The IIT Human Brain Atlas (v.3) contains a sharp, high-resolution DTI brain template in ICBM-152 space, which can be used as a reference for spatial normalization of DTI data from multiple subjects. Based on our analysis, the recommended registration tool for use with our atlas is DTI-TK. You will find simple step-by-step instructions on how to use DTI-TK to register DTI data to our atlas at the "Docs" link. For additional questions please feel free to use this forum.

The files of the 256x256x256 version of the atlas were updated on August 9th 2013 to maximize compatibility with DTI-TK.

In the "HOW TO register DTI data to the IIT Human Brain Atlas (v.3) using DTI-TK" documentation you have shown how to run the rigid, affine and nonlinear steps, but not the step that combines the affine and nonlinear warps together (i.e. dti_warp_to_template_group step). Is this unnecessary?

You are correct. The previous version of the HOW TO document did not include the step for combining the affine and nonlinear transformations. This step allows the user to transform an individual DTI dataset to the space of the IIT Human Brain Atlas (v.3) using one interpolation instead of two (i.e. one for the affine registration and one for the nonlinear registration). This is an important step, and is now included in the updated HOW TO document. Thanks for your comment!

I have two questions regarding the IIT3 template, dti-tk and and ICBM space:
1) You mentioned, that one can use the IIT3mean_tensor_256 image for directly registering individual tensor files using dti-tk. Now, I am wondering if I could use the non_256 file in a similar way. If not, what are the precise differences
2) In the file description, it is mentioned, that the IIT3 template is in ICBM156 space. Unfortunately, it is not in the ICBM_81 space for white matter tracts. There just one voxel per dimension missing in order to fit the non_256 image to DTI_ICBM81 space. Is there a way to have an atlas reference for white matter tracts in the IIT3 space.

Greetings and a thanks for this very nice templates, Marcus Heldmann

Hi Marcus,

1) The main reason we constructed the 256 version of the atlas was to make it compatible with DTI-TK, since DTI-TK requires 128x128 or 256x256 images. The 256 and non_256 versions of the IIT v.3 atlas have exactly the same contents. The only difference is in the matrix size. 
2) The non_256 version of the IIT v.3 atlas has 182x218x182 dimensions and is in ICBM152 space. If you would like to use the ICBM-81 white matter labels, you can actually find them in the 182x218x182 matrix size in FSL (look in the fsl/data/atlases/JHU directory). This should fix the issue of the single voxel difference you mentioned.
Also, please note that, very soon, we will be releasing white matter tract labels customized for the IIT v.3 atlas (in 256 as well as in non_256 space). Check back for those soon.

Regards,
Konstantinos

Hi,

I'm trying to set up a DTI-TK protocol where I register a Groupwise template to the ICBM- 152 standard space.
My groupwise template created by registering within subject templates which were created for each subject from each of the subjects two time points.
I've found that after the deformable alignment of the Group wise to the ICBM- 152 standard space the new GW to ICBM template looks very odd with no real definition of the brain structures.
It also leads to weird subject images when the are normalized to this template.
Is this normal or is there a way to correct for this?
Thanks
Lauren

Hi Lauren,

Could you send us the group wise template that you are trying to register to the IIT atlas?
You can attach a file to a message to this forum.

Thank you,
Konstantinos

HI Konstantinos,

Thank you for your response.
Please find attached my groupwise template (GW_template) before registering to the IIT atlas.

This was created from 10 within subject templates where each subject had 2 time points.
I followed the pipeline from the HOW-TO-register-DTI-data-to-IIT-atlas-using-DTITK.pdf.

I can attach the resulting image in another message.

Thanks,
Lauren

Please find attached the resulting image from registering my group wise temp to the IIT atlas.

Best,
Lauren

Sorry it wouldn't attach to my previous message.

