External Email - Use Caution
I have a dataset consisting of individuals scanned at two time points 7 years apart. At the first time point, a single MPRAGE scan was collected for each individual; at the second, the same MPRAGE scan was collected, as well as an additional set of 4 shorter MPRAGE scans (collected with the aim of increasing measurement reliability). I’m looking to determine the best way to model the data with the longitudinal pipeline in order to detect reliable change between the two time points (there is test-retest data in a subset at both time points for assessing this).
I’m specifically wondering whether it makes sense to compute a base “template” using all 6 scans from a given individual at once (i.e. the matching MPRAGE from tp1 and tp2 + the 4 short MPRAGEs from tp2). My intuition is that doing so would bias the template toward tp2 given that 5/6 images are from that time point, which is undesirable. I am also not sure how the different acquisition types affect things, as the shorter scans do produce systematically [slightly] different measurements. The plan is rely primarily on residual change scores for these analyses, to account for the slight differences in the acquisitions.
Any thoughts / suggestions are appreciated – thanks!
Annchen Knodt Research Associate Laboratory of NeuroGenetics Duke University
You are right that using all the scans will bias toward tp2. I think I would just run the cross for each time point using all the data you have. This will not bias the base because an average of the images within each time point will be used to make the base. If possible, you will want to make sure the scans between the time points are as similar as possible, but that might be a moot point 7 years apart.
On 12/6/2023 5:59 PM, Annchen Knodt wrote:
External Email - Use Caution
I have a dataset consisting of individuals scanned at two time points 7 years apart. At the first time point, a single MPRAGE scan was collected for each individual; at the second, the same MPRAGE scan was collected, as well as an additional set of 4 shorter MPRAGE scans (collected with the aim of increasing measurement reliability). I’m looking to determine the best way to model the data with the longitudinal pipeline in order to detect reliable change between the two time points (there is test-retest data in a subset at both time points for assessing this).
I’m specifically wondering whether it makes sense to compute a base “template” using all 6 scans from a given individual at once (i.e. the matching MPRAGE from tp1 and tp2 + the 4 short MPRAGEs from tp2). My intuition is that doing so would bias the template toward tp2 given that 5/6 images are from that time point, which is undesirable. I am also not sure how the different acquisition types affect things, as the shorter scans do produce systematically [slightly] different measurements. The plan is rely primarily on residual change scores for these analyses, to account for the slight differences in the acquisitions.
Any thoughts / suggestions are appreciated – thanks!
Annchen Knodt
Research Associate
Laboratory of NeuroGenetics
Duke University
Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer
External Email - Use Caution
Ah, that makes sense, thanks! We hadn’t really considered feeding them all in at the cross-sectional step, since to this point for these types of analyses we have been doing all averaging after freesurfer (on the individual extracted measures). But I can definitely see how it’s more appropriate to combine them at the cross-sectional step than to treat them as separate longitudinal time points. Thanks again!
From: Annchen Knodt annchen.knodt@duke.edu Date: Wednesday, December 6, 2023 at 5:59 PM To: freesurfer@nmr.mgh.harvard.edu freesurfer@nmr.mgh.harvard.edu Subject: Longitudinal pipeline with multiple scans at one time point I have a dataset consisting of individuals scanned at two time points 7 years apart. At the first time point, a single MPRAGE scan was collected for each individual; at the second, the same MPRAGE scan was collected, as well as an additional set of 4 shorter MPRAGE scans (collected with the aim of increasing measurement reliability). I’m looking to determine the best way to model the data with the longitudinal pipeline in order to detect reliable change between the two time points (there is test-retest data in a subset at both time points for assessing this).
I’m specifically wondering whether it makes sense to compute a base “template” using all 6 scans from a given individual at once (i.e. the matching MPRAGE from tp1 and tp2 + the 4 short MPRAGEs from tp2). My intuition is that doing so would bias the template toward tp2 given that 5/6 images are from that time point, which is undesirable. I am also not sure how the different acquisition types affect things, as the shorter scans do produce systematically [slightly] different measurements. The plan is rely primarily on residual change scores for these analyses, to account for the slight differences in the acquisitions.
Any thoughts / suggestions are appreciated – thanks!
Annchen Knodt Research Associate Laboratory of NeuroGenetics Duke University
freesurfer@nmr.mgh.harvard.edu
-
Annchen Knodt -
Douglas N. Greve