Parcellating using dscalar

[Andraž Matkovič]

Dear HCP team,

Is there a way to use the wb_command to parcellate the dtseries.nii file using the dscalar.nii file, which contains functional maps based on soft parcellations, such as ICA or NMF? I managed to parcellate the files using Python (Nibabel + NumPy), but I was wondering if there is a direct way to do this using wb_command.

Best regards,

Andraž

[Glasser, Matthew]

We do not currently have a method of “weighted” parcellation beyond RSNRegression (which works for sICA, tICA, and PFMs using its various modes).

 

Matt.

--
You received this message because you are subscribed to the Google Groups "HCP-Users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to hcp-users+...@humanconnectome.org.
To view this discussion visit https://groups.google.com/a/humanconnectome.org/d/msgid/hcp-users/dbe8310a-bfa3-4690-b516-7dff0ba57e2bn%40humanconnectome.org.

 

---

The materials in this message are private and may contain Protected Healthcare Information or other information of a sensitive nature. If you are not the intended recipient, be advised that any unauthorized use, disclosure, copying or the taking of any action in reliance on the contents of this information is strictly prohibited. If you have received this email in error, please immediately notify the sender via telephone or return mail.

[Tim Coalson]

The cifti parcellated format effectively encodes a hard parcellation in its XML, so you'd need to generate that in addition to the computation of the soft parcellation values.  You can then combine them, either in python/matlab (or other language with a cifti library), or with -cifti-convert -from-text.  wb_command doesn't currently have the algorithms for ICA or NMF.

Hard parcellations don't overlap, so that is a simple computation of (usually) averaging the values (weighting by vertex area for correctness).  For generic soft parcellations, you may want to solve it as a multiple regression problem, such that reconstructing from the result gets close to the original timeseries data.  Of course, if the soft parcellation comes from a decomposition like ICA/NMF, the component timeseries are probably already available.

Tim

To view this discussion visit https://groups.google.com/a/humanconnectome.org/d/msgid/hcp-users/34954656-9565-48D1-820A-0DEBD5D3751E%40wustl.edu.

[Andraž Matkovič]

Thank you for the replies. It's indeed best and simplest to use the component time series provided by the algorithm. I realized that, at least for regularized NMF, these are not the same as what one would get with the regression approach (i.e., solving X = U*V' for V, where X is the fMRI time series, U are the component functional maps, and V are the component time series). This is because the algorithm iteratively optimizes both U and V; thus, X ≈ U*V, not X = U*V.

Best,

Andraž

torek, 13. januar 2026 ob 21:23:28 UTC+1 je oseba tim.c...@gmail.com napisala:

[Glasser, Matthew]

Typically, NMF produces both U and V.  You can use non-negative regression to do dual regression-like things, but we have not implemented that in RSNRegression.

To view this discussion visit https://groups.google.com/a/humanconnectome.org/d/msgid/hcp-users/87db2706-8141-48a7-b71d-3b9524b9d77cn%40humanconnectome.org.

[Andraž Matkovič]

That's true, but what I'm saying is that the U*V output of the NMF algorithm only approximately equals X. Therefore, the component time series (V) obtained by computing pinv(U)*X are not necessarily the same time series as those provided by NMF directly. In any case, this is not an issue for me at the moment because NMF and other similar algorithms typically output both matrices. I was interested in the more general case because I also wanted to estimate V based on group-based functional maps for a comparison with the individual ones.

V V čet., 15. jan. 2026 ob 13:31 je oseba 'Glasser, Matthew' via HCP-Users <hcp-...@humanconnectome.org> napisala:

You received this message because you are subscribed to a topic in the Google Groups "HCP-Users" group.
To unsubscribe from this topic, visit https://groups.google.com/a/humanconnectome.org/d/topic/hcp-users/Fsdv2bZ0i24/unsubscribe.
To unsubscribe from this group and all its topics, send an email to hcp-users+...@humanconnectome.org.
To view this discussion visit https://groups.google.com/a/humanconnectome.org/d/msgid/hcp-users/8BF17979-2F2C-4461-B1AE-0FA255C85E0C%40wustl.edu.

[Glasser, Matthew]

That is always true.  All these types of decompositions always have a + E term.

To view this discussion visit https://groups.google.com/a/humanconnectome.org/d/msgid/hcp-users/CAGicDTYjsop0N8rAzsXC0m3zm49TCTs6EuLtip1HsgK3%2BWesiw%40mail.gmail.com.