jcbeer
commented
1 year ago Hi Hao,
Great questions.
Since scanner is completely confounded with time (except for the 2nd and 3rd scans), it is going to be very difficult to separate the scanner effect from the time effect. The linear mixed model including both time and scanner will be overparameterized, and basically you can only estimate the joint scanner 1 / time 1 and scanner 3 / time 4 effects. You can however separate the time 2 and time 3 and the scanner 2 effects.
If you have any other scans from scanner 1 and 3 that are not confounded by time, you may be able to use that as training data. Methods for that have not been implemented for longitudinal ComBat, but there are methods for regular ComBat, to use a control group to estimate the harmonization model and then apply that to another set of data. Basically you would need a set of scans for individuals drawn from the same population across all 3 scanners, to train the harmonization model.
You might try looking at the unharmonized individual trajectories for cortical thickness across time, to see how "smooth" they look. If the biological changes are much bigger than the scanner effects, you may be able to see those without any harmonization. But you would have to make some assumptions in that case. I'm not sure I have any better suggestions, sorry about that!
Joanne
On Fri, Jul 14, 2023 at 2:31 AM Hao Li @.***> wrote:
Dear Experts in Combat Analysis,
I am interested in employing the longcombat methodology to correct the scanner effects in my longitudinal study regarding cortical thickness. In my study, all participants underwent scanning on a 1.5T scanner in 2006. They were subsequently scanned using another 1.5T scanner in 2011 and 2015. The final scan was performed in 2020, this time employing a different 3.0T scanner. In summarly, each participant was scanned four times: the first scan used Scanner 1, the second and third scans used Scanner 2, and the fourth scan used Scanner 3.
In this situation, would the application of longcombat be an effective way to adjust for the differing scanner effects? Also, would this method inadvertently eliminate changes in cortical thickness over time after adjustment, considering that in my study, the scanner is intrinsically tied to the time point?
Thank you for your time and input on this matter.
Kind regards, Hao Li PhD Candidate
Radboudumc | Donders Institute for Brain, Cognition and Behaviour Department of Neurology Reinier Postlaan 4 Postbus 9101, 6500 HB Nijmegen E: @.*** T: 0619789454
— Reply to this email directly, view it on GitHub https://github.com/jcbeer/longCombat/issues/7, or unsubscribe https://github.com/notifications/unsubscribe-auth/AA2EB2ASWFBPJIIOWBV2D3LXQEGVXANCNFSM6AAAAAA2KCRUXY . You are receiving this because you are subscribed to this thread.Message ID: @.***>
lyhoo23618-csu
Dear Experts in Combat Analysis,
I am interested in employing the longcombat methodology to correct the scanner effects in my longitudinal study regarding cortical thickness. In my study, all participants underwent scanning on a 1.5T scanner in 2006. They were subsequently scanned using another 1.5T scanner in 2011 and 2015. The final scan was performed in 2020, this time employing a different 3.0T scanner. In summarly, each participant was scanned four times: the first scan used Scanner 1, the second and third scans used Scanner 2, and the fourth scan used Scanner 3.
In this situation, would the application of longcombat be an effective way to adjust for the differing scanner effects? Also, would this method inadvertently eliminate changes in cortical thickness over time after adjustment, considering that in my study, the scanner is intrinsically tied to the time point?
Thank you for your time and input on this matter.
Kind regards, Hao Li PhD Candidate
Radboudumc | Donders Institute for Brain, Cognition and Behaviour Department of Neurology Reinier Postlaan 4 Postbus 9101, 6500 HB Nijmegen E: hao.li@radboudumc.nl T: 0619789454