# Can we use two different seed options to generate 5M streamlines contributed from each option
**URL:** https://community.mrtrix.org/t/can-we-use-two-different-seed-options-to-generate-5m-streamlines-contributed-from-each-option/8768
**Category:** Development
**Tags:** tractography
**Created:** [May 20, 2026, 2:26am UTC](https://community.mrtrix.org/t/can-we-use-two-different-seed-options-to-generate-5m-streamlines-contributed-from-each-option/8768 "2026-05-20T02:26:00Z")
**Posts on this page:** 5
**Page:** 1
### Author:  [@suhail](https://community.mrtrix.org/u/suhail)
#### Post date: [May 20, 2026, 2:26am UTC](https://community.mrtrix.org/t/can-we-use-two-different-seed-options-to-generate-5m-streamlines-contributed-from-each-option/8768/1 "2026-05-20T02:26:00Z")
Dear Experts,
**Command** :
**tckgen** -act 5tt\_coreg.mif -backtrack \
$wmFOD $tckFName - **seed\_gmwmi** gmwmi.mif - **seed\_image** maskImage.nii.gz -select 10M -cutoff 0.05 --force
**Queries :**
1.Can we use two different seed options (seed\_image and seed\_gmwmi ) in one tckgen command run?
a) If yes does the order of the seed options influence the outcome ?
b) Can we generate the 10M streamlines, with 5M streamlines contribution from each option?
2.Generating 5M streamlines separately with the above seed options and merging both would be equivalent to above mechanism ?
3.Which above approaches could be efficient technically, if next step was to use tckedit with SIFT2 based filtering?
Thank you,
Suhail
---
### Author:  [@mblesac](https://community.mrtrix.org/u/mblesac)
#### Post date: [May 20, 2026, 10:14am UTC](https://community.mrtrix.org/t/can-we-use-two-different-seed-options-to-generate-5m-streamlines-contributed-from-each-option/8768/2 "2026-05-20T10:14:59Z")
Hi @suhail,
The approach you propose is perfectly valid, it was implemented [here](https://journals.plos.org/ploscompbiol/article?id=10.1371/journal.pcbi.1004692). You can do as many parameter permutations as you want.
In you specific scenario, I would run two tckgen commands each one with 5M streamlines and a different seeding strategy, then combine them with tckedit and run tcksift2 in the combined tractogram. I hope this helps.
Best regards,
Manuel
---
### Author:  [@suhail](https://community.mrtrix.org/u/suhail)
#### Post date: [May 21, 2026, 2:04am UTC](https://community.mrtrix.org/t/can-we-use-two-different-seed-options-to-generate-5m-streamlines-contributed-from-each-option/8768/3 "2026-05-21T02:04:25Z")
Hi Manuel,
Thank you for providing the insights on possible implementation and also directing to the ensemble tractography concept.
Best regards,
Suhail
---
### Author:  [@anagrammarian](https://community.mrtrix.org/u/anagrammarian)
#### Post date: [May 22, 2026, 1:04pm UTC](https://community.mrtrix.org/t/can-we-use-two-different-seed-options-to-generate-5m-streamlines-contributed-from-each-option/8768/4 "2026-05-22T13:04:41Z")
Hi Manuel and others,
One concern we had was choice of #streamlines for each approach. 10M for whole brain is fine. For the seed\_image, if we choose too much, it will be artifically boosted (thalamus is our seed which is much smaller than whole GM). One thought is to do a volume of GM/volume of thalamus and use this ratio to divide 10M and use that. In any case, how much mismatch can SIFT2 handle as we will SIFT2 after combining the tracks from the 2 methods.
We are already finding (thanks to rsmith comment in another thread) that seed\_dynamic is much better and fusion is not helping the big nuclei but small nuclei seem to improve which makes sense but we need to look at this on 10 or 20 subjects and see if it holds.
Thanks,
manoj
---
### Author:  [@ottogriffin59](https://community.mrtrix.org/u/ottogriffin59)
#### Post date: [July 28, 2026, 3:37pm UTC](https://community.mrtrix.org/t/can-we-use-two-different-seed-options-to-generate-5m-streamlines-contributed-from-each-option/8768/5 "2026-07-28T15:37:28Z")
One thing about running them separately—you can validate each seed’s output before merging. If one distribution looks off, you just re-run that command instead of tweaking everything. SIFT2 handles mixed distributions fine anyway.