[Mristudio-users] Fw: excluding bad images

[susumu mori]

Getting corrupted images degrade your results no matter what you do.
If you include them, your FA values deviate from real value.
If you remove it, you loose SNR, which is known to increase FA.

You have to judge which is worse. For example, if you acquire 30 DWIs (30
orientation, 15 orientation x 2, 10 orientation x 3, it doesn't matter much)
and remove 1 DWI, SNR loss could be minimum. So, you may not have to worry
much. If you acquire only 12 DWIs and remove one, the SNR impact may be
visible, but if you leave the bad image in the calculation, the impact could
be even bigger.

I have an impression that removing the bad images (sacrifice SNR) is usually
the better choice, but again, once you get corrupted images, you always
loose some.

Another thing you may want to try is "pixel rejection" instead of "slice
rejection". If the image corruption is very localized in a small region, the
pixel rejection approach can minimize the impact only in that area. There is
a detailed description at
https://www.mristudio.org/wiki/changelog/dtistudio-3.0.2 describing
different types of artifacts and correction methods.

On Thu, Sep 23, 2010 at 7:34 AM, Christopher Latsis <latsisc at yahoo.com>wrote:

> Dr Mori I sent my question below to the mail list but I guess it did not go
> through. I hope you would not mind this direct email to you for this one
> time.
> Thanks in advance for your time and kind attention. Chris.
>
>
>
>
> Hi All,
> I am trying to understand Dtistudio's quality control window. I understand
> that
> if an image corruption is found, I can simply click on the bad image and
> the
> image is not used for subsequent calculation. How would this effect the FA
> ?
> I am planning on doing a VBA type group analysis. Some datasets have signal
> loss
> at certain slices/gradients. Can I exclude these bad images, calculate FA
> for
> these certain subjects and compare them to those with good quality data
> with no
> signal loss? Any good references you could recommend on this
> matter? Thanks very much. Chris.
>
>
>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.mristudio.org/pipermail/mristudio-users/attachments/20100923/a30e296f/attachment.html