Event-related design example

From FMRI Methods

This provides an example of a methods section that conforms to the guidelines laid out by Poldrack et al. (2008). It is loosely based on an unpublished study by Poldrack, Stover, and Cazalis, but some details have been fabricated to demonstrate as many of the guidelines as possible.

Participants

Fourteen healthy subjects were included in this study (mean age 22.4 years, range 19-35; 10 females). Four additional subjects were recruited but excluded from analysis (for excessive head motion in the scanner, non-completion of study, inadequate task performance (< 70% accuracy) and structural abnormalities, respectively). All volunteers gave informed consent according to procedures approved by the UCLA Office for Protection of Research Subjects. All volunteers were native English speakers, and right-handed as determined by the Edinburgh handedness inventory. Subjects had no history of neuropsychiatric disorders and were not currently taking psychoactive medications.

Task design

The study design is derived from previous studies of skill acquisition in the mirror-reading task (Poldrack et al., 1998; Poldrack & Gabrieli, 2001). Subjects participated in two scanning sessions separated by two weeks, during each of which they were presented with 6 imaging runs. During each run, the subject was presented with a word on each trial and asked to decide as quickly as possible whether the word named a living or nonliving entity, using a keypad button-box to register their response. Each run included 32 plain and 32 mirror-reversed words, varying in length from four to seven letters. There were a total of 12 word lists (6 runs in 2 sessions); order of presentation of the 12 word lists was counterbalanced across subjects/sessions, and word length was equated within each list. This ensures that no words are repeated from the first to second training session, such that any learning effects reflect general skill rather than item-specific learning. Stimuli were presented in a pseudorandomly mixed fashion within each run, and there was no warning on each trial whether the item would be in mirror-reversed or plain text. Switching between stimulus types (mirror-reversed and plain text) occurred on 34% of trials. The timing and order of stimulus presentatation was optimized for estimation efficiency using custom MATLAB code (Dale, 1999); the response window was 3.25 seconds, and the stimulus-onset asynchrony (SOA) varied across trials from 4 to 11.5 seconds according to an exponential distribution (mean SOA=6.28 seconds). These stimulus onset lists were also counterbalanced across runs over subjects.

Following the initial scan, subjects participated in three behavioral training sessions, during which they were presented with ten passages written entirely in mirror reversed text. The subjects are instructed to read the passages, which were each several paragraphs long, as quickly as possible; speed to read each passage is recorded. After each passage, subjects were given a multiple-choice question related to the content of the passge, to ensure that they actually read the passage. The three training sessions were spaced over a period of two weeks, with no more than one session on any single day. Following the final scan, subjects performed a recognition memory test for items encountered during scanning; results from that test are not reported here

Stimulus presentation and timing of all stimuli and response events were achieved using the MATLAB Psychophysics Psychtoolbox (www.psychtoolbox.org) on an Apple PowerBook running Mac OS 9 (Apple Computers, Cupertino, CA). Visual stimuli were presented using MRI-compatible goggles (Resonance Technologies, Van Nuys, CA), and the computer was synchronized with the onset of each functional run to ensure accuracy of event timing. Response times and accuracy were measured using a fiber optic button box (Current Designs, Inc).

fMRI acquisition

Imaging was performed using a 3T Siemens (Erlangen, Germany) Allegra MRI scanner at the UCLA Ahmanson-Lovelace Brain Mapping Center. In each of six imaging runs, we acquired 205 whole-brain functional T2*-weighted echoplanar images (EPI) [slice thickness, 4 mm; 1mm skip; 34 slices; repetition time (TR), 2 s; echo time (TE), 30 ms; flip angle, 90°; matrix, 64 x 64; field of view (FOV), 200 mm]. Two additional volumes were discarded at the beginning of each run to allow for T1 equilibrium effects. In addition, a T2-weighted matched-bandwidth high-resolution anatomical scan (same slice prescription as EPI) and magnetization-prepared rapid-acquisition gradient echo (MPRAGE) were acquired for each subject for registration purposes (TR, 2.3; TE, 2.0; FOV, 256; matrix, 192 x 192; sagittal plane; slice thickness, 1 mm; 160 slices).

Data analysis

Behavioral data were analyzed using repeated measures analysis of variance (ANOVA) in R (http://www.r-project.org). fMRI data were preprocessed using the FSL toolbox (http://www.fmrib.ox.ac.uk/fsl) Version 3.2 (Smith et al., 2004). BET (brain extraction tool) was used to extract the brain from the image of the skull and surrounding tissue. Motion correction was performed using MCFLIRT, using a normalized correlation ratio cost function and linear interpolation. The direction and magnitude of motion for each subject over the course of each scan were examined, and functional runs in which a subject moved over 1mm in any direction were excluded from the analysis (a total of 5 runs were excluded). Data were spatially smoothed using a 5 mm full-width-half-maximum Gaussian kernel to reduce noise. Registration was conducted through a 3-step procedure, whereby the mean EPI image was first registered to the matched-bandwidth high-resolution structural image, then to the MPRAGE structural image, and finally into standard [Montreal Neurological Institute (MNI)] space (FSL 3.2 MNI avg152 template), using a correlation ratio cost function and 12-parameter affine transformations (Jenkinson et al., 2001). This process was performed separately for each imaging run. Statistical analyses were performed in native space, with the statistical maps normalized to standard space prior to higher-level analysis. The normalized intersection mask image (demonstrating voxels present for all subjects in the analysis) is presented in Figure 1. Anatomical regions were identified by manual inspection using the Duvernoy atlas and the Harvard-Oxford Probabilistic Atlas (from FSL 4.0).

Figure 1. Normalized intersection mask overlaid on MNI avg152 template.

Whole-brain statistical analysis was performed using a multi-stage approach to implement a mixed-effects model treating participants as a random effect (Beckmann et al., 2003). Statistical modeling was first performed separately for each imaging run. Regressors of interest were created by convolving a delta function representing trial onset times with a canonical (double-gamma) hemodynamic response function, along with their temporal derivative. Time-series statistical analysis was carried out using generalized least squares in FILM (FMRIB's Improved Linear Model) with local autocorrelation correction (estimated locally at each voxel, tapered and regularized in space: Woolrich et al., 2001) after highpass temporal filtering (Gaussian-weighted LSF straight line fitting, with sigma=33.0s). Contrasts were performed to compare mirror-reversed versus plain items, and switch versus non-switch trials.

For each lower-level analysis, a second-level analysis was performed that combined all sessions for each participant, treating imaging runs as a random effect and pooling the between-runs variance estimate across subjects. The contrast maps from the second level were analyzed at the third level using a one-sample t-test at each voxel for each contrast of interest, using mixed-effects inference with a Bayesian 2-level model with fast approximation to the posterior probability of activation (Beckmann et al., 2003; Woolrich et al., 2004).. Whole-brain familywise error was controlled at p<.05 using nonparametric randomization tests (FSL randomize tool: Nichols & Holmes, 2002) with variance smoothing of 8 mm, a cluster-forming threshold of t > 2.0, and 5000 iterations. The 0.05 FWE-corrected critical cluster size was 75 voxels; The search region included 394,053 voxels. Surface renderings of group statistical maps were created using multifiducial mapping to a population atlas using the CARET software (http://brainmap.wustl.edu/caret/: Van Essen, 2005).

ROI analysis

Regions of interest were defined from significant clusters in the mirror vs. plain contrast, for the purpose of signal characterization. Percent signal change was computed from average parameter estimates using the height of an isolated event as the scaling factor, and is relative to the voxel mean.