Hi,
I am trying to learn the VBM toolbox of SPM and I have strictly followed all the instructions from VBM8 manual(dbm.neuro.uni-jena.de/vbm8/VBM8-Manual.pdf) for Gray matter(GM) white Matter(WM) and Cerebro  spinal fluid(CSF) segmentation. I am trying to look at the difference in GM volume for 2 groups.

For the segentation, i have used all the defaults in Estimate and Write options in VBM8. After i have done the segmentation, i also used the "check data quality" to check my segmentation results. After smoothing, I used the 2 sample T test as my statistical model. I have strictly followed the manual for defining the parameters in this model. After running this model, i find that I have got a SPM.mat file in my analysis directory. After i Estimate my statistical model, I find that following images are generated in my analysis directory:
1. beta_0001.hdr
2. beta_0001.img
3. beta_0002.hdr
4. beta_0002.img
5. con_0001.hdr
6. con_0001.img
7. con_0002.hdr
8. con_0002.img
9. con_0003.hdr
10. con_0003.img
11. con_0004.hdr
12. con_0004.img
13. mask.img
14. mask.hdr
15. ResMS.hdr
16. ResMS.img
17. RPV.hdr
18 RPV.img
19. spmT_0001.hdr/img
20  SPMT_0002.hdr/img
21. SPMT_0003.hdr/img
22. SPM_0004.hdr/img

Also, when i define the contrast for group A vs group B (-1 1) and set the p value to  0.001(unc), then i find that the brain regions where i find significant difference in gray matter are not properly registered. Also, when i try to see only the brain for group A, I find that the brain is way out of the box and is not properly registered(I guess).
I have attached the images for your further understanding.
Following are my queries:

1.I am curious to know what these images mean which are generated after i have finished with the Estimation of my model.
2. Why am i facing this problem of getting improper results when i am strictly following the tutorial.Also, how to fix this problem.

Please reply and help

Thanks
Tanmay

Dear Tanmay,

Regarding 1, the images generated in the model estimation are:
- mask, a binary image indicating which voxels were considered in the analysis, (see e.g. http://www.ncbi.nlm.nih.gov/pubmed/18848... for more on this with regard to VBM)
- beta, the estimated linear model parameters at each voxel, in your case of a two-sample design, these are just the two group mean images
- con, the contrasts of the parameters, which in your case of a [-1 1] contrast will simply be beta_0002 - beta_0001
- spmT, the t-statistic maps formed by dividing the contrast image by its estimated standard error
- ResMS, the estimated variance (residual mean squares), from which the above-mentioned standard error is derived (in more detail, the standard error is proportional to the square root of ResMS and a term that relates to the design matrix X and contrast c as sqrt(c'*pinv(X'*X)*c), which in your case roughly reflects the fact that the standard error is inversely proportional to the number of subjects you have)
- RPV, the resels-per-voxel image related to the smoothness (roughness) estimate needed for random field theory (see e.g. http://www.ncbi.nlm.nih.gov/pubmed/15193... for more on RPV)

Regarding 2, I think the most likely problem is incorrect orientation of the original images, perhaps reflecting incorrect conversion from DICOM (or other format) to NIfTI. To check this, you can use the "check reg" button, go into the SPM templates directory and pick the T1 template, then pick one or more of your images. If the orientation is very far off, you won't be able to see both the template and the original at the same time, in which case use the "Display" button and just pick one of your images. To fix the orientation, you can either right-click your image after Check Reg and there is a context menu option for reorientation, or you can edit the orientation related parameters in the Display GUI.

Manual reorientation is (somewhat laboriously!) demonstrated in the first 10 minutes of following video:
 http://www.fil.ion.ucl.ac.uk/spm/course/video/#DemoPreproc

I hope that helps, best wishes,
Ged

Originally posted by Tanmay Nath:

Dear Ged,
Thanks for the reply.

I went through the solution you suggested me. I checked(using the check Reg button in SPM) all my nifti files and found that many were not registered to the template(in fact almost all the images).The template which i was looking at was the Avg152T1.nii in the Canonical directory of SPM.  Then using the Display button i manually altered the RIGHT(mm),FORWARD(mm),UP(mm),PITCH(rad),ROLL(rad),YAW(rad) values to register my nifti images to the T1 template.Once the image looks registered, i press the Reorient Images button to save the changes in the respective header file. Even after the manual registration process, i was not 100% accurate in registration, but I believe that my registration were 95% correct. There were some amount of mis registration which i was not manually  able to remove(as the template image is smoothed image and it is difficult to look at the exact anatomical locations). Also  the VBM tutorial states that there is some amount of freedom as far as manual registration is concerned using the Display button.

