Thirty healthy adults volunteered for an fMRI study. A self report questionnaire was used to screen participants for known psychiatric and neurological conditions and medications that could affect their performance. For all participants, English was the first language and vision was normal or corrected to normal. Informed consent was obtained prior to the experiment and all participants received compensation. Data from six participants were excluded due to technical problems with the images.
Of the remaining 24 participants (13 females and 11 males), the mean age was 24.0 years (standard deviation [SD] = 3.6, range = 19.4–33.9) and the mean level of education was 17.3 years (SD = 2.0, range = 13–22.5). The Edinburgh Handedness Inventory  indicated 23 participants were right handed and one was mixed handed (laterality quotient mean = 85.0, SD = 16.4, range = 28.6–100). All participants completed an exit questionnaire and were debriefed after the experiment. The study had ethics board approval.
Experimental Design and Task
The experiment was designed to manipulate IT. Stimuli were presented to either the right visual field (RVF) or left visual field (LVF). Word and face stimuli were used to elicit lateralized processing in the left and right hemispheres, respectively. Participants were asked to respond with the right or left hand. As a result, four IT conditions were examined:
low IT (e.g., LVF faces and left hand response);
visual IT (e.g., RVF faces and left hand response);
motor IT (e.g., LVF faces and right hand response); and
high IT (e.g., RVF faces and right hand response).
Participants were instructed to maintain fixation for the duration of the experiment. Pilot eye tracking data confirmed that participants were able to maintain fixation. A central fixation was presented continuously. Stimuli were presented to the left or right of the fixation (150 ms duration), with the inside edge at least 2.3 degrees away from centre to avoid stimulating the vertical meridian . For each trial, participants were asked to judge stimulus format, in which they discriminated words versus pseudowords, and faces versus scrambled faces. To keep the response requirements simple, stimulus format was indicated using the index and middle fingers (counterbalanced across subjects). Responses to stimuli of the same type (i.e., words/pseudowords versus faces/scrambled faces) were made with the same hand, depending on the IT condition (see motor IT details below). Participants were instructed to respond as quickly and accurately as possible.
The experiment was presented in a block design format, with blocks of low IT, visual IT, motor IT, and high IT conditions (eight blocks per condition). Eight stimuli (two each of words, pseudowords, faces, and scrambled faces) were presented per block. The inter-stimulus interval (ISI) was varied pseudorandomly within each block (1.85, 2.85, or 3.85 s, mean ISI = 2.6 s). Each block lasted 22 s, with an 18 s rest period between blocks.
Blocks were grouped together into two runs, with sixteen blocks per run. Due to response hand requirements, one run contained the low and visual IT blocks, and the other run contained the motor and high IT blocks. Each participant completed both runs (counterbalanced order). Response hand assignment was only changed between runs (low and visual IT run: right hand for word stimuli, left hand for face stimuli; vice versa for the motor and high IT run). Participants performed practice tasks (with feedback) at the beginning of each run to ensure adequate performance (at least 50% accuracy). Each run was divided into two sub-runs to allow participants a chance to rest (330 s per sub-run). Each sub-run started with an 18 s fixation.
Stimuli were presented using E-Prime software (Psychology Software Tools, Inc.). Stimuli were back projected onto a screen that was affixed to the front of the bore of the scanner. Subjects viewed the screen through a mirror mounted to the head coil. Behavioural data were collected using an in-house magnetic resonance-compatible button pad and recorded using E-Prime.
Sixty-four monosyllabic, four letter words were selected from the Medical Research Council (MRC) Psycholinguistic Database [48, 49]. A number of linguistic factors were controlled across conditions: imageability, concreteness, Kucera-Francis written frequency, and familiarity ratings from the MRC database; and pronunciation regularity from the Centre for Lexical Information (CELEX) English lexical database . Pseudowords were derived from a second set of 64 words (matched with the real words on the linguistic factors) by replacing one or two letters. The pseudowords were manually verified to ensure they were not proper nouns or homonyms of real words. The word stimuli were presented in black uppercase letters on a white background (24-point Arial font). Words were presented vertically (e.g., [51–54]) with visual angles of approximately 3.0 degrees high by 0.9 degrees wide.
Sixty-four front view, expressionless, and unfamiliar faces were selected from the Max Planck Institute for Biological Cybernetics Face Database . The digitized images were converted to grayscale and the image contrast was increased so that the features could be rearranged without creating the appearance of seams in the images. Scrambled faces were created by rearranging the order of the eyes, nose, and mouth within the face outline for each of the 64 faces. The eyes were never the bottom most feature to ensure that they fit within the face outline. Face stimuli occupied a visual angle of approximately 3.4 degrees high by 2.8 degrees wide.
Behavioural Data Analysis and Task Validation
Reaction time data were lost due to equipment failure. Accuracy data were submitted to a repeated measures analysis of variance (ANOVA) with four factors (visual field, response hand, stimulus type, and stimulus format). The significance threshold was set at p < 0.05. Post-hoc paired t-tests were performed where appropriate (Bonferroni correction).
