# Probability Weighting

Hello AFNI Experts,

I would like to examine activation within the amygdala subregions (centromedial amygdala, basolateral amygdala) to different stimuli (CS+, CS- in a fear conditioning paradigm). I have probability information, so that I know which voxels are least likely to compose of each of the subregions and which voxels have the lowest probability. I would like to incorporate this probability information into my analysis. I would like for example, voxels that are more likely to compose a subregion to carry more weight to the averaged BOLD activation than voxels less likely to compose the subregion. I have averaged tents for the CS+ and CS-. Can I simply multiply the probability weights with each of these tents. Then run 3dROI stats as usual on the probability weighted tents to get the averaged BOLD activation? I hope this is making sense. I would appreciate any guidance.

Thanks,

Emily

Emily,

I have averaged tents for the CS+ and CS-.

Did you use TENT as basis function to estimate BOLD response at the single subject level? Are you trying to average those betas associated with the TENT basis functions for each stimulus condition?

I have probability information, so that I know which voxels are least likely to compose of each
of the subregions and which voxels have the lowest probability.

Do you mean that you have two probability values at each voxel, each of which is the likelihood the voxel belongs to one of the two subregions?

For what purpose do you use the resulting “weighted” average?

Hi Gang,

Yes I used the TENT function as the basis function to estimate BOLD response at single subject level and have already averaged those betas for each stimulus condition.

Yes, each voxel was assigned to a specific subregion based on probability information. Each voxel belongs to only one subregion, so that there is no overlap. I guess what I am asking is that I would like the voxels that are most likely to belong to the subregion (like that voxels that have a 90% chance of belonging to the basolateral amygdala) to carry more weight to the average than those voxels that have only a 50% probability of belonging to the basolateral amygdala.

What I would like to apply something similar that is done with resting state data to task data. For example, Roy (2009) does the following:

To minimize effects due to interindividual anatomic variability, each voxel’s time series was weighted by the probability of inclusion in a given amygdala subdivision, based on the interindividual variability of the ten subjects used to construct the original anatomic atlas. In other words, those voxels most reliably located in a given region made the greatest contribution to its signal. In each subject, mean time series were then extracted by averaging across all voxels’ probability-weighted time series within each subdivision.

Thanks!

Emily