Print

Print


Dear Fuyu
There is no 'correct' way to do VOI extraction - you can do what seems sensible to you, as long as you can justify it in your paper. Here's a sensible procedure:

1. View your group level results showing the contrast that interests you (FWE corrected < 0.05). This could be a T-contrast or an F-contrast.

2. In the Results window, click save, then "all clusters (binary)". Give the image a filename.

3. View the results of one subject, using a the same contrast as step 1 (now uncorrected p < 0.05). For "apply masking" select "Image", and select the image from step 2.

4. Move the cursor (the crosshairs) to your first ROI of interest. Then right click on the "glass brain" (the transparent black-and-white brain at the top of the Graphics window), and click "goto nearest local maximum". You might like to overlay sections to ensure that this peak is sensible.

5. Click eigenvariate. Give the ROI a name and for adjustment, select your Effects of Interest F-contrast (as detailed in the previous email I linked to). You can think of each row of the F-contrast being combined using the 'OR' operator. So it includes effects covered by row 1 of the contrast, OR row 2, OR row 3 etc.

6. Go back to Step 4 and repeat for each ROI with the group-level mask you are interested in.

7. Go back to step 3 and repeat for the next subject

If you wish to have different ROIs identified using different contrasts, then repeat from step 1.

Best
Peter

From: #KWOK FU YU# [mailto:[log in to unmask]]
Sent: 03 April 2017 02:28
To: Zeidman, Peter <[log in to unmask]>
Cc: [log in to unmask]
Subject: Re: Creating VOI for DCM
Importance: High

Dear Peter,

Not sure if you received my previous email.

I was wondering for the manual option, to create the group level mask, do we have to first mask over a ROI that we're interested in (E.g. FG) at the second level? That way, if i'm interested in 7 ROIs, I will create 7 different group level mask.

Also, for the group level mask-binary mask, do you mean the save-> all clusters (binary)?

Lastly, for DCM, why do you recommend F-contrast as oppose to T-contrast?

Kind Regards,
Fuyu

On 30 Mar, 2017, at 10:15 pm, #KWOK FU YU# <[log in to unmask]<mailto:[log in to unmask]>> wrote:


Dear Peter,

Thank you for your detailed explanation. Think I'd stick with the manual option.

To create the group level mask, do we have to first mask over a ROI that we're interested in (E.g. FG)? That way, if i'm interested in 7 ROIs, I will create 7 different group level mask.

Also, for the group level mask-binary mask, do you mean the save-> all clusters (binary)?



On 30 Mar, 2017, at 9:03 pm, Zeidman, Peter <[log in to unmask]<mailto:[log in to unmask]>> wrote:


Dear Fuyu

Thanks for your advice. I'm not fully sure that I understand how the batch file and script work. It would be great if you could clarify some of my doubts:
1. For the batch file (example_voi_batch.mat), do I have to create it for every ROI in the individual subject? (i.e if I have 10 participants and each participant i'm creating 7 VOI, I would need 70 of the *_batch.mat files)?

Yes, the instructions I gave creates a batch for 1 VOI in 1 subject. However, the idea is to use this batch .mat file as a template. Then you use a script to loop over subjects and ROIs, updating the batch appropriately, and extracting the ROIs.
*         In the batch editor, for the SPM.mat, I presume we select the individual subject .mat file here -am i right?
Yes
*         For the adjust data, i'm not sure what do I put here. (i'm guessing "1" ?) I did both T and F contrast at the 1st level (one contrast each). I was interested at looking at High load > Low low so for the F contrast, it was   (1 0                                                                                                                                                                                                                                                                                                                                          0 1)
You should put the index of your Effects of Interest F-contrast. Please see the detailed instructions in my original email to Elisabeth and Zuo (step 1).  It should not just be high load > low load, but rather should be an identity matrix with 1s over each interesting regressor.
*         For the sessions, do I just put "1" if i've only 1 session?
Yes
*         For the contrast under the Region(s) of Interest, do I set it as "1" again since i've only 1 contrast?
Yes, it's the index of the contrast you wish to use for finding the peak. This could be the same as the effects of interest contrast (above) or it could be the index of a t-contrast, e.g. high load > low load.
2. For the script file (example_voi_script.m), I'm not sure how it works. It seems like it requires the *_batch.mat files.

The script file loads the template batch, and loops over subjects, extracting the ROI for each subject. So for your 7 ROIs, you could create 7 copies of this script. Please go through the script in detail before using it - I wouldn't want you ending up with data without knowing where it came from!

Lastly, if I were to do this manually, how should I go about using the local peak for each individual subject? Could I apply a ROI mask and obtain the peak activation within the ROI and create the VOI based on that? If so, how do I make sure it's within some maximum (i.e. 10 or 20mm) radius of the global peak?

To do this manually, you can of course open the results from each subject and click Eigenvariate. If you are not comfortable with the batch and scripting, this might be the best option. To guide you, you could save a mask of your group level results (click save -> binary mask). This will create a nifti image of the group level results. Then when you view the single subject's results, when you're asked if you want to mask results, select the nifti image from the group. You can then right click on the transparent (glass) brain and click Global Maximum.

Let me know if anything's unclear.

Best
Peter

Many thanks!

Kind Regards,
Fuyu

On 30 Mar, 2017, at 6:05 pm, Zeidman, Peter <[log in to unmask]<mailto:[log in to unmask]>> wrote:



Dear Fuyu
Best practice is to use the local peak for each individual subject, within some maximum radius of the global peak. If you'd like a script to do this automatically, see this recent post:

https://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=ind1703&L=spm&P=R55250&1=spm&9=A&J=on&d=No+Match%3BMatch%3BMatches&z=4

Best
Peter

-----Original Message-----
From: #KWOK FU YU# [mailto:[log in to unmask]]
Sent: 30 March 2017 05:56
To: Zeidman, Peter <[log in to unmask]<mailto:[log in to unmask]>>
Cc: [log in to unmask]<mailto:[log in to unmask]>
Subject: Re: Creating VOI for DCM

Dear Peter,

Thanks. I was just wondering if it would be better to reduce the threshold to obtain activation at the specific peak coordinate or would it be better to create the VOI based on the next nearest point of activation (if it's still within the ROI).

Kind Regards,
Fuyu


On 28 Mar, 2017, at 4:04 pm, Zeidman, Peter <[log in to unmask]<mailto:[log in to unmask]>> wrote:



Dear Fuyu
Nice to hear from you. I hope you don't mind me CC'ing the SPM mailing list so others can benefit. (I am always happy to help, but where possible, please could I ask you to email the mailing list rather than sending direct emails with queries?)



1. When creating the VOI (based on peak coordinates from 2ndlevel) for individual subjects, how low can we set the uncorrected threshold to be in order to obtain the activation at the peak coordinate/ within the area?
2. For the T-value at the VOI, is there a minimum value that should be obtained?

You could set the uncorrected threshold to infinitely low (i.e. p = 1). The purpose of the threshold here is to select which voxels to include in the ROI, rather than performing a statistical test, which you have already done at the group level. It can be advantageous to only include voxels which show some experimental effect. Otherwise, there's an increased risk that the first principle component represents a physiological or other noise process rather than relating to the task. So there's no right answer, but how about going for p = 0.05 as your most liberal value. And then relax it further if you can't get signal from every subject.

Best
Peter