[Mristudio-users] Head Motion issue

Tue Mar 28 09:40:45 EDT 2017

The short answer is, unfortunately, "difficult" and "you need to rely on
your eyeballs", but there are some metrics you can use as a not perfect but
somewhat reliable tool. Here is why;

1: As simple as linear co-registration of 3D volumes by AIR or eddy current
corrections, in the end of the day, you use your eyeballs to see if it
works well or not. You may think, "but there must be a way to monitor the
goodness of registration or correction quantitatively." Yes, such metric
exists and all registration programs have their own functions to monitor
the goodness of registration. Registration algorithms try to maximize their
goodness functions while moving and deforming the target images. The
goodness function is as simple as subtraction of target and template images
and minimizes the subtraction errors. However, we all know that
registration algorithms are not perfect and in the end of the day, we use
our eyeballs to see if it is working. If we have a good metric that can
substitute this eye inspection, or as good as our eyes, we would have used
such a function in the registration algorithms to begin with.

2: Having said that, there are several objective metrics you can use to
infer the goodness of your data and registration results. First, you can
monitor how many voxels you needed to move 3D images to register them.
Second, you can monitor the results of the goodness functions. Third, if
your software has automated pixel rejection, you can monitor how many
pixels were rejected.

These metrics are not perfect. For example, there are papers describing how
many voxels AIR needed to move images for registration and use the metrics
(such as translational and rotational motions) to report the patient
motions. This is a good attempt but not perfect because it assumes that AIR
(or any registration software) works perfectly, extracting accurate
information about the motion. However, in reality, if the subject moved too
much, AIR may not work and sometimes it gives up, reporting 0 motion. If
the 3D images are corrupted, there is no point of doing cross-3D
registration to begin with. If you know these limitations of such metrics,
they should give you a good guidance about the quality of the data.

In DtiStudio and MRICloud, there is a very comprehensive reports about the
metrics mentioned above. As a matter of fact, it may be too comprehensive
with a flood of numbers. We tried to create a wrapper software that
summarize and visualize the reports, but haven't completed the work.

If anybody is interested in pursuing this topic based on the quantitative
reporting of DtiStudio/MRICloud, please contact Yue and ask for the wrapper
software.

I assume that other well-regarded software such as RESTORE also has similar
reporting, but I'm not sure.

On Tue, Mar 28, 2017 at 9:06 AM, Shaimaa Abdelsattar <shaimaa96 at hotmail.com>
wrote:

