[Mristudio-users] DTI

[susumu mori]

1) Conventional method (all steps will be performed in the native space)

>Load DTI data to DtiStudio, do tractography based on a manual protocol.
>Use "Statistics" function to quantify pixel values along the reconstructed
tract. If you load an FA map, you'd get one averaged FA value of the entire
tract.
>Use "Profile" function to obtain the report of averaged FA values at each
slice (axial, sagittal, coronal). This method works only when your tract is
more or less straight. For example, if you are measuring the corticospinal
tract (CST), the axial slice-by-slice report is most informative.
>Support the CST spans axial slice #5 (pons) to slice #50 (motor cortex),
you have 46 values. Save this information in the first column of a Excel
sheet. This is the result from Patient A.
>Repeat the process for Patient B. You may find the tract runs slice #8 -
#51. You have 44 values, which are copied to the second column.
>Now you have an issue; the slice locations are different and the tract
length are different. This is a typical "registration" problem.
>Manually identify landmark slices. For example, for Patient A, slice#5 is
pons, #12 is midbrain, #30 is internal capsule. For Patient B, the
corresponding slides are 8-13-27. Then you can align these slice levels
between A and B inside the Excel sheet. Slice locations are adjusted but
you still have tract length problem; pons-midbrain of A is 7 slices and B
is 5 slices. You can do interpolation calculation inside Excel to extend
the 5 slice values in B to 7 slices.

This is what we did in Stieltjes paper. Obviously a lot of work. The
problem is cross-subject registration. This registration issue doesn't
exist if you average the entire FA values of a tract. As soon as you try to
recover location information, the registration issue arises.

2) Tractography after registration

Things get much easier if you do brain-to-brain registration first and then
do tractography. For this purpose, you use linear registration that adjust
brain locations, rotations, and sizes (12-parameter or 9-parameter
(recommended) affine transformation) . After this registration procedure,
the slice locations and tract lengths are equal (almost) across all
subject. So you won't need the registration procedure described above.

You need to use DiffeoMap for this registration. You need to register two
tensor fields. After two tensor fields are aligned, you calculate FA and
vectors for tractography. The procedures are described in "Getting Started"
in www.mristudio.org.

3) Some more thoughts

As mentioned above, you don't need brain registration if you use only the
overall FA value from a tract. As soon as you try to retrieve more location
information, you have to worry about location registration. If you want
even finer location information beyond slice-by-slice report, or if your
tract is not linear (U-shape) and slice-by-slice report is not enough, you
need to further divide the tract into smaller units. The smallest unit is a
voxel. The ultimate goal of image registration is voxel-to-voxel
registration. For any arbitrary voxel you pick in one brain, you can find a
corresponding voxel in the other. If your voxel-to-voxel registration is
perfect, you don't need tractography; you can compare FA values at each
pixel and find abnormalities; why do you bother spending time for
tractography if you can find a cluster of voxels with abnormal FA values
without tractography?

There are two reasons. First you can't always assume the brain registration
is accurate after the automated brain-to-brain registration. You can argue
tractography can identify corresponding pixels between two brains more
accurately. *Namely, tractography is a cross-subject registration tool with
which you can extract corresponding group of voxels from multiple
subjects.*Another point is that voxel-to-voxel FA comparison is too
noisy and needs
voxel-averaging to enhance SNR. Usually the voxel-based anlaysis applies
isotropic voxel-grouping filter, like 5x5x5 filters. Instead of applying
the blind isotropic filtering, tractography groups only voxels that belong
to the same tract. *Namely, tractography is structure-specific smart filter
to enhance SNR.* Therefore, the automated voxel-based and
tractography-based approaches could have different accuracy and sensitivity.

On the other hand, tractgraphy is also a noisy tool; the number of voxels
clustered by this method is highly variable among subjects. So, there is no
perfect solution. If you used a probabilistic tract information stored in
our atlas, you are not affected by the tractography variability because you
apply the same probabilistic map to all subjects. However, you invite the
inaccuracy of the brain-to-brain registration. In other words, you throw
away the "registration function" of the tractography I described above, but
you still retain its "filtering function".

I hope these arguments are helpful.

Susumu



On Sun, Feb 24, 2013 at 10:57 AM, Shaimaa Abdelsattar <shaimaa96 at hotmail.com
> wrote:

