was 10 cm from the center of the aperture (10 vs 50) or the leading target’s initial x-offset was 50 cm from the center of the aperture (50 vs 10), either with correct responses or incorrect responses (Figure 3).

AG: angular gyrus, IFG: inferior frontal gyrus, IPG: inferior parietal gyrus, ITG: inferior temporal gyrus, MFG: middle frontal gyrus, MOG: middle occipital gyrus, MTG: middle temporal gyrus, ParaCL: paracentral lobule, PostCS: postcentral cortex, PreCS: precentral cortex, SFG: superior frontal gyrus, SMG: supramarginal gyrus, SOG: superior occipital gyrus, SPG: superior parietal gyrus.

Figure 3 ((a) condition 10 cm vs 10 cm; (b) 10 cm vs 50 cm; (c) 50 cm vs 10 cm; (d) 50 cm vs 50 cm) shows the activation maps of trials with correct responses with a threshold t-value 1.943, extent of 10 voxels (p < 0.05, uncorrected), across all the subjects. Table 1 and Table 2 show the corresponding coordinates of the local maxima of these clusters, in MNI space.

In general, across all the subjects, the activation areas were distributed along the motion processing pathway [21] . Consistent with previous research [22], the activation was not only showing along the ventral pathway (also known as “what” pathway), it also showed along the dorsal pathway (also known as “where” pathway). Activation during the stimulus motion was significantly elevated compared to baseline in the occipital lobe, including the superior occipital gyrus (SOG), temporal lobe including middle temporal gyrus (MTG), inferior temporal gyrus (ITG) and superior temporal sulcus (STS), parietal lobe including superior parietal gyrus (SPG) and intraparietal sulcus (IPS), frontal lobe including middle frontal gyrus (MFG) and inferior frontal gyrus (IFG). In addition, the pre-central sulcus and post-central sulcus were activated bilaterally across all subjects.

For all x-offset target conditions, when we compared the activation for 0% coherence and 75% coherence conditions, the latter resulted in more distributed activation aroundthe bilateral pre-central and post-central temporal areas, MTG, IFG and IPS.

Based on previous functional imaging data related to TTP and TTC processing (e.g. Field & Wann, 2005), we defined several functional regions of interest based on the Automated Anatomical Labeling (AAL) atlas. Figure 4 shows the BOLD percentsignal changes in these functional ROIs. By and large the BOLD percentsignal changes increased from 0% to 75% coherence levels in bilateral precentral, postcentral, and middle temporal areas, whereas they decreased in superior and middle frontal areas. In the inferior frontal and parietal cortical regions, including intra-parietal sulcus, there was not much difference between the two coherence levels. Specifically, in the left precentralsulcus (PreCS: p = 0.045), left postcentralsulcus (PostCS: p = 0.027) and bilaterally in the middle temporal (MTG: p = 0.025) cortical regions, percentsignal changes at 75% coherence were significantly larger than those at the 0% coherence level, for the symmetric condition (10 vs 10). In the right hemisphere MTG: p = 0.024), significant activations were also found in the asymmetric condition (10 vs 50) at both coherence levels. Bilaterally in the middle frontalgyrus (MFG: left: p = 0.012, right: p = 0.005) and right PreCS (p = 0.046) areas, there were significant differences between the 0% and the 75% coherence level in the asymmetric condition (50 vs 10). In the left inferior frontal gyrus (IFG: 0.034), percentsignal changes were also found significantly higher under 75% coherence than 0% coherence, in the symmetric condition (50 vs 50). In MFG in the left hemisphere (p = 0.030), percent signal changes were found to be significantly higher at 0% coherence than at 75% coherence, in the symmetric condition (10 vs 10). Interestingly, bilateral significant differences in percent signal change were found in hMT areas for asymmetric conditions, but not for symmetric conditions (left hMT: p = 0.013 in 10 vs 50, right hMT: p = 0.032 in 10 vs 50 and p = 0.036 in 50 vs 10).

The percent signal changes suggest that bilateral precentral and postcentral sulci as well as a MTG are highly involved in the processing of global optic flow. The activation in hMT bilaterally suggests that more complicated visual processing is performed when there is more than one cue that subjects might use (e.g. global

optic flow, object velocities, and symmetry heuristics) for their TTP judgments. The activation of IFG and MFG may be underlying the process of decision making for solving the task.

For each condition described above, we also compared the trials in which subjects gave the correct responses to those with incorrect responses. For the symmetric conditions (offsets of 10 cm vs 10 cm and 50 cm vs 50 cm), the activations were more extended for trials with correct responses than for those with incorrect responses. The activations in precentral postcentral, IPS, MTG and MFG regions, in both correct and incorrect trials, suggest the involvement of these areas in solving the underlying task. For asymmetric conditions with the leading target at 10 cm x-offset (10 vs 50), invalid image velocity information was provided to the subjects, resulting in more distributed activation in incorrect trials than in correct trials, at 0% coherence. When coherent background dots were presented (75% coherence), thereby providing global motion information, subjects exploited multiple cues, resulting in more distributed activation in correct trials than in incorrect trials. Conversely, for the asymmetric conditions with the leading target at 50 cm x-offset (50 vs 10), valid image velocity information was provided, and activations were more distributed in correct trials than incorrect ones in both 0% and 75% coherence level.