differential RDI banding?

[Hannah Lindsey]

Hi! 

I'm running differential tractography and have been overlaying the results on percent decrease slices I calculate using fslmaths after exporting the registered ses01_ses-02 metrics:

#Absolute decrease

fslmaths ses-01_rdi.nii.gz -sub ses-01_ses-02_rdi.nii.gz dec_rdi.nii.gz

#Percent decrease

fslmaths dec_rdi.nii.gz -div ses-01_rdi.nii.gz pdec_rdi.nii.gz

I've checked to make sure the calculated percent change datasets give accurate results by comparing them with manually calculating percent decrease from the ses-01 and ses-02 stats exported from DSI Studio across the whole brain and various individual tracts. 

I've noticed some banding of percent and absolute change in RDI in the affected regions. A multishell acquisition was used, and we're expecting to see decreased RDI in language tracts.

This particular brain is the worst I've seen it in so far, but I've noticed it in others too. I've checked other metrics (QA, dti_FA, RD), and it doesn't happen with those. I've attached some images showing how it looks with different overlays.

Any thoughts on what could be causing this and/or if it's something I need to be worried about in terms of validity of results?

Thanks!

Hannah

[Frank Yeh]

Thanks for your patience with my delayed reply.

It's possible that data quality might be a factor in the findings
you're observing, as issues there can sometimes lead to unexpected
results across different brain regions.

Would you be comfortable sharing the dataset with me and also the
steps you used to get these results? I'd be happy to take a look to
see if I can identify any potential quality concerns.

I can also run the analysis on my end to check if I get similar
results, which might help us pinpoint the cause.

> --
> 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/08019445-ca42-4f6d-b284-6042a2ecd4c4n%40googlegroups.com.

[Hannah Lindsey]

Sure. I'll upload the preprocessed niftis and the sz and fz files, along with the percent change maps I created using fslmaths. 

I've recently been testing qsiprep and comparing the output with those from our usual preprocessing pipeline, so qsiprep was used on this dataset. I've attached the script I used to perform it. Other than converting dicoms to nifti, the only other step that was taken prior to running qsiprep was that the affine was copied from the AP to the PA datasets (using fslcpgeom) due to an FOV mismatch error that occurred during qsiprep, based on recommendations from their help forum.

Let me know if you need anything else. Thank you!!

Hannah

[Hannah Lindsey]

oh, I just realize I didn't really explain the steps I got to get those results. 

They show up when you perform longitudinal differential tractography, using really any differential tracking parameters, but again, I only noticed them when applying the absolute and percent change in RDI color maps, which I created using fslmaths. I just exported the ses-02 images from the ses-01 after the linear registration was performed and used those when calculating the change maps.

I was specifically looking at percent decrease in RDI, so (ses-01-ses-02)/ses-01. I'm pretty sure all the tracking parameters were set to default. I do believe I ACPC aligned the preprocessed data during reconstruction, and I reconstructed using GQI.

[Frank Yeh]

There are slice artifact in the sum of all DWI image from the ses-02.sz:



This caused the slice artifact in iso and rdi in ses-02.fz



It is also visible from sub-030_ses-02_space-ACPC_desc-preproc_dwi.nii.gz


Do you have the .sz file before eddy/topup? 

Frank

To view this discussion visit https://groups.google.com/d/msgid/dsi-studio/e02a79a9-43da-4b05-b2b6-9ee13d9257b0n%40googlegroups.com.

[Hannah Lindsey]

Interesting. I don't have pre-topup/eddy sz but I just uploaded the original niftis

[Hannah Lindsey]

Okay, I see it on the raw ses-02 data. Is it a multiband acquisition issue? I'm going to reach out to the person in my lab who usually handles preprocessing and see what she thinks. Do you have ideas on how to fix it in the meantime?

[Frank Yeh]

I will check out the raw data you uploaded and see if there is a solution.

Frank

To view this discussion visit https://groups.google.com/d/msgid/dsi-studio/3d993262-6039-43bb-b47b-580797afa0afn%40googlegroups.com.

[Frank Yeh]

I've examined the data and explored various correction methods. Regrettably, I don't see an effective solution for the slice-wise signal drop.

[Hannah Lindsey]

Okay thanks so much for doing this. I’ve looked at some other data collected around the same time, and they all have it, but to a lesser extent, so I think it’s a scanner issue. 

Do you think the data is unusable?

_________________________________

Hannah M. Lindsey, PhD

Research Associate

TBI and Concussion Center

Department of Neurology

University of Utah

On Sep 5, 2025, at 3:47 pm, Frank Yeh <fran...@gmail.com> wrote:



[Frank Yeh]

Sorry for missing your question. 
It seems that the data is still usable for group wise analysis like correlational tractography.
But for more individualized analysis, the artifact won't look good on figure.

Just my two cents.

[Hannah Lindsey]

Yes, the figure would not look good. I’m relieved to hear that you don’t think the dataset itself is bad though! What concerns me is that the differential tractography showed differences almost explicitly between the two upper bands of signal loss. As I said before, we were expecting so see reduced RDI in the language tracts, so I’m hoping that that result is related to the treatment and not the artifact. Unfortunately, the signal loss appears consistently across all diffusion data (regardless of sequence) collected on that specific scanner over at least the past five years, so if you have any thoughts or concerns about trusting those data, I’d be very interested in hearing them!

_________________________________

Hannah M. Lindsey, PhD

Research Associate

TBI and Concussion Center

Department of Neurology

University of Utah

On Sep 16, 2025, at 11:29 pm, Frank Yeh <fran...@gmail.com> wrote:



[Frank Yeh]

One thing I would do is to see if the neighboring DWI correlation (NDC) of all scans have outliers, and consider just dropping those. 
If there are substantial NDC differences between the control and patient, I may also include NDC as a covariate.

Hopefully this may reduce the quality concern.

Frank