Forum

Notifications
Clear all

GLM Issue and Analysis Queries

3 Posts
3 Users
0 Likes
997 Views
Owais Ahmed Khan
Posts: 2
Topic starter
(@ok25775)
New Member
Joined: 2 years ago

Hello David and Meryem,

I had the following queries about the GLM function in Homer3:
1. I'm getting the following MATLAB output on running the hmrR_GLM function in Homer3: "Design matrix is VERY poorly scaled...(RCond=0.000000e+00), cannot perform computation". Concurrently, when I try to view the HRF in the Homer3 GUI, I am unable to do so, although I can still view the concentration values in the Homer3 window. Could you point out why this seems to be the case? Surprisingly, I face this issue only during processing some, but not all files. Is there something I am missing out on, or should be doing?

2. Does the GLM function account in any way for motion artifacts and baseline shifts? Our lab works primarily with children, so our data can get quite contaminated with artifacts. I know we partly account for systemic physiology through the driftorder in GLM (we unfortunately do not have auxiliary measurements or SS channels for our data), but should I be running some motion correction functions (e.g. Spline+Wavelet or SplineSG) prior to running the GLM function? 

3. For the files I can successfully run the GLM function for, how can I obtain the beta and p-values? Based on our training video, we only discussed extracting the HRF following the processing. However, if I want to use the GLM in contrast to block-averaging, how should I go about the analyses? I'm new to the GLM-based analyses, any guidance you could provide would be much appreciated!

Thank you!
Owais

Topic Tags
glm
2 Replies
Meryem Yücel
Posts: 141
Admin
(@mayucel)
Estimable Member
Joined: 3 years ago

Hi Owais,

 

Please see my responses below:

1. I'm getting the following MATLAB output on running the hmrR_GLM function in Homer3: "Design matrix is VERY poorly scaled...(RCond=0.000000e+00), cannot perform computation". Concurrently, when I try to view the HRF in the Homer3 GUI, I am unable to do so, although I can still view the concentration values in the Homer3 window. Could you point out why this seems to be the case? Surprisingly, I face this issue only during processing some, but not all files. Is there something I am missing out on, or should be doing?

Please check the number of trials left for GLM in the runs that you get the design matrix warning. It typically happens when you have only a few trial left.

 

2. Does the GLM function account in any way for motion artifacts and baseline shifts? Our lab works primarily with children, so our data can get quite contaminated with artifacts. I know we partly account for systemic physiology through the driftorder in GLM (we unfortunately do not have auxiliary measurements or SS channels for our data), but should I be running some motion correction functions (e.g. Spline+Wavelet or SplineSG) prior to running the GLM function? 

We suggest using motion correction before GLM. Using a drift order can help with baselineshifts not necessarily with systemic contamination. Since you do not have SS channels you can try using the average of long channels as a global regressor in your GLM. You can do this by choosing option 2 for nuisances regression and setting the ss channel threshold to a high value that will allow long channels to be considered short.

 

3. For the files I can successfully run the GLM function for, how can I obtain the beta and p-values? Based on our training video, we only discussed extracting the HRF following the processing. However, if I want to use the GLM in contrast to block-averaging, how should I go about the analyses? I'm new to the GLM-based analyses, any guidance you could provide would be much appreciated!

GLM outputs beta values. After running homer3 at group level, you will have a groupresults.mat file in your main data directory. You can extract subj/run level beta values by loading this mat file and extracting it from the relevant field:

group.subjs(1).runs(1).procStream.output.misc

Beta output dimensions are: # of basis functions for HbO/R (will change depending on the HRF model you used) X # of channels X # of conditions.

 

Hope this helps.

 

Meryem

 

 

 

 

Reply
David Boas
Posts: 249
(@dboas)
Reputable Member
Joined: 2 years ago
Posted by: @mayucel

'm getting the following MATLAB output on running the hmrR_GLM function in Homer3: "Design matrix is VERY poorly scaled...(RCond=0.000000e+00), cannot perform computation".

When Rcond=0 that would likely mean you have no stim trials for that condition! If it is a very small number (not zero), then it means your design matrix is ill-conditioned which would result if you stimuli are close together and the intervals not randomized.

Reply
Share:
en_USEnglish