Hi Lauren,

Is the groupwise template you generated and trying to register to the IIT atlas a tensor template? I assumed it is because you are using DTI-TK for registration, but then I noticed that the file you attached is an anatomical T2-weighted image... Please clarify.
We'll figure this out :)

Regards,
Konstantinos

HI Konstantinos,

All the steps I've taken before creating the groupwise are as follows:
-Took raw DWI for each subject time point and converted to 4D FSL NIfTI format.
-Corrected for EDDY currents using EDDY.
-Ran DTIFIT for tensor reconstruction.
-Converted DTI into DTITK format.
-Created within-subject template by bootstrapping both time points to the IXI-aging template then using dti_affine_population and dti_diffeomorphic_population.
-Created Group wise template by bootstrapping the 10 within-subject templates to the IXI-aging template then using dti_affine_population and dti_diffeomorphic_population.

I think the format of the image I uploaded only shows one component of the tensor so it looks similar to a T2 weighted image and therefore probably isn't the best way to visualise alignment. I plan to compare the FA maps for each template and registration.
I've attached the FA map for the group wise template before registering to the IIT atlas.

Regards,
Lauren

Hi Lauren,

Your steps look fine. So the only question is what parameters you are using in DTI-TK to register to the IIT atlas. It would also help to see the FA map after registration to IIT space.

Regards,
Konstantinos

Hi Konstantinos,

dti_rigid_reg /home/lauren/Imaging/IITmean_tensor_mask_256_templates/IITmean_tensor_256.nii.gz GW_template.nii.gz EDS 4 4 4 0.01

dti_affine_reg /home/lauren/Imaging/IITmean_tensor_mask_256_templates/IITmean_tensor_256.nii.gz GW_template.nii.gz EDS 4 4 4 0.01 1

dti_diffeomorphic_reg /home/lauren/Imaging/IITmean_tensor_mask_256_templates/IITmean_tensor_256.nii.gz GW_template_aff.nii.gz /home/lauren/Imaging/IITmean_tensor_mask_256_templates/IITmean_tensor_mask_256.nii.gz 1 5 0.002

These are the commands I've used. I know that the IIT atlas has significantly bigger dimensions compared to my GW template. Would that be a problem?
I've attached the FA map for the GW template after being registered to the IIT space.

Thanks,
Lauren

Hi Lauren,

I don't see any problem with your steps. And the result of registration, although very wrong in some regions, is generally "good". I think the problem lies in the fact that your group template is relatively blurry and you are trying to do precise spatial matching of a blurry image to a sharper image.
What I would suggest you do is this: instead of generating group templates and registering the group templates to the IIT atlas, skip the group template step and register single subject data directly to the IIT atlas. I hope this works for your project.

Regards,
Konstantinos

Thanks Konstantinos,

I will try this and let you know how it goes.

Best,
Lauren

Hi Konstantinos,

I've tried registering my subject templates directly to the IIT atlas and still had the same problem.

Please find attached a snap shot of the IIT atlas (left) and a single subject's template after deformable alignment to the IIT atlas (right).

I've also been trying to alter the regularisation parameters for the diffeomorphic step which seems to have improved it in that it looks more normal. Please find attached a screen shot of the IIT atlas (left) and the group wise deformably aligned using regularisation parameters 200 times the weaker ones provided.

There are still regions not well aligned with the IIT atlas and the image is quite blurry. We think this is because there is quite a lot of atrophy in the brains, particularly in the caudate as the are huntington's disease patients, and the ventricles are enlarged.

I'd be interested in your opinion on this template and whether you know of ways to factor in brain atrophy when registering into the IIT standard space?

Kind Regards,
Lauren

Hi Lauren,

It looks like you are getting closer. Few things:
1) In this last example, did you first generate the group template and then register that to the IIT atlas? Or did you register individual subjects to the atlas and then took their mean? I am trying to understand where the blurriness you mentioned comes from.
2) Although blurry, your group wise template looks like it matches the IIT space relatively well.
3) Have you tried other values for the regularization parameters?
4) Atrophy certainly reduces normalization accuracy. The magnitude of this effect depends on the amount and pattern of atrophy. I suggest you experiment with regularization a bit to find a good balance that works for your project.

Regards,
Konstantinos

HI Konstantinos,

1) That was just one of the subjects registered to the IIT atlas so I could take a mean of them.
3) I've tried a range of regularisation parameters (200x being the highest) I wasn't sure if there was much point increasing further as I couldn't tell if it was improving the template any more.
4) so is the fact that atrophy affects registration just accepted or is there a way to mask it to improve alignment?

