Fig. 5
From: Individual Brain Charting dataset extension, third release for movie watching and retinotopy data
Statistical validation of naturalistic-stimuli tasks with FastSRM. (a) Person-correlation coefficients (ρnormalized) of FastSRM prediction for Raiders task, compared to noise ceiling. The noise ceiling was estimated as correlations between run pairs 1-11, 2–12 and 3–13 that refer to the same video excerpts. The correlation between runs 1, 2 and 3 with their reconstructed runs from FastSRM are expressed as a fraction of this noise ceiling. These ratios were computed for every vertex and subject. We then took the median across subjects, normalized and took the median across runs. Results are thresholded at 0.1. (b) Group-level, Pearson-correlation coefficients (ρ) between the original and reconstructed data for Raiders and Clips tasks. Coefficients were obtained for every vertex from a double K–fold cross-validation experiment across subjects (K = 3) and runs (K = 2) for each task. Data of test subjects performing test runs were reconstructed from the projection of the shared response of train subjects while performing test runs onto the individual components of test subjects while performing train runs. Predictions between original and reconstructed data were estimated for every subject and run. To obtain the group-level estimation of the coefficients, the vertex-wise median of the coefficients were firstly taken within split-halfs, secondly between split-halfs for every subject, and finally across subjects. To assess the group-level significance of these estimates, we computed a mass-univariate non-parametric analysis, then derived an FDR-corrected p-value. Coefficients are only displayed for vertices with q ≤ 0.05. (c) Group-level z–maps displaying brain activation significantly different between Raiders and Clips tasks. Results were determined through a vertex-wise paired t–test between the individual Pearson-correlation coefficients of the two tasks and standardized afterwards. Statistical significance was assessed using an FDR-corrected threshold q = 0.05. Clusters depicted in orange/yellow represent brain activation significantly higher for Raiders than Clips, whereas those depicted in dark/light blue represent brain activation significantly higher for Clips than Raiders. Orange-yellow clusters surpass in number and size the blue clusters, highlighting the additional cognitive recruitment in Raiders related to auditory and language comprehension specific to this task.
