[Neurobureau-hubs] [rs-fMRI QC protocol]

[Pierre Bellec]

Dear hubs,

I am posting on this list regarding one of the brainhack 2013 (ongoing)
project: quality control (QC) of processed rs-fMRI. I just heard a terrific
talk from Frissoni, and I think what we would want is something like that:
http://www.hippocampal-protocol.net/SOPs/index.php

Basically
(1) set up a QC form,
(2) get a panel of experts to use it on a large database,
(3) get them to discuss conflicts,
(4) loop until you reach a consensus with more directions on QC choices
(and possibly new QC metrics as well), which you document in a painfully
detailed operations manual.
(5) Then you get an online interface were "naive" people who just read the
manual are presented with test cases to train and eventually certify that
people get it (when they reach a good-enough score).
(6) you grade intra-inter-rater consistency for naive raters that got
through the certification to check that it actually works.
(7) Finally, you test what is the effect of QC on a number of analysis, to
check if it does impact the results (and that it was worth doing all of
that).

Frissoni, Duchesnes and coll. have done something like that for hippocampus
segmentation instead of QC, and I think this is a really good way to help
standardize the production of imaging outcomes. For the QC, the same
approach could work for a variety of packages, as NIAK/DSPARF/CPAC and
others have a lot of the operations in common, just using different tools.
Results of QC would actually be a great way to compare the
accuracy/robustness of different pipelines (one of the main things we have
examined so far is coregistration T1->stereotaxic and T1->BOLD).

We have started working on this project with a PhD student, Yassine
Benhajali, in collaboration with some folks at NYU (Cameron Craddock, Chao
Gan Yang, Qingyang Li) as well as Maarten Mennes. This initiative started
at brainhack2012, and has been an important component of the "ABIDE
preprocessed" release (yet-to-come ... but getting there ...). Yassine has
drafted an operation manual that he is revising, and we have a number of
metrics that could be used in the first round of the panel. So we kind of
went through 1-4 once, and we should get a preliminary idea on 7 soon,
while preparing the "ABIDE preprocessed" paper.

What we need to do next is to get an actual QC report form going (we just
agreed loosely on what to look at for now), ideally web-based, and update
the operation manual. And hopefully build the panel, select a dataset
(which is likely going to remain ABIDE ...) and go through one or more
iteration. If anybody is interested to be part of that, please get in
touch. I am going to post a project description on brainhack.org, but I
prefer to start first with an email on this list in case I get some good
feedback for the project description.

Sorry about the long post. Best,

Pierre
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://www.nitrc.org/pipermail/neurobureau-hubs/attachments/20130826/3cd35ca4/attachment.html