Surface-based methods may boost your statistical power enough (by better alignment, exclusion of irrelevant tissue, and smoothing that doesn't cross sulcal banks, if you decide you need smoothing) that you may not need to rely as much on existing ROIs. Parcel-based statistics have a lot of power, because the multiple comparisons are orders of magnitude smaller, spatially independent noise averages out, and the signal averages together. We believe that a lot of old data would benefit from reanalysis using surfaces.
However, our paper is mainly focused on specificity and continuous data. If you have a binary volume ROI and you only need a rough guess of it on the surface, you can get approximate answers, in a way that should reduce false negatives (and give more false positives) from the surface/volume transition problems. You can map the ROI to the anatomical MNI surfaces of a group of subjects, and take the max across subjects. Each individual may miss the expected group ribbon location in any given location, but it is very likely that every point in the expected group ribbon location will overlap with at least one subject in the group. If this isn't enough, you can dilate the volume ROI a few mm first. Tim On Fri, Feb 23, 2018 at 11:18 AM, Glasser, Matthew <glass...@wustl.edu> wrote: > Hi Mike, > > We have a preprint out on this exact question and the conclusion is that > it is really hard to do this accurately for most brain regions: > > https://www.biorxiv.org/content/early/2018/01/29/255620 > > Really the best idea is probably to go back and reanalyze the old data > without volume-based smoothing and aligned across surfaces. Erin Dickie, > CCed is working on tools to make this a little easier, but still there are > issues like needing a field map to get accurate fMRI to structural > registration. The good news is that one’s statistical power should be much > better if brains are actually lined up, and using parcellated analyses > instead of smoothing offers further benefits. > > Matt. > > From: <hcp-users-boun...@humanconnectome.org> on behalf of "Stevens, > Michael" <michael.stev...@hhchealth.org> > Date: Friday, February 23, 2018 at 8:58 AM > To: "firstname.lastname@example.org" <email@example.com> > Subject: [HCP-Users] Best Approach for using old volumetric data to pick > parcels-of-interest > > Hi everyone, > > > > There’s been a lot posted here over the past year or two on the challenges > and limitations of going back-and-forth between volumetric space and > HCP-defined surface space, with solid arguments for moving to (and sticking > with) CIFTI-defined brainordinates. Here, I’m asking a slightly different > question… The field has decades of research using volume-space fMRI > timeseries analyses that helps to define where to look in the brain to test > new hypotheses. Has anyone got a well-thought-out approach for mapping > such volume-space ROIs to the parcels within the new HCP 180 atlas? I ask > because the specificity of the HCP atlas sometimes offers a half dozen > candidate parcels for hypothesis-testing for what we previously thought of > as just one or two regions. Even though our group currently has a half > dozen newer NIH-funded studies that use HCP compliant sequences, most of > that work is still predicated on a “region-of-interest” approach because > the study groups sizes are less than a hundred, not in the thousands > typical of the HCP grantwork. So we still have to contend with the > statistical power limitations inherent in any ROI approach. It would be > great to be able to use our prior volume-space data to have greater > confidence in selecting among the various parcel-of-interest candidates > when testing hypotheses. > > > > I’m wondering if anyone’s yet worked out a step-by-step approach for a > series of warps/surface-maps/transformations that can take ROIs from MNI > space and give a “best guess” as to which HCP 180 atlas parcel(s) should be > queried in such instances. It would be a nice bridge from older work to > newer HCP-guided work, that would allow researchers to circumvent the added > burden of having to go back and collect new pilot data using HCP > sequences. A thoughtful list of the analytic or conceptual pros/cons of > something like this would be helpful as well. > > > > Thanks, > > Mike > > > > *This e-mail message, including any attachments, is for the sole use of > the intended recipient(s) and may contain confidential and privileged > information. Any unauthorized review, use, disclosure, or distribution is > prohibited. If you are not the intended recipient, or an employee or agent > responsible for delivering the message to the intended recipient, please > contact the sender by reply e-mail and destroy all copies of the original > message, including any attachments. * > > _______________________________________________ > HCP-Users mailing list > HCP-Users@humanconnectome.org > http://lists.humanconnectome.org/mailman/listinfo/hcp-users > > _______________________________________________ > HCP-Users mailing list > HCP-Users@humanconnectome.org > http://lists.humanconnectome.org/mailman/listinfo/hcp-users > _______________________________________________ HCP-Users mailing list HCP-Users@humanconnectome.org http://lists.humanconnectome.org/mailman/listinfo/hcp-users