Email Icon Facebook icon Twitter Icon GooglePlus Icon Contact

User Top Menu

The Baby and the Bathwater: Signal and Noise in Psychiatric Neuroimaging

Led by John Krystal. Posted on 15 Jun 2016

John Krystal

On June 6, SRF advisor John Krystal of Yale University moderated a discussion among Alan Anticevic, Deanna Barch, Randy Buckner, and Daniel Weinberger on critical issues in the interpretation of magnetic resonance imaging (MRI) data in psychiatric research.

For some years now, the methodologists in the neuroimaging community have been working to resolve a well-known confounder in magnetic imaging (MRI) research: the fact that head movement and breathing are hard to control for. This issue was laid out in detail for the psychiatry research community in a recent review in the American Journal of Psychiatry by Daniel Weinberger and Eugenia Radulescu of the Lieber Institute for Brain Development. They mention other confounders to contemplate, writing that factors such as smoking, medical comorbidities, drugs, and perhaps even the experience of being in an MRI machine vary between patients and control groups, and conclude that, "the evidence that findings are neurobiologically meaningful is inconclusive and [these] may represent artifacts or epiphenomena of uncertain value."

We invite you to read commentaries on the Weinberger and Radulescu review by Alan Anticevic and Matthew Glasser and by Godfrey Pearlson, Vince Calhoun, and David C. Glahn. These researchers concur with some of the assertions, but push back against others, arguing that the field is solving some of the confounds and that others are not as serious as Weinberger and Radulescu present. We welcome the submission of comments.


Listen to the Webinar


Alan Anticevic's Presentation

Last comment on 6 Jun 2016 by Angus MacDonald


Submitted by Jared Van Snellenberg on

I think one important way we can move forward with neuroimaging measures in order have greater confidence that observed differences in these measures are meaningful, even in a simple case-control design, is to show that the neuroimaging measure in question is correlated with another more directly meaningful non-neuroimaging measure. For example, if a BOLD signal difference is associated on an individual-subject basis with working memory capacity or with delusional symptomatology, then we can have more confidence that that neuroimaging measure reflects some meaningful aspect of a brain-behavior relationship. Regarding Dr. Weinberger's comments, I think it is absolutely the case that linking neural measures to behavior or symptoms, or some other independently measurable phenomenon that is interpretable separate from the neuroimaging data, gives us a much stronger ability to interpret that neuroimaging data. This can raise our confidence that what we are measuring is not simply movement related or due to some other systematic confound.

Submitted by Angus MacDonald on

The pile-on tone of this webinar derives entirely from our reliance on pseudo-experimental designs in psychopathology. As Dr. Barch noted, this isn't about imaging per se, but imaging has a specific set of limitations in such designs. Can we get more quantitative measurements to compare the effect sizes associated with confounds versus diagnosis or symptoms? Would that guide us to know which of these concerns are confounds in principle versus real?

Many of the confounds expounded by Weinberger largely affect global signals. Under these circumstances, unaffected voxels (components, or regions) serve as controls for affected voxels. So there are internal controls in this regard. The extent to which the confounds don't affect global signals could presumably be mapped (e.g., using angiography etc.). Would it be useful to have a set of maps that could be compared to activation maps or structural differences to "guide" interpretation of differences that are found?