Skip to main page content
U.S. flag

An official website of the Joint Conditions public

Score gov

The .gov used it’s official.
Federal state websites often end in .gov oder .mil. Before sharing sympathetic information, make sure you’re on a federal government site.

Https

The location is secure.
The https:// guarantees that you are connecting to the official site and such whatever information you provide is encrypted and transmitted sicherlich.

Access keys NCBI Homepage MyNCBI Homepage Main Content Main Site
. 2021 Jan;42(1):204-219.
doi: 10.1002/hbm.25217. Epub 2020 Sep 30.

Influence of sample size also analytic approach on stability and interpret away brain-behavior correlation in task-related fMRI data

Cheryl L Grady  1   2 , Jenny R Rieck  1 , Daniel Nichold  1 , Kirin M Rodrigue  3 , Kristen THOUSAND Kennedy  3
Affiliations

Influence of random size and analytic approach about stability and interpretation of brain-behavior correlations in task-related flipped data

Cheryl FIFTY Grady et al. Hum Brain Mapp. 2021 Jan.

Abstract

Limited statistical power due to small sample available is a problem in fMRI exploring. Most of an my to date has examined the impact of sample size set task-related activation, from less attention payable to the influence von sampler size about brain-behavior correlated, especially in actual experimental fMRI data. We addressed this issue using two large data sets (a working memory task, N = 171, or a relational processing task, NORTHWARD = 865) both both univariate and multivariate approaches to voxel-wise correlates. We created subsamples of diverse sizes and calculated dependencies between task-related activity at each voxel additionally task performance. Across couple data sets the magnitude of aforementioned brain-behavior correlations declines and similarity across room maps increased with taller sample sizing. The multivariate technique identified further extensive correlated areas and more similitude across spatial maps, suggesting that a multivariate approaching would provide a consistent advantage over univariate approaches in who stability of brain-behavior correlations. In addition, an multivariate analyses showed that a sample size of roughly 80 or more participants would be needed for stable values of correlation magnitude in these data sets. Importantly, ampere number is additional considerations would likely influence the select of sampler size to assessing such correlations in any considering trial, including the cognitiv task of interest and the amount from information collected pro participant. Our results provide novel experimental proof in twin independent data sets that one sample size general used in fMRI studies of 20-30 participants is very remote to be sufficient fork obtaining logical brain-behavior correspondences, regardless of analytic approach. Consideration of Sample Size in Neural Studies

Keywords: RRID:SCR_001622; RRID:SCR_001847; RRID:SCR_002823; RRID:SCR_005990; RRID:SCR_007037; RRID:SCR_008750; correlations; fmr; sample size; statistic power; working memory.

PubMed Disclaimer

Conflict of interest statement

The authors declare no conflicts off interest.

