issue with eddy

[Frank Garcea]

Hi Frank,

I'm running the Hou version (July 25, 2026) on my mac and have also tested and replicated the issue with the ubuntu 22.04 version. I've noticed two problems: 

(1) After reconstruction from dicoms to SRC, the mask is consistently missing frontal cortex. This is not a huge issue since I can adjust the thresholding, erode, dilate, etc., to get the whole brain. But this is a new problem, as I never observed this behavior using older version of DSI_Studio (with the same subject dataset).

(2) Topup/eddy correct takes hours. Like 12+ hours. I routinely checked with the 'top' call to see if eddy_openmp was running. It was but the correction never completed. DSI_Studio gave me warning that it could not find the corrected image because of insufficient memory. I've never encountered this issue before and had no problem using topup/eddy correction with the same 259-direction DSI SRC and PA B0 image. It's only after upgrading to the Hou version (July 25, 2026) version that I encounter the issue.

Please let me know if there is anything else I can do to help troubleshoot what might be going on here.

Thanks,
FG

[Frank Yeh]

Is it possible to send me a sample SRC file to test?

--
You received this message because you are subscribed to the Google Groups "DSI Studio" group.
To unsubscribe from this group and stop receiving emails from it, send an email to dsi-studio+...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/dsi-studio/9a4fd247-7ffb-41a0-bc9d-61781fdb5343n%40googlegroups.com.

[Frank Garcea]

That would be great. How would you like me to send it? 

FG

[Frank Yeh]

Hi Frank,

There is a private upload link provided on the DSI Studio website that
you can use to share the data for troubleshooting.

Best regards,
Frank

[Frank Yeh]

I would need bval and bvec

On Thu, Jul 30, 2026 at 12:23 PM Frank Garcea <frank....@gmail.com> wrote:
>
> Hey Frank,
>
> Thanks for that update.
>
> I uploaded a sample DSI dataset and a B0 image (both are nii.gz files).
>
> Best,
> FG

> --
> You received this message because you are subscribed to the Google Groups "DSI Studio" group.
> To unsubscribe from this group and stop receiving emails from it, send an email to dsi-studio+...@googlegroups.com.

> To view this discussion visit https://groups.google.com/d/msgid/dsi-studio/5abc0102-184a-4b76-952c-9fbdaf0ada19n%40googlegroups.com.

[Frank Garcea]

Hey Frank,

I just uploaded the SRC, bval, and bvec file.

Thanks,

FG

[Frank Yeh]

Hi Frank,

The mask issue is caused by signal inhomogeneity. Running [Correction]
-> [Bias Field Correction] resolves it immediately and updates the
mask, which I have verified through testing. Older versions may have
applied this correction universally, but this is not always ideal as
some users prefer to avoid it.

Once the mask issue was resolved, I tested TOPUP using the b0 image
and it completed successfully (TOPUP took 3 minutes, and saving the
NIfTI took 1 minute, though I am using a GPU so it is much faster).
Please note that Eddy is generally not recommended for DSI-style
multi-shell data and may not work well.

I recommend opening the console window first to monitor the progress;
if it gets stuck, you will be able to see exactly where the issue
occurs.

Best regards,
Frank

> To view this discussion visit https://groups.google.com/d/msgid/dsi-studio/38645aec-96aa-4ab6-a9a2-b745556d6c24n%40googlegroups.com.

[Frank Garcea]

Hey Frank,

Thanks for your help--I followed your directions and it worked without issue!

We use eddy and topup when pre-processing single shell DWI data.

Is there a good resource or current standards for pre-processing DSI data?

Or is there consensus that topup is sufficient before reconstruction with GQI?

I suppose QCing the data after topup is wise to check that data quality is high.

Best,

FG

[Frank Yeh]

Hi Frank,

It is difficult to define what qualifies as "sufficient." For DSI
data, images are typically acquired using a bipolar pulse, which
mostly cancels out eddy currents. The remaining factor is subject
motion. If subject motion is significant, the best practice is to
re-collect the data.

Best regards,
Frank