To start the VBM analysis in baby steps,i used only 4 datasets(Rather than 20 subjects) from each group(There are 2 groups) and segmented the GM,WM using the VBM,as taught in the VBM manual.To eliminate any complexities for now, i did not add any nuisance.

I estimated my 2 Sample T test model and set the contrast as 1 -1. To check the brain within the bounding box, i initially set the p value to 1 and unfortunately, i found that the result is same as the earlier results.i.e the brain is outside the glass brain box.

I have attached the pdf which illustrates the  data quality check of sample homogeneity using covariance and the final result.

Please provide some solution

Thanks
Tanmay

Hi Tanmay,

It's not as far outside the brain as I had initially understood. This is probably just the issue of low variance outside the brain. See e.g. Reimold's or my paper:
  http://www.ncbi.nlm.nih.gov/pubmed/16208...
  http://www.ncbi.nlm.nih.gov/pubmed/22037...
And try making a tighter mask (are you using absolute/relative threshold masking or an explicit mask, or just implicit?)

Which image are you overlaying your results onto? I would recommend using ImCalc to create a mean (or median) of the spatially normalised bias corrected images, as this is probably the most representative image for your subjects.

Best,
Ged

Dear Ged,
Thanks for the reply.

I was using Absolute threshold of thresholding as 0.1. I was also using the implicit mask and was not using the explicit mask.Also, i was using the Avg152T1.nii image in the canonical directory of the SPM as my image to overlay my result. I will look at the documentation of Imcalc() and try to make the average template of my spatially normalised bias corrected images which i got during the segmentation process. 
I will try your suggestions and also try to make a tighter mask.

Thanks
Tanmay

Hello,
 
So I am actually having a similar problem as you did at first. I am using the VBM8 toolbox in SPM8. I have manually reoriented all my images to the ACPC line, matching it to the DARTEL template (Template_1_IXI550_MNI.nii or something along those lines). When I want to run my Estimate and Write batch, it still gives me the error message spm_affreg.
 
I am out of options and can use some advice.
 
Thank you,
Rayana

Hi,
I also did a VBM analysis on GM images and I have a question If you are kindly enough to respond. Have you noticed that the resulted image from the GLM analysis (a 2-sample T test) is orientated different from the orientation of the preprocessed images? Do you know how come is that? In the first step, the preprocessing, the images are orientated the same as in the aquisition process, but in final after the GLM analyses, are re-orientated as in FSL. If I overlay the Avg152T1 template over the results the orientation is still different from the initial one, even though at the beginning the template was orientated the same as a the aquisition. The images were converted with spm so is all from the program.

Please help and reply!
Thank you!

Kindly regards,
Ana

Hi Ana,

Do you mean that the glass-brain MIP and the "overlays -> sections" views are displayed differently (e.g. with the axial image on its side with anatomical left at the top) compared to using the Display or Check Reg buttons on the processed images (where e.g. axial is upright, with anatomical left on screen left)?

These are actually just display conventions; the image orientation itself, in terms of the mapping from voxel indices to mm coordinates, should be consistent. E.g. if you use Check Reg to compare one of the images from the SPM results directory (e.g. the mask image) to one of the processed images that go into the statistical model, you should find they are aligned.

As to why the MIP/overlays convention differs from the Display/CheckReg one, that is a good question... It's possible that SPM12 will offer the user a choice of either of these conventions (perhaps others too) and then use the chosen one consistently, but I can't promise this...

Best wishes,
Ged

Hi Ged,
So the MIP/overlays convention are displayed in "radiological" convention? And the Check Red displays the images in "neurological" convention? I'm asking because I have to compare the VBM result from FSL (which are displayed in "radiological" convention) and SPM8, and I'm a little confuse about the orientation of the final results in SPM. 
Thank you very much for you help!

Best regards,
Ana

Hi Ana,

No, SPM does consistently use the neurological convention: in the coronal views, anatomical-left is on screen-left, for both CheckReg and MIP, and even in the "sideways" axial in the MIP, the anatomical left is to the left with respect to the direction the eyes are facing.

You can also check where you are by noting that the sign of the x-coordinate is positive for right hemisphere and negative for left-hemisphere (again, consistent between CheckReg, MIP, results table, etc.), assuming properly converted NIfTI images. This will also hold in FSLview and other software that correctly handles NIfTI, regardless of its display convention.

Best wishes,
Ged

Hi Ged,
Thank-you very much for your help! It was very helpful.

Best regards,
Ana