Insight Meta-Analysis fMRI Maps
Sprugnoli, G., Rossi, S., Emmendorfer, A., Rossi, A., Liew, S.-L., Tatti, E., di Lorenzo, G., Pascual-Leone, A., Santarnecchi, E. (2017).
“Neural correlates of Eureka moment. Intelligence” https://doi.org/10.1016/j.int ell.2017.03.004
Fluid Intelligence fMRI Maps
Santarnecchi, E., Emmendorfer, A., Pascual-Leone, A., on behalf of Honeywell SHARP Team authors (2017)
“Dissecting the parieto-frontal correlates of fluid intelligence: A comprehensive ALE meta-analysis study”. Intelligence https://doi.org/10.1016/j.intell.2017.04.008
ALE map computation
The quantitative evaluation of spatial fMRI patterns was carried out using the activation likelihood estimate (ALE) technique implemented in the GingerALE software v2.3.2 (www.brainmap.org) (Eickhoff et al., 2009; Eickhoff et al., 2012). The method yields a statistical map that indicates the set of brain voxels that are more active than would be expected by chance. The ALE method assumes that for each study of interest there is a given spatial distribution of activity and an associated set of maximal coordinates. Therefore, the algorithm tests to what extent the spatial locations of the activation foci correlate across independently conducted fMRI studies investigating the same construct.
First, the lists of coordinates were carefully checked for duplication of data across publications, in order to avoid artefactual inflation of a given foci significance. Coordinates collected from studies reporting activation foci in Talairach space were converted into the MNI space using the tal2mni algorithm implemented in GingerALE. Activation foci from each study were modeled as Gaussian distributions and merged into a single 3D volume. The ALE algorithm modeled spatial uncertainty of each activation focus (Turkeltaub et al., 2012), using an estimation of the inter-subject and inter-study variability usually observed in neuroimaging experiments, rather than applying an apriori full-width half maximum (FWHM) kernel. Therefore, the number of participants in a given study influenced the spatial extent of the Gaussian function used. We first modeled the probability of activation over all the studies at each spatial point in the brain, returning localized “activation likelihood estimates” or ALE values. Values were then compared to a null distribution created from simulated datasets with randomly placed foci, in order to identify significantly activated clusters (permutations test = 1000 run). Following Eickoff and colleagues arguments supporting a better balance between sensitivity and specificity for Cluster-based corrections over False-Discovery-Rate (FDR) and Family Wise Error (FWE) approaches (Eickhoff et al., 2012), we applied cluster correction for multiple comparisons with a p < 0.001 threshold for cluster-formation and a p < 0.05 for cluster-level inference. Only clusters with a size exceeding the cluster size recommended by ALE were reported (range 500–1000 mm3). Given the potential overlap of the different ALE maps (e.g. vGf and vsGf, RI and RA), specific statistical comparisons have been computed in order to identify segregated neurobiological signatures of each component as well as conjunction maps showing significant overlaps. The procedure involves the creation of a combined map including the two maps of interest (i.e. including all the activation foci), using the voxel-wise minimum value of the input ALE images. Then, two contrast images are created by directly subtracting the two ALE maps, together with a map showing their statistically significant overlap. It is important to notice that the resulting subtraction image does not take into account differences in the dataset sizes between the two original maps. Therefore, simulated datasets are created by pooling all the available foci and randomly dividing them into two groupings with the same sizes as the original datasets. An ALE image is created for each new dataset, subtracted from the other and then compared to the real data. The process is computed 10,000 times, and a voxel-wise p value image is obtained. Values in each voxel represent the position of real data with respect to the distribution of values obtained during the permutation test. To ease the comprehension of the results, ALE contrast images are converted to Z scores.
This procedure was applied to each of the aforementioned coordinate lists. Results are then express as clusters of activation using Z score values in the image statistics and maxima value. Anatomical labels of final cluster locations were provided by the Talairach Daemon (http://www.talairach.org/daemon.html). ALE maps were visualized using MriCron (Rorden & Brett, 2000) on an MNI standard brain. Moreover, in order to provide anatomical mapping of activation foci according to the recently developed new multimodal parcellation of the human brain (Glasser et al., 2016), ALE volumetric maps have been converted to surface space and overlap to the new atlas using Connectome Workbench software.