> Thank you very much for your detailed explanation
> Primarily,  I would like to master the first approach ( manual), how can I
> load multiple patients  for registration and then slice by slice processing
> to do group analysis, is  AIR program included within DTI studio, or is a
> separate program. Please can you tell me step by step like 1, 2, .....
>
> Thank you very much
>
> On Feb 23, 2013, at 5:07 PM, "susumu mori" <susumu at mri.jhu.edu> wrote:
>
>
> I would like to ask if  1. I can measure FA value, ADC or MD in a specific
>> ROI or along a specified tract in order to have data in an objective way?
>>
>>
> There are multiple approaches;
>
> 1) Manual approach in the native space:
>
> You need to establish a protocol for manual placement of ROIs. After
> manual fiber reconstruction, as Dorian pointed out, DtiStudio can report
> averaged pixel intensities (e.g. FA, MD, etc) of the all pixels that
> contain the fibers or report slice-by-slice numbers.
>
> You can find some protocols in this paper;
> Wakana S, Caprihan A, Panzenboeck MM, Fallon JH, Perry M, Gollub RL, Hua
> K, Zhang J, Jiang H, Dubey P, Blitz A, van Zijl P, Mori S. Reproducibility
> of quantitative tractography methods applied to cerebral white matter.
> Neuroimage 2007;36(3):630-644, PMC2350213
>
> When you use the slice-by-slice reports, instead of one value averaged
> over an entire tract, you have to "register" data from different subjects
> because, say, an averaged FA of a tract at "axial slice#10" of one subject
> may not be the same anatomical slice of "axial slice#10" of the other
> subject. An example of this operation can be found in this paper;
>
> Stieltjes B, Kaufmann WE, van Zijl PC, Fredericksen K, Pearlson GD,
> Solaiyappan M, Mori S. Diffusion tensor imaging and axonal tracking in the
> human brainstem. Neuroimage 2001;14(3):723-735
>
> 2) Automated approach in the native or MNI space:
>
> These papers describe automated ROI placements;
>
> Zhang W, Olivi A, Hertig SJ, van Zijl P, Mori S. Automated fiber tracking
> of human brain white matter using diffusion tensor imaging. Neuroimage
> 2008;42(2):771-777, PMC2585359
>
> Zhang Y, Zhang J, Oishi K, Faria AV, Jiang H, Li X, Akhter K, Rosa-Neto P,
> Pike GB, Evans A, Toga AW, Woods R, Mazziotta JC, Miller MI, van Zijl PC,
> Mori S. Atlas-guided tract reconstruction for automated and comprehensive
> examination of the white matter anatomy. Neuroimage 2010;52(4):1289-1301,
> PMC2910162
>
> The idea is, you can define ROIs (or use our pre-defined brain
> parcellation maps) once in an MNI atlas and warp these ROIs to each subject
> for automated tracking. Yajing Zhang has many pre-defined ROI sets in our
> brain parcellation maps. I believe you can download them from our websites.
>
> 3) Probabilistic approach:
>
> One issue of #1 and #2 approaches is that the streamline generation has a
> large amount of variability. Another common issue is, if there are lesions
> with low FA, tractography is influenced by that. Our atlases have
> probabilistic tract locations. For example, we have reconstructed the
> corticospinal tracts in normal subjects and the results are registered into
> the MNI space, creating probabilistic map of the corticospinal tract in the
> MNI space. If you normalize your patient brains to the MNI space, you can
> superimpose these probabilistic maps on the patient brains and quantify
> averaged pixel intensities.
>
> There are pre-defined probabilistic maps of many tracts in RoiEditor,
> which can be applied for automated pixel intensity calculation of various
> white matter tracts. Downside of this approach is, it assumes that tract
> locations of your patient population are not significantly altered and
> therefore after normalization to the MNI space, the probabilistic maps can
> accurately define the properties of each tract of interest.
>
> These papers describe this approach;
>
> Hua K, Zhang J, Wakana S, Jiang H, Li X, Reich DS, Calabresi PA, Pekar JJ,
> van Zijl PC, Mori S. Tract probability maps in stereotaxic spaces: analyses
> of white matter anatomy and tract-specific quantification. Neuroimage
> 2008;39(1):336-347, PMC2724595
>
> Zhang Y, Zhang J, Oishi K, Faria AV, Jiang H, Li X, Akhter K, Rosa-Neto P,
> Pike GB, Evans A, Toga AW, Woods R, Mazziotta JC, Miller MI, van Zijl PC,
> Mori S. Atlas-guided tract reconstruction for automated and comprehensive
> examination of the white matter anatomy. Neuroimage 2010;52(4):1289-1301,
> PMC2910162
>
>
>> 2. Can I load an anatomical sequence over DTI data eg. Color map or fA
>> map etc...?
>>
>>
> Of course, you first have to register your anatomical image (like T1) to
> your DTI (or register DTI to T1). Then you can load both T1 and DTI to
> RoiEditor. We don't have image-to-image overlay functions with transparency
> control. You have to define "object (ROI)" using one of the image (like
> tract locations from DTI or thresholded high FA regions) and then
> superimpose the object (ROI) to the other image (like T1). In RoiEditor,
> the tract streamline information have to be converted to 3D image format
> (like a masking file in which pixels that contain a fiber are "1" and all
> other are "0", or each pixel contains a probabilistic value). If you load
> the co-registered T1 to DtiStudio, you can superimpose streamline
> information on T1. For T1-DTI co-registration, you can use our DiffeoMap or
> other software like FSL and SPM.
>
> Please be careful because all MriStudio family programs use Radiology
> convention (right is left) while many other programs follow Neurology
> convention (right is right).
>
> susumu
>
>
>> Thanks
>> _______________________________________________
>> mristudio-users mailing list
>> mristudio-users at mristudio.org
>> http://lists.mristudio.org/mailman/listinfo/
>> Unsubscribe, send a blank email to:
>> mristudio-users-unsubscribe at mristudio.org
>>
>
> _______________________________________________
> mristudio-users mailing list
> mristudio-users at mristudio.org
> http://lists.mristudio.org/mailman/listinfo/
> Unsubscribe, send a blank email to:
> mristudio-users-unsubscribe at mristudio.org
>
>
> _______________________________________________
> mristudio-users mailing list
> mristudio-users at mristudio.org
> http://lists.mristudio.org/mailman/listinfo/
> Unsubscribe, send a blank email to:
> mristudio-users-unsubscribe at mristudio.org
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.mristudio.org/pipermail/mristudio-users/attachments/20130224/85b9b4be/attachment.html