Computational protocol: Associative Processing Is Inherent in Scene Perception

Similar protocols

Protocol publication

[…] Due to participant drowsiness and movement during the later course of the experiment, as well as adaptation to the stimuli, only fMRI data from the first two association runs were analyzed. For a more detailed description of why only run 1 and run 2 were included in the analysis, please refer to the S1 Supporting Methods.Preprocessing. Functional data was analyzed using SPM8 ( All data were realigned to correct for minor head motion by registering all images to the mean image and the anatomical image was co-registered with the functional images. For the ROI analyses, functional data from the experimental conditions were subjected to no additional preprocessing steps. Thus, all ROI analyses were only preprocessed for correcting for motion and did not have additional smoothing. For the group average of the whole brain analysis, the association data were normalized to the MNI template for averaging purposes, and smoothed with using an isotropic Gaussian kernel 4mm FWHM. Finally, the scene localizer fMRI data were smoothed using an isotropic Gaussian kernel (FWHM = 6mm).General Linear Model. fMRI data were analyzed in a block design paradigm using a canonical hemodynamic response function. Each event was modeled within a 16s time window, and a high pass filter using 128s was implemented. The six output parameters from realignment were used as nuisance regressors within the model. The general linear model incorporated a robust weighted least squares (rWLS) algorithm [] which yields estimated the noise covariates and temporal auto-correlation for later use as covariates within the design matrix. The association design modeled nine conditions: SPID, SP, ID, NA, scene category (3), objects, and scrambled. The scene localizer design modeled two conditions: scenes and objects. For the whole brain analysis in the group average the contrasts were passed to a second-level random effects analysis that consisted of testing the contrast against zero using a voxel-wise single-sample t-test. All group maps presented are whole brain analysis using an FDR correction of q < .05, minimum cluster size k = 10. For visualization purposes group average maps were rendered onto 3D inflated brains using the CARET software [].Region of interest (ROI) analyses. All ROI analyses were performed at the individual level using the MarsBaR toolbox ( and analyzed within native space. Data from the contrast of scenes versus objects in the separate scene localizer were used to define scene-selective regions within the PPA, RSC, and OPA for each individual. Typically, a threshold of FWE p < .01 was used to define the set of voxels, and there were no overlapping voxels across ROIs. For the anterior and posterior analysis of the PPA region, the PPA ROI was defined by first anatomically labeling the parahippocampal cortex (PHC) in each individual [,]. Using the voxel coordinates from the y-axis, the PHC was then divided in four equal (in y-domain) sections. These subregions were then functionally masked to only include voxels that were significantly active in the scene versus object contrast of the localizer. However, scene related activity within the parahippocampal/lingual region extended beyond the posterior border of the parahippocampal cortex and into the lingual gyrus. We included these more posterior voxels as the most posterior section, giving us a total of five sections of the parahippocampal place area from anterior to posterior regions. Average number of voxels across the subregions was 127 in the LH, and 141 in the RH. There were significant differences in the number of voxels across the subregions (LH: F(4,56) = 8.89, p < 0.000012; RH: F(4,56) = 19,85, p < .0000001). These differences, however, arise from middle subregions containing more voxels than the end subregions. Critically, in the LH the most anterior and most posterior regions were not significantly different in number of voxels (planned comparisons; p > .55), in the RH the anterior region had slightly more voxels than the most posterior region (p < .05). These ROIs were then applied in the analyses, paired t-tests and repeated measures ANOVAs, of the association data to extract weighted parameter estimates (i.e., beta values) averaged across all voxels for each condition compared with the baseline or the NA condition.In the cross correlation analysis, unthresholded t-values from each of the voxels of a specific ROI were extracted from the contrasts of interest. These t-values were then cross-correlated on a voxel-by-voxel level within each ROI across the different contrasts as specified in the results section (e.g., Scenes vs. baseline correlated with SPID vs. NA). R-values for each individual contrast comparison, for each ROI, were then Fisher corrected to perform additional paired t-tests and repeated measures ANOVAs. […]

Pipeline specifications

Software tools SPM, caret
Databases SPiD
Application Functional magnetic resonance imaging
Organisms Homo sapiens