Event duration in GAM model (-stim_times)

AFNI Message Board
1

AFNI version info (afni -ver): 24.3.10

Hi,

I’m running a GLM in AFNI using the -stim_times option with 'GAM' as the response model (Code at the bottom).

From my understanding, 'GAM' is a fixed-shape gamma variate model that does not take explicit durations from the input file - just the onset times. In the GAM model, what “actual” duration is assumed for each event?

Additionally, could you explain the theory behind using a fixed-duration model like 'GAM'? I’m trying to understand the reasoning for treating events without specifying a duration.

For context, it is for a lexical decision task (LDT) experiment with a fast event-related design. On each trial, a word or pseudoword is presented very briefly, and participants press a button immediately after stimulus onset to indicate whether it is a word or a pseudoword. The trials are short, and the interstimulus intervals are jittered.

The below code is my 3dDeconvolve code.

3dDeconvolve \
        -input ${dir_preproc}/pb06.${subj}.r0?.scale+tlrc.HEAD \
        -censor ${dir_preproc}/motion_${subj}_censor.1D \
        -ortvec ${dir_preproc}/motion_demean.1D 'motion_demean' \
        -polort A \
        -float \
        -allzero_OK \
        -num_stimts 3 \
        -stim_times 1 ${dir_behav}/word_${subj}.1D 'GAM' -stim_label 1 word \
        -stim_times 2 ${dir_behav}/nonword_${subj}.1D 'GAM' -stim_label 2 nonword \
        -stim_times 3 ${dir_behav}/nogo_${subj}.1D 'GAM' -stim_label 3 nogo \
        -gltsym 'SYM: +word +nonword -nogo' -glt_label 1 WordNonword_Nogo       \
        -gltsym 'SYM: -word -nonword +nogo' -glt_label 2 Nogo_WordNonword       \
        -gltsym 'SYM: +word -nonword' -glt_label 3 Word_Nonword \
        -gltsym 'SYM: +word -nogo' -glt_label 4 Word_Nogo       \
        -gltsym 'SYM: +nonword -nogo' -glt_label 5 Nonword_Nogo \
        -fout -tout \
        -x1D X.xmat.${subj}.1D \
        -xjpeg S.${subj}.jpg \
        -bucket stats.WordNonwordNogo.${subj}

Thank you in advance!

2

Hello,

The GAM function is Mark Cohen's 1997 model of a response to a 1s light flash, as measured in the visual cortex.

For events that are close to the same duration (including a duration of 0.0001 s and 1 or maybe even 2 s), we do not much variation in the shape of a BOLD response. But for short events, we do expect a large effect of duration on the magnitude of the response. Of course, whether a given shape applies to different parts of the brain is a separate and large question.

So if the events are all the same (short) duration, using a fixed-shape regressor might be reasonable. However, it sounds like your event durations might vary a bit, based on how long it takes for a subject to make a decision. What is the range of response times? What is the difference among stimulus classes?

-rick