[Homer-users] GLM gaussian filter, CreateAuxRegressor, Wavelet

[Defenderfer, Jessica]

        External Email - Use Caution        

Good morning and thanks in advance for any advice/direction in response to my questions:

I have read the posts on this forum about Wavelet, and I’m still unclear about which value to input into the Homer2 version of Wavelet. The description of hmrMotionCorrectWavelet (when hovering over in EasyNIRS) states that the typical value used is (iqr) 0.1; also the default value of iqr 0.1 is already input in the EasyNIRS stream.  However, other posts on this forum explain that homer version of Wavelet is slightly different and uses IQR set to 1.5. Alpha and IQR have opposite effects on how many coefficients are set to 0, so I want to be clear on which one I should be using.

Second, I’m using hmrDeconvHRF_DriftSS to regress out my short channel. Ive read that using idxbasis1 with Gaussian filters gives me more flexibility (and also appears to look more similar to the raw data waveforms compared to idxbasis2 gamma function). I have 27 beta coefficients because I used .5 as my spacing value- what is the protocol to analyze the beta coefficients? Do you objectively select the highest one?

Last question (I promise) :)  I am working with data obtained from a speech perception task. I have about 180 trials (across 5 different conditions). Each trial has two time stamps: the first time stamp marks onset of the stimulus (they hear a short sentence) and this data is recorded in aux(:,1). The second time stamp cues the participant to repeat the sentence; this data is recorded in aux(:,2). I have the trials sorted into their various conditions in the s matrix, and put all of the time stamps associated with repeating the sentence into it’s own column. My question is how do I use the aux regressor to remove HRF associated with the speaking portion (second time stamp) from the HRF associated with hearing the sentence (first time stamp)? So far, I’ve used the hmrCreateAuxRegressor and entered [2  0  5] assuming that the deconvolution step will look to column 2 of aux and regress it out of column 1 of aux. However, there are no differences in the HFRs when I run my processing stream with the aux regressor compared to running it without the aux regressor. I think I’m missing something. For reference, here is the order of my processing stream:

IntensitytoOD
hmrBandpassFilt [.01  2]
hmrMotionCorrectWavelet [1.5]
hmrCreateAuxRegressor [2  0  5]
hmrBandpassFilt [.01  .05]
hmrOD2Conc [6  6]
hmrBlockAvg [-2  12]
hmrDeconvHRF_DriftSS
trange [-2 12]
glmsolvemethod [1]
idxbasis [1]
paramsBasis [.5  .5]
rhoSD_ssThresh [15]
flagSSmethod [1]
driftORder [3]
flagMotionCorrect [0]
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mail.nmr.mgh.harvard.edu/pipermail/homer-users/attachments/20180705/80a03f18/attachment.html