Regards,
Lauren

Hi Lauren,

1) What was the voxel-size of the raw data?
4) Well this is a tough question. It all depends on how precise registration you need. In general, we want to maximize registration accuracy to maximize the ability to see small differences across people. If however you are looking for big effects, then you could say that you accept some mis-registration due to atrophy. In which of the two categories does your project belong to? About masking out the atrophic brain regions to improve alignment, I don't think DTI-TK does anything like that.

Konstantinos

Hi Konstantinos,

The volume info for the raw data is:
size: 96x96x60, voxel size: 2.29167x2.29167x2.3, origin: [-48, 41.2169, 27.6666]

For the statistics we want to do to assess group change we don't particularly need the registration to be as precise as all data points are from patients. We need a precise alignment to a standard space though in order to do further analysis such as ROI analysis.

Many thanks,
Lauren

Dear Konstantinos and others:

Like the others who have commented before myself, I am trying to use the IIT template in DTI-TK. I have read the guide on "How to spatially normalize individual DTI data to the IIT HUman Brain Atlas Using DTI-TK". I assume this is more for those who want to do the registration to the IIT template and not in the case whereby one has generated the population-specific atlas and is trying to register that to the IIT template. I have a few questions:

1. Referring to the guide provided, I assume one has to already perform the bootstrap before the suggested rigid alignment steps. Prior to this, if I preprocessed my data using FSL, I would have also used the script fsl_to_dtitk (http://dti-tk.sourceforge.net/pmwiki/pmw...). I assume these steps are compatible with the later steps in the analysis as suggested in the guide?

2. I am also trying to integrate TBSS with DTI-TK, to perform group wise analyses. In this case, after I have completed Step 7 (combining affine and diffeomorphic transformation), how would I then integrate the output with TBSS? In DTI-TK, there is a integration pipeline (http://dti-tk.sourceforge.net/pmwiki/pmw...). Are these steps still relevant, or is there a way to smoothly proceed on to do the TBSS analysis?

Thanks very much.
Rowena

Hi Rowena,

You are right, our HOW-TO document describing how to spatially normalize individual DTI data to the IIT Human Brain Atlas using DTI-TK describes how to register directly to the atlas and does not cover the case where a population template is first constructed and then registered to the IIT atlas. Before answering your specific questions, let me pause here to comment on the two approaches above. I tend to prefer the approach covered by our HOW-TO document. The reasons are the following. A study-specific template is theoretically most representative of the data under study and may lead to the best spatial normalization accuracy. Based on that, some users first register to a study-specific template and then register the study-specific template to the IIT atlas. However, the accuracy of spatial normalization to a study-specific template is high only when the study-specific template is carefully constructed and of high quality. Poorly constructed study-specific templates (e.g. when small number of subjects are used, or when suboptimal template-building procedures are used) are actually not representative of the individual data under study and lead to low spatial normalization accuracy. On the other hand directly registering to a high quality standardized template like the one in the IIT atlas, is shown to perform very similar to registering to a high-quality study-specific template and then registering that to the IIT atlas, and is also consistent and fast (no need to make a study-specific template). In brief, both approaches are appropriate, but if you decide to make a study-specific template first, you need to make sure that the study-specific template you are constructing is of high quality. If you cannot achieve that for whatever reason, or you don't want to worry about it, or you want to be able to compare results across studies, then register directly to the IIT atlas as shown in our HOW-TO document.
Now, in terms of your specific questions:
1) Yes, you should be able to combine the steps you mentioned with the later steps of our HOW-TO document.
2) Yes, you can perform TBSS after you have normalized to the IIT atlas. Just read the HOW-TO use the IIT atlas in TBSS document. It walks you through the steps you'll need to follow. And once you are done with your TBSS analysis in IIT space following the steps in our HOW-TO document, you can use our white matter atlas resources to help you understand your findings. For example, if you find some significant effects somewhere in white matter, you can use the "regionstat" program to give you the list of the most probable connections going through the region with the significant effects (See HOW-TO-extract-pairs-of-GM-labels-with-connections-through-WM-ROI). It does not require any tractography and is very fast. It works by interrogating our 4D white matter atlas to extract the most probable connections through your white matter ROI.

Let me know if you have more questions.
Regards,
Konstantinos