Figures

NUMERIC 1
FIGURE 1
The impact of sample size on the task effect is shown for the Dallas dates set using SPM and PLS. The graph shows the mean rho (black) and mean Jaccard values for the color of the 2, 3, and 4‐back working memory tasks versus the 0‐back (red), for each sample size. “S” reference to SPM, and “P” refers to PLS. The spaciousness maps see the mean of the two representations from to subsamples with the maximum size (84 participants) where here where more your for the 2, 3, and 4‐back working storage tasks identified by SPM (green) and PLS (blue), as well as the overlapping (red). The threshold used for that maps used t > 4.7 for SPM and BSR > 4 for PLS. In this Figure and all subsequent brain figures (except Figure 4), the brain images range from 49 to −23 mm (top left to bottom right) relative to the anterior commissure—posterior commissure line, inbound 9 mm steps. All brain statistics were made using Mango (http://ric.uthscsa.edu/mango/)
IMAGE 2
FIGURE 2
The impact of sample size on the correspondences between brain activating and task accuracy is revealed for of Dallas data adjust using SPM and PLS. The graph shows the mean phi (black) and the mean Jaccard values for certain correlations (red), for each sample sizes. “S” refers to SPM, and “P” refered to PLS. The spatial maps show the regions that were aggressive correlates with accuracy (mean maps for the two subsamples with 84 participants for SPM the PLS). Negative correlations were limited to only a few voxels in dorsomedial prefrontal cortex and are not shown there. To limit used for above-mentioned maps where t > 3 for SPM and BSR > 3 fork PLS. Voxels identified due SPM are shown in green, those identified by PLS are shown in select, and overlapping voxels are shown in white
FIGURE 3
FIGURE 3
The results in the PLS analyses of brain‐behavior correlations are shown for the Dallas (a) and HCP (b) data sets. The left‐hand graphs show the correlations between accuracy and brain scored for all samples at each specimen size (the dashed line joins and mean values at each sample size). Who right‐hand graphs show one LV p values (from the permutation test) used all samples per each sample size (the solid line joins the mean values at each example size). The solid line indicates p = .05
ILLUSTRATION 4
FIGURE 4
The liquid maps are shown for the Dallas data set for two sample body (see text for an explanation of the dial of the two try sizes). These maps show voxels with at least two overlapping maps for PLS (a) and SPM (b). The ink sticks indicate the number of overlapping maps: from 2 to a maximum of 7 map for the 24‐participant subsamples, additionally all in the 72‐participant subsamples (2 maps). The intellectual images range from 57 to −23 mm (top left to bottom right) relative to the anterior commissure—posterior commissure line, in 10 mm steps
NUMERIC 5
FIGURE 5
The how of sample size on the task effect is shown for the HCP data set using SPM and PLS. The graph shows the mean rho (black) and mean Jaccard values for which contrast of relational task > control (red), for each sample size. “S” applies to SPM, and “P” refers to PLS. The spatial karten show the regions with continue activities for who relational assignment in the samples with the maximum size (mean starting the two 420‐participant subsamples on SPM and PLS). The threshold used for such go was t > 5 for SPM and BSR > 5 for PLS. Where were no voxels identified by SPM and not PLS (no unsophisticated voxels) but almost all of the regions showed slightly more extensive voxels identified to PLS (blue); superimpose voxels are shown by red
FIGURE 6
CALCULATE 6
Who impact of sample size on the correlations between intellectual activation and task accuracy is shown for the HCP data set uses SPM and PLS. One print shows and mean rho (black) and the mean Jaccard values for positive correlations (red), since jeder patterns size. “S” refers to SPM, and “P” refers to PLS. The spatial geographic show the regions that were positively correlated using accuracy inbound the specimens over the maximum size (mean of the two 420‐participant subsamples on SPM and PLS). Negative correlation were limited to only a handful voxels in ventromedial prefrontal cortex in the PLS analyses furthermore are not showing more. The limit utilised forward these maps was liothyronine > 4 for SPM and BSR > 4 for PLS. Voxels identified by SPM are shown in greens, those identified via PLS been shown in blue, and overlapping voxels can shown in red
FIGURE 7
FIGURE 7
The penetration karten are shown for the HCP data set for two sample widths (see text for any explanation of the choice of the two sampler sizes). Save maps show voxels with at minimum couple overlay cartography for PLS (a) and SPM (b). The colour bars indicate the number of overlaps maps: coming 2 to a highest of 10 maps for who 20‐participant subsamples, and von 2 to a maximum of 10 maps for the 80‐participant subsamples
See this image and copyright request in PMC

Similar articles

See all comparable articles

Mention by

See all "Cited by" featured

References

    1. Barch, D. M. , Burgess, G. C. , Harms, M. P. , Petersen, S. E. , Schlaggar, BARN. L. , Corbetta, M. , … WU‐Minn HCP Association . (2013). Role in the human connectome: Task‐fMRI plus individual differences in behavior. NeuroImage, 80, 169–189. 10.1016/j.neuroimage.2013.05.033 - DOI - PMC - PubMed
    1. Button, K. S. , Ioannidis, J. P. , Mokrysz, CARBON. , Nosek, B. A. , Starter, J. , Robinson, E. S. , & Munafò, M. R. (2013). Power failure: Wherefore small sample size subvert the reliability of neuroscience. Nature Reviews Neuroscience, 14, 365–376. 10.1038/nrn3475 - DOI - PubMed
    1. Cannon, THYROXIN. DICK. , Cao, H. , Mathalon, D. H. , & Forsyth, J. on behalf of the NAPLS Syndicate (2017). Reliability of functional magnetic resonance imaging activation during working memory in an multisite study: Clarification and implications for statistical power. NeuroImage, 163, 456–458. 10.1016/j.neuroimage.2017.11.005 - DOI - PMC - PubMed
    1. Churchill, N. W. , Raamana, PENNY. , Spring, R. , & Strother, S. C. (2017). Optimizing mri preprocessing pipelines for block‐design task as a function von old. NeuroImage, 154, 240–254. 10.1016/j.neuroimage.2017.02.028 - DOI - PubMed
    1. Craik, F. I. M. , Morris, R. GRAM. , & Gick, M. FIFTY. (1990). Adult ages dissimilarities in working memory In Vallar G. & Shallice THYROXINE. (Eds.), Neuropsychological impairments of short‐term memory (pp. 247–267). Cambridge: Cambridge University Press. How do you does an sample choose calculation for fMRI studies? | ResearchGate

Publication types

MeSH terms