> Thank you very much for your reply, so movement is inevitable problem that
> can't be completely solved by AIR.
> Is there a way to know objectively the degree of motion in each dti study,
> eg. Motion index calculation (may be implemented with AIR), in order to
> perform group comparison, taking in concideration this motion index
>
> Shaimaa
>
> Sent from my ASUS
>
> -------- Original Message --------
> From:Susumu Mori
> Sent:Mon, 27 Mar 2017 16:48:04 +0200
> To:"DTI Studio, ROI Editor, DiffeoMap Questions/Support"
> Subject:Re: [Mristudio-users] Head Motion issue
>
> Yes, motion always degrades MR images. It's not just DTI, but DTI is one
> of the most susceptible.If data degrade in the time domain, there is only
> so much you can do in postprocessing, which is always "after the fact"
> remedy.
>
> 1> Any MR image degrades if the subject moves more than 1 pixel during the
> scan (I mean "DURING" the scan, not in between the scan). This is a
> macroscopic motion in the order of 1-2mm or more. It is not a DTI specific
> issue. During one phase encoding to the next phase encoding, if the subject
> moved more than 1 voxel, the fourier transform won't work. When we talk
> about motion-correction for DTI, this type of motion correction is not
> included. This effect is more severe for 3D T1 scans which usually take 2-5
> min for one image, during which the subject could move. For DTI, we
> consider this effect is much less because we usually use a single-shot EPI,
> which takes only less than a second. To correct this motion, we need a
> real-time motion monitoring, which usually involves external motion tracker
> or extra imaging schemes, but we assume this problem is negligible for DTI.
>
> 2> There are true-3D MRI and multi-slice 3D MRI approaches. T1 is usually
> the former and T2 and DTI are usually the latter. The former suffers more
> of the #1 problem. The latter has it's own problem; between-slice
> misalignment within one 3D volume. As I mentioned in #1, one 2D DWI is
> taken within less than a second. However, to obtain a whole 3D volume with
> multi-slice acquisitions, I believe it takes 10-15 s. Any macroscopic
> motion (>1-2mm) during this time period would cause among-2D-slice
> misalignment within a 3D volume. In DTI motion correction, this type of
> misalignment is NOT corrected while this problem does happen. To correct
> this problem, a real-time motion tracker and real-time FOV alignment DURING
> the data acquisition is needed. And even with such a method, it may be
> difficult to eliminate the problem completely. There are some technique
> already presented but the important point is, this problem cannot be
> corrected after the scan, just like the problem #1.
>
> 3> Motion correction between 3D volumes. This is a problem common to all
> quantitative MRIs that require calculation using multiple 3D MRI data, like
> DTI, fMRI, quantitative T2 map, etc. What is specific to this group is the
> mis-alignment among 3D data. Here we assume that each 3D data is free from
> #1 and #2 problems and therefore each data are not degraded. This is one of
> the motion corrections we routinely do for DTI and all other quantitative
> MRI. This method is well established and we can expect the errors are
> minimized. Of course, this correction can introduce resolution degradation
> but we usually ignore it.
>
> 4> Image distortion correction. There are B0 susceptibility and
> eddy-current distortion, while the former is common to all EPI-based
> acquisition (DTI, fMRI, etc) and the latter is specific to DWI. These
> corrections are approximation at best but they do substantially reduce the
> error. So, it is a good idea to do it. The degradation by eddy current can
> especially be bad in old days and the correction schemes helped us a lot.
> B0 problem does not degrade DTI data because all DWIs and b0 are deformed
> in the same way, but it can cause some interpretation errors of the data.
> Morphologic results (like volumes) could be inaccurate without correction,
> although, we can't expect perfect correction by postprocessing regardless
> of the method we use because sometimes the degradation occurs in the time
> domain (can't retrieve in the spatial domain anymore).
>
> 5> motion-caused phase error and intensity drop. This is DWI specific.
> Microscopic motion (like brain pulsation) less than one pixel (<1-2mm)
> causes signal drop IF it happens during a pair of diffusion-weighting
> gradients. Because our brains are constantly moving, this is an significant
> issue. The higher the b-value, more problem we have. Thus, HARDI-type scans
> with higher b-values suffer more. Once this happens, the only thing we can
> do is to discard the image or affected pixels. Many DTI postprocessing
> software has pixel-rejection programs. This is again, an after-the-fact
> remedy, not a fundamental solution. Even if it works perfectly, we suffer
> from loss of data points. For example, if we acquire 30 DWIs and 3 of them
> are badly affected and removed, we loose SNR. For high-b-value scans with
> poor SNR, I'm not sure how you can reliably detect affected voxels or
> images. I believe there are some papers discussing about it.
>
> So, overall, you can see that the motion problems for DTI (and MRI) are
> not solved but many software tries to reduce it.
>
> As some of the issues are inherent to MRI, which is one of the slowest
> imaging techniques, we can't dismiss MRI entirely due to its sensitivity to
> motions. We always need to be careful when we interpret the data,
> especially when you suspect that the patient group may have more
> motion-related problems. Otherwise, we would end up in using MRI and
> sophisticated analysis only to say, "patients moved more."
>
>
>
> On Sun, Mar 26, 2017 at 2:05 PM, Shaimaa Abdelsattar <
> shaimaa96 at hotmail.com> wrote:
>
>> Hello,
>> There are papers in the literature discussing the effect of head motion
>> on DTI analysis as it can induce spurious differences among groups. I would
>> like to ask if this motion is still a concern even after eddy current
>> correction and registration into b0 image using AIR tool in DTI studio for
>> tractography.
>> Thank you in advance
>> Dr.Shaimaa Mohammad, MD
>>
>> Sent from my iPad
>> _______________________________________________
>> 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 mr
>> istudio.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@
> mristudio.org
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mristudio.org/pipermail/mristudio-users/attachments/20170328/8f27b6f7/attachment.html>