Release3 restored T2w change?


exciting times - the release 3 is out :). Thank you very much for all the hard work required to get this wonderful dataset out for the community!

We managed to successfully download subset of subjects but have noticed number of differences in the dhcp_anat_pipeline files between releases 2 and 3. Namely, it seems that the field of view in the restore T2w file was changed compared to the same subject in the release 2. Also, you do not seem to provide DrawEM generated brain masks. Is there more information on specific changes between the releases?

Additionally, we are using MCRIBS segmentation approach for the cortical surface parcellation. In the release2, this was performing very well. However, the exact same pipeline fails on the release3 surfaces (the parcels seem to be shifted across the surface). We have noticed this error is probably coming from the curvature alignments between the template and subject within the MCRIBS => the curvatures computation leads to ‘reversed’ maps between same subject of release2 vs release3 (the sulci and gyri are inverted). This made us think that there might be differences in the direction of surface normals between release2 and 3 but we are not really sure. Do you have any opinion or advice on this?

Thank you very much in advance for your input.
Wishing you a lovely day,

Hi Andrea - am looking into this, will get back to you asap.

Hello Andrea,

The release 2 docs claimed that everything was in T2 space, but in fact the restore images were in a mix of T1 and T2. We fixed that in release 3 so that everything is now T2 and the docs are (hopefully) correct.

You’re right that a document summarizing the changes between release 2 and 3 would have been a good idea. We’ll bear this in mind for next time.


Hello @jcupitt
thank you very much for the info.
@EmmaR thanks for looking into the surfaces problem. Any advice will be very appreciated :slight_smile:

Wishing you both a lovely day,

Thanks a lot for the release 3!
We’re wondering if you could have to look at the differences between surfaces and curvatures of releases 2 and 3. We haven’t been able to fix the problem, although Andrea has explored the issue in depth. Is it exactly the same structural pipeline that ran on both releases? Indeed, the one described on github seems to be 4 years old.
Thanks in advance and have a nice day!

Hi Jessica

Hopefully @jcupitt or @EmmaR will respond with more detail, however looking at the structural pipeline GitHub there are several branches that have been updated more recently. I am not sure exactly which branch was used for release 3, but I will suggest to the team that the version used for release 3 gets tagged as such.

Cheers, Sean

Hi Jessica, the struct pipeline sources are here:

I’ll double-check later today, but from memory git master is the version used for release 3. The binary used for rel 3 is here:

Hi, in brief, I started speaking the MCRIB team who said 'Look’s like there is a difference in T2w between rel2 and rel3 […] rel3 T2 images are LPI whereas rel2 are RPI '. Unfortunately I haven’t had time to look since. Its on my todo.

Thanks a lot for your answers.
I let @AdaG reply for the git version…
And the information on LPI vs RPI is very useful: we’ll investigate this and keep you update!
Thanks for your time! Jessica

Thanks all for you replies and help.

We have investigated the problem in more depth. It is indeed coming from the LPI vs RPI difference. Changing this (in the header) resolves the parcellation problems.

Thanks again.