Okay - here it goes. Basically what you need to do is an element-by-element multiplication of the singular images from the two LVs that you want to assess. You will need to apply a mask to the images first so that you are emphasizing those voxels that are reliable in both LVs. This cannot be done in the GUI so you will need to do this from the command line in matlab.
Here are some detailed instructions:
1. Make a copy of the results file containing the LVs of interest
cp mytext_fMRIresult.mat mytext2_fMRIresult.mat
where "mytext" is the prefix you created for the results file of interest
2. Start matlab and load the copied results file
load mytext2_fMRIresult
3. Once done if you type "whos" you will get a list of the variables that are stored in the results file. The ones we are most interested in are:
brainlv (the singular images)
s (the singular values)
boot_result.compare (the bootstrap ratio images)
Note that the dot between boot_result & compare is intention as boot_result is structure array4. Create a binary mask of the bootstrap ratio images. We'll use the absolute value
mask_lv1=abs(boot_result.compare(:,1)>=3);
where the value '3' is your threshold - use what you think is valid - also make sure you end the line with a semi-colon ';' otherwise the numbers get written to the screen
repeat this for the other LVs of interest
5. At this point there are two options - create the conjunction of boostrap images or singular images
BOOTSTRAP
LV1=mask_lv1.*boot_result.compare(:,1);
LV3=mask_lv3.*boot_result.compare(:,3);
conj_LV1LV3=LV1.*LV3;
you have to use the '.*' and not just '*' for these operations. The dot '.' means element-by-element
SINGULAR IMAGE
LV1=mask_lv1.*(brainlv(:,1)*s(1));
LV3=mask_lv3.*(brainlv(:,3)*s(3));
conj_LV1LV3=LV1.*LV3;
the operation (brainlv*s) rescales the brainlv by its singular value putting it back into 'covariance' space.
The reason for using one vs another has to do with statistical purity. Technically, the singular image approach is more valid since it is the conjunction of the parameter estimates, which, in principle, can be subjected to a statistical test. The conjunction of bootstrap ratios cannot be as readily tested because the exact value depends on both the parameter estimate and the standard error.
6. Replace the fields in the results file with the conjunciton image:
BOOTSTRAP
boot_result.compare(:,1)=conj_LV1LV3;
SINGULAR IMAGE
brainlv(:,1)=conj_LV1LV3;
s(1)=1;
replacing the singular value with '1' will prevent rescaling of the conjunction image in the display
7. Save the results file
save mytext2_fMRIresult
8. Start plsgui and load this results file. If you created the bootstrap conjunction, this should come up and you can view it and run the cluster report on that image. If you created the singular image, then you click "View:View Brain LV" and then do your cluster report.
It is worth emphasizing that neither of these approaches is a statistical assessment of the conjunction ala SPM. It is a description of the overlap between images. However, it should suit your needs. One could do the PLS conjunction in the course of bootstrap resampling, but that is not part of the current package. I would have to evaluate it a little more before it would get implemented.