Pat Gunn (dachte) wrote,
Pat Gunn
dachte

Trial Length Analysis

For those without a computer science or psychology background, this will probably be unbearably dull:Monday's imaging meeting (at work) was a continuation of our last imaging meeting, where we primarily discussed different approaches to analysis of a recent study my group did. As background (anyone who wants to come by my office can hear the long version and see the tools I use), the type of research my group does is based on multiple subjects, each of which is exposed to a long sequence of problems while in the MRI, with some of the characteristics of each problem representative of the conditions we're trying to understand. Each individual brain image takes a certain fixed amount of time (called TR, measured in seconds - my group generally chooses coverage of the brain that results in a TR of 1.5 or 2) to collect. With a temporal resolution measured in the seconds, we have a fairly crude (but spatially accurate) way to measure brain activity, but given that the underlying phonomena that we're measuring is closely related to bloodflow, that's not such an issue. The simplest kind of studies for us tend to be designed so that each trial (generally a "problem" as described above plus some time to let neural activity fall down to a baseline and possibly a distractor) takes time that's a clean multiple of the TR - that makes equivalent times within different trials line up, and makes for a clean analysis (further details of which I'll skip here, although they're computationally interesting). This simple case is most appropriate for problems where the problem is simple and should in theory take a set amount of time to study. A slightly more complex variant which my most recent experimet used had two trial lengths, 8 scans for a correct trial and 11 for incorrect. For analysis, this isn't particularly difficult because most analysis tends to focus on trials where the subject performed the task correctly, and even when that's not the case, it's not difficult to pad (at the break point) the 8-trial scans up to 11 for most types of analysis by repeating values from one or both sides of the gap. The particular study that we were discussing was a fairly complex math/memory experiment which was relatively free-form in its temporal structure, with subjects taking anywhere from a few seconds (very rare, as I understand) to a bit over a hundred per successful trial. Trying to map these into a single temporal space involves figuring out an appropriate heuristic for shrinking/expanding trials to a mean (or median) length for analysis, and that was what the meeting was about - some didn't preserve temporal structure in a neurologically plausible way, some had bad distortions at various parts of the trial structure, etc. To an extent, we would expect some of the data to be impossible to warp to the trial structure (people who somehow manage to go through each step in a few seconds have so little scan data that the shape of their activations just isn't likely to be there in the data, and they may also be somehow thinking about the problem in a very different way, while people who take a long time might be starting over a number of times), but it's clear that we're not fishing in the dark - there are characteristic shapes with each of the methods we tried that match our predicted activations, and the distortions from the predictions are all predictable (separate from our experiment) based on what we'd expect from those methods of redoing the scan order. It's a messy (and thus very interesting) problem trying to decide how to deal with the different heuristics open to us (or to find better heuristics).

It's going to be interesting seeing what Europe will look like with Blair and Chirac replaced...

Tags: science, work
Subscribe

  • Still alive

    Been feeling a bit nostalgic. Not about to return to LiveJournal - their new ownership is unfortunate, but I wanted to briefly note what's been up…

  • Unplugging LJ

    It's about time I pulled the plug on the LJ version of my blog: 1) I'm much more active on G+ than I am with general blogging. I post many times a…

  • Mutual Trust

    I don't know which should be considered more remarkable: That a cat should trust a member of a far larger and stronger species that it can't…

  • Post a new comment

    Error

    Anonymous comments are disabled in this journal

    default userpic

    Your reply will be screened

    Your IP address will be recorded 

  • 0 comments