Question about neonatal UF tractography differences in DSI Studio

[shashank bansal]

Dear Dr. Yeh,

I am working with diffusion MRI data from infants aged approximately 1–18 months, I am using DSI Studio to analyze the uncinate fasciculus (UF).

I generated UF tracts in two ways:

1. Using the DSI Studio neonate group template.

2. Using my own neonatal data reconstructed with QSDR into a human neonatal template.

My acquisition is single‑shell DTI with 15 directions at b=800 s/mm². Even when I apply AutoTrack with the UF atlas in template space, the UF rendered on my data looks noticeably different from the one on the DSI neonate template (more fragmented and with a different overall shape).

Could you please advise on the main reasons this discrepancy might occur (e.g., limitations of 15‑direction b=800 data, QSDR registration issues, or tracking thresholds tuned for the group template) and suggest parameter adjustments or best practices to obtain a more reliable UF reconstruction in this acquisition setting?

Best regards,
Shashank Bansal

[Frank Yeh]

b=800 is much more challenging especially since neonatal axons mostly
have low anisotropy.
Even with a higher b value, the task is still not trivial.

Honestly I don't have a good solution except for multiple trial and
error: adding ROI/ROA, reconstruct data in high resolution, try
different tracking parameters, and tolerance distance,...etc

On Mon, Mar 9, 2026 at 7:06 AM shashank bansal

> --
> 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/1aa54ee6-1049-4873-9ab7-3e358b23abd6n%40googlegroups.com.

[shashank bansal]

Dear Dr. Yeh,

Thank you for your guidance.

Based on my current experiments, I have found that increasing the default Otsu threshold to around 0.8–0.9 and constraining the tract length between 20–200 mm seems to produce the most consistent results for the uncinate fasciculus in my data.

I also had a few additional questions regarding connectivity analysis:

I am currently investigating brain connectivity across different regions. My initial approach was to reconstruct the data using QSDR in DSI Studio, register it to a neonatal template, and generate a connectivity matrix using the AAL atlas. I then used this matrix for downstream analysis (control vs. disease groups) with machine learning in Python.

However, since my cohort spans a relatively wide developmental range (2–18 months), I am concerned that using a single neonatal template may not be optimal. Do you think this could significantly affect the validity of my results?

As an alternative, I am considering reconstructing the data using GQI in subject space, generating streamlines, and then applying an age-appropriate atlas for parcellation.

In this context, I had a few specific questions:

• Would performing the analysis in subject space (instead of template space) affect the reliability or comparability of connectivity measures across subjects?

• Are there any recommended tools or pipelines for structural connectivity analysis?

• In your opinion, what would be the most robust approach given the limitations of my acquisition (single-shell, 15 directions, b=800)?

I would greatly appreciate your insights.

Best regards,
Shashank Bansal

[Frank Yeh]

Sorry for the late reply.

>
> I am currently investigating brain connectivity across different regions. My initial approach was to reconstruct the data using QSDR in DSI Studio, register it to a neonatal template, and generate a connectivity matrix using the AAL atlas. I then used this matrix for downstream analysis (control vs. disease groups) with machine learning in Python.
>
> However, since my cohort spans a relatively wide developmental range (2–18 months), I am concerned that using a single neonatal template may not be optimal. Do you think this could significantly affect the validity of my results?

I don't think so, as long as all subjects are spatially normalized to
the same framework accuracy (even using an adult template). The
analysis results should be very consistent.

>
> As an alternative, I am considering reconstructing the data using GQI in subject space, generating streamlines, and then applying an age-appropriate atlas for parcellation.
>

This should also works.

> In this context, I had a few specific questions:
>
> Would performing the analysis in subject space (instead of template space) affect the reliability or comparability of connectivity measures across subjects?

No, the results should be consistent (won't be identical) if the
effect size is sufficient.

If you are aiming for a high-impact journal. The ultimate solution is
to do both. Then report one result in the main text and another in the
supplement.

>
> Are there any recommended tools or pipelines for structural connectivity analysis?

I don't have a recommendation because the differences are not very substantial.

>
> In your opinion, what would be the most robust approach given the limitations of my acquisition (single-shell, 15 directions, b=800)?

b=800 is challenging. I would have some trial and error and maybe
calculate a test-retest reliability to make sure I get good
reliability.



Hope this help,
Frank

> To view this discussion visit https://groups.google.com/d/msgid/dsi-studio/06d54eb1-0ad9-4bde-8d9a-7a8e48724ab7n%40googlegroups.com.