help > Why is seed autocorrelation not suppressed?
Showing 1-1 of 1 posts
Jan 15, 2018 08:01 PM | Stephen L. - Coma Science Group, GIGA-Consciousness, Hospital & University of Liege
Why is seed autocorrelation not suppressed?
Dear Alfonso,
I transmit here a question we got from a reviewer, which I found quite interesