AFNI version info (afni -ver): Precompiled binary linux_ubuntu_16_64: Apr 13 2025 (Version AFNI_25.1.03 'Maximinus')
Dear AFNI community,
I very recently started using AFNI specifically for the 3dISC function and the ability to apporopriately model the dependencies in ISC data.
However, I am not sure if 3dISC can be applied to my problem.
I have data of N = 48 participants from a 2 (Hierarchy: Shots, Scenes) x 3 (Duration: 4s, 12s, 36s) within-subject design, so from 6 conditions in total. My intial idea, coming from lme4 analysis in R was to simply model the data like this (simplifed Code):
But, if I understand it correclty, having to factors does not work, right?
The bigger issue however seems to be that through the within-subject design I have non-unique pairs in my data table. Thus, I always get the follwing error message: Error: the number of rows/files, 6768, is not equal to 1128 - the possible subject pairs!
Is there even any possible way I can model my 2 x 3 within-subject design in the 3dISC framework? Are there other options I have overlooked?
3dISC was originally designed to handle one input per individual pair. Could you clarify what effects you're aiming to investigate with your 2 × 3 within-individual design: specific main effects and interactions?
thank you for your fast response!
The idea of the analysis is to use ISC like in the papers on TRWs to identify areas that are consitently engaged by of differnt levels of duration (4s, 12s, 36s) and hierarchical structure (Shots and Scenes) to answer the question, whether hierarchically nested stimuli (like movies, in our case) are processed by the brain according the temporal duration of the individual levels, based on the different structural levels alone, or based on the combination or interaction of the two.
So, we presented the participants with 6 different sequences of stimuli that either contained only shots or entire scenes (2) of different temproal durations (3).
We now want to compare the ISC maps between the 6 conditions to - if possible - be able to identify the main effect of hierarchy, so how the ISC differs between scenes and shots, the main effect of duration, if there are differences between the 3 levels of duration, as well as possible interactions. The factor "Duration" can maybe be modelled continously even though I would prefer the 3 levels.
Regarding the specific comparisons:
I created a model, where I tried to encode the factors already in the data table to circumvent the limit of one factor in 3dISC and came up with this set-up inspired by the Example 4 in the 3dISC documentation:
The program 3dISC is designed to handle one input file per pair, so the model you specified wouldn’t work as intended. I suggest the following: for each of the estimations in the lines below from your script,
the approach works very well and the first results look promising. But, I was wondering how I should correct the results for multiple comparisions, first across voxels and then across the multiple separate analyses?
For each result I was thinking about the approach that Paul Taylor suggested on the NeuroStars Discussion Board:
I would start with going back to 3dFWHMx on the residuals (“errts”) time series, averaging the ACF parameters, and going to 3dClustSim with that. I think a standard place to start would be p=0.001, bisided, alpha=0.05; whichever NN level you feel is appropriate is certainly fine, it just has to be maintained consistently with 3dClustSim and 3dClusterize.
So, I would firstly use the residual files that went into the respective ISC maps involved in each anaylsis to estimate the smoothness and then simulate the approproate cluster threshold. Is this approach reasonable?
Secondly, then the question remains if I need to correct for multiple comparisons across the separate analyses. Could I just adjust the alpha level of the FWE corrected threshold to accommodate for the number of contrasts?
Thank you very much in adnvance!
Best wishes,
Falko
You’re raising a controversial and challenging issue in neuroimaging. While the dominant view in the field has favored stringent multiple testing adjustments, I tend to disagree with both the underlying assumptions and the methodological approach.
The prevailing modeling strategy, massive univariate analysis, treats each voxel independently, assuming no shared information across the brain. This assumption is fundamentally flawed and gives rise to the multiple testing problem. Although many adjustment methods attempt to account for local spatial structure via smoothness, the resulting penalties are often excessively conservative.
Ideally, multiple testing should be addressed through hierarchical modeling. We’ve demonstrated a proof of concept at the region level in this paper, and we hope to extend this framework to voxel-wise analysis in future work.
Gang Chen
The
National Institute of Mental Health (NIMH) is part of the National Institutes of
Health (NIH), a component of the U.S. Department of Health and Human
Services.