Accuracy data are presented as mean +/- standard error. Overall, participants were able to perform the task accurately (82.0 +/- 1.6%). There was a main effect of visual field (F(1,23) = 7.50, p < 0.05), with greater accuracy in the RVF (83.2 +/- 1.8%) than the LVF (80.8 +/- 1.6%). This effect can be explained by the visual field by type interaction (F(1,23) = 6.78, p < 0.05). Post-hoc tests revealed greater accuracy for word stimuli in the RVF than the LVF (t(23) = 3.30, p < 0.05, Bonferroni corrected), with no significant effects of visual field for face stimuli. Finally, there was a main effect of format (F(1,23) = 17.49, p < 0.001), with significantly more accurate responses for words and faces (86.9 +/- 1.6%) than pseudowords and scrambled faces (77.1 +/- 2.3%). In summary, there was an effect of visual IT for word stimuli, highlighting the interaction between visual IT and stimulus type.
Imaging was performed on a 4 T magnet (Oxford Magnet Technology) using an INOVA™ console (Varian, Inc.), 36 mT/m imaging gradients (Tesla Engineering), and a transverse electromagnetic quadrature radiofrequency coil (Bioengineering, Inc.). A two-shot spiral sequence was used to acquire whole brain fMRI data (repetition time [TR] = 1000 ms, effective TR = 2000 ms, echo time [TE] = 30 ms, flip angle = 60 degrees). Twenty axial slices (6 mm thick, 0.6 mm gap) were acquired with a 240 × 240 mm field of view (FOV) and a 64 × 64 matrix. For each sub-run, 165 volumes were acquired, for a total of 660 volumes per participant. At the beginning of each sub-run, four dummy scans preceded the acquisition. A navigator echo and a field map correction were applied during image reconstruction. A whole brain, T1-weighted anatomical image was also acquired using a magnetization prepared fast low angle shot (MPFLASH) sequence (TR = 10 ms, TE = 5 ms, inversion time = 500 ms, flip angle = 11 degrees, 240 × 240 mm FOV, 256 × 128 matrix, 203 axial slices, 0.94 mm thick).
Imaging Data Analysis
Statistical Parametric Mapping '05 (SPM5) (Wellcome Department of Cognitive Neurology, Institute of Neurology) [56, 57] running on Matlab 6.5.1 (R13) was used for pre-processing and statistical analysis. Images were realigned to correct for head motion using a six parameter, rigid body transformation. A two pass procedure was employed: images were initially realigned to the first image and then registered to the mean image. The participant's anatomical image was then coregistered to their functional images, using the mean functional image as the reference. To transform the data to Montreal Neurological Institute (MNI) space, the mean functional image was warped to a functional image template using an optimized 12 parameter affine transformation followed by a series of nonlinear deformations. This transformation was then applied to all volumes (voxels resampled to 3 × 3 × 3 mm) and the coregistered anatomical image. The functional data were spatially smoothed using a 7.5 mm isotropic field-width-at-half-maximum Gaussian kernel.
A global scaling factor was used to account for signal drift between scanning sessions within each subject. A highpass filter (period: 128 s) was applied. An autoregressive (order 1) plus white noise model was used to account for short range serial temporal correlations [45, 58]. Regressors for modelling the observed responses were created by convolving stimulus onset vectors for each condition with SPM5's hemodynamic response function (HRF) and its time and dispersion derivatives . Six movement parameters (output from the realignment procedure) were included as regressors of no interest. For each subject, t-contrasts of the HRF regressors were created to compare activation associated with each condition against baseline. T-contrasts were also created to compare conditions (e.g., high IT > low IT). Unless otherwise stated, the activation threshold was p < 0.05 (FWE corrected, extent = 2). To assign anatomical labels to activations, the MNI coordinates were converted to Talairach space using a nonlinear transformation  and anatomically identified using the Talairach Daemon Client 2.0 [61, 62] and the MNI Space Utility .
At the group level, a random-effects analysis was employed. For each t-contrast, an ANOVA was performed with three factors: the HRF regressor and its time and dispersion derivatives. Group effects were then assessed with t-contrasts for the HRF regressor (time and dispersion derivatives were factors of no interest).
For the individual level analysis, whole brain task related activation was examined. To identify a cluster as callosal activation, the majority of the cluster's voxels had to be located in the corpus callosum (defined by a region of interest [ROI] created in WFU PickAtlas ). This reduced the potential contribution of partial volume effects (i.e., clusters that were located primarily in gray matter were excluded).
In addition, voxels with significant movement related activity were also excluded. Movement related signal changes were identified by performing an F-test on the motion regressors (p < 0.05, FWE corrected, extent = 2). Two of the five subjects with callosal activation had overlapping movement related activity (S03: 14 out of 66 voxels; S04: 1 out of 23 voxels). These voxels were excluded from the description of the callosal activation (Table 1) and the time course analysis (Figures 3 and 4; see below). In addition, a sixth subject had task-related callosal activation, but was excluded because all of the activated callosal voxels had significant movement-related activity.
Time Course Analysis
Time course data were extracted from ROIs based on both functional and structural boundaries using MarsBaR . The white matter ROIs were derived from the corpus callosum clusters (described above). The gray matter ROIs were derived from activation clusters that were the best match to the corpus callosum activation clusters in terms of intensity and extent (Table 1). White matter voxels were excluded from these clusters by performing an intersection with the subject's gray matter segmentation map. For one subject (S03), the most comparable gray matter cluster was much larger than the callosal cluster. In this case, a subset of gray matter voxels was selected by performing another intersection with a partially overlapping cortical region (defined using the WFU PickAtlas such that the resulting ROI was approximately matched for extent with the white matter ROI). Signals were extracted from all four sub-runs and demeaned to account for global signal differences. The sub-runs were averaged together for each subject. The frequency spectra of the time courses were created using fast Fourier transforms (FFT).