Comments on: specificity statistics for ROI analyses: a simple proposal http://www.talyarkoni.org/blog/2009/12/13/specificity-statistics-for-roi-analyses-a-simple-proposal/ ...or you get no soup for one year! Fri, 10 Feb 2012 12:37:32 +0000 hourly 1 http://wordpress.org/?v=3.3.1 By: tal http://www.talyarkoni.org/blog/2009/12/13/specificity-statistics-for-roi-analyses-a-simple-proposal/comment-page-1/#comment-95 tal Wed, 23 Dec 2009 05:19:43 +0000 http://www.talyarkoni.org/blog/?p=144#comment-95 Yeah, I think that's basically the same as what I suggest in the footnote, which is to quantify the percentile ranking of your ROI relative to all other voxels, as an alternate measure of specificity. Really the critical part is to make an honest effort to situate ROIs within the larger context of whatever else the rest of the brain is doing; as long as you're committed to that, there are probably lots of good ways to go about it. Yeah, I think that’s basically the same as what I suggest in the footnote, which is to quantify the percentile ranking of your ROI relative to all other voxels, as an alternate measure of specificity. Really the critical part is to make an honest effort to situate ROIs within the larger context of whatever else the rest of the brain is doing; as long as you’re committed to that, there are probably lots of good ways to go about it.

]]> By: Neuroskeptic http://www.talyarkoni.org/blog/2009/12/13/specificity-statistics-for-roi-analyses-a-simple-proposal/comment-page-1/#comment-93 Neuroskeptic Tue, 22 Dec 2009 09:53:06 +0000 http://www.talyarkoni.org/blog/?p=144#comment-93 This is an extremely good idea. You do get papers reporting that a certain ROI is activated by a task & then claiming that it's specific on the grounds that the other ROIs <i>they used</i> didn't activate - which sounds convincing, but it could just mean the other ROIs they used were in the wrong place (or too big, or too small). A specificity statistic would be a much better approach. Another approach might be to progressively lower the threshold on your whole-brain analysis until a blob appears. If your ROI really does identify a specific area of activation, what should happen is that the first blob to appear in the whole-brain would be in your ROI. If it was anywhere else, you'd know you'd missed something. An advantage of this is that you'd see the shape and size of the blob; one thing I worry about with ROI analysis is that you might find activation in your "amygdala ROI" but actually the blob is a whole swathe of the temporal lobe, it just so happens that the amygdala is in the middle of it. This is an extremely good idea.

You do get papers reporting that a certain ROI is activated by a task & then claiming that it’s specific on the grounds that the other ROIs they used didn’t activate – which sounds convincing, but it could just mean the other ROIs they used were in the wrong place (or too big, or too small). A specificity statistic would be a much better approach.

Another approach might be to progressively lower the threshold on your whole-brain analysis until a blob appears. If your ROI really does identify a specific area of activation, what should happen is that the first blob to appear in the whole-brain would be in your ROI. If it was anywhere else, you’d know you’d missed something. An advantage of this is that you’d see the shape and size of the blob; one thing I worry about with ROI analysis is that you might find activation in your “amygdala ROI” but actually the blob is a whole swathe of the temporal lobe, it just so happens that the amygdala is in the middle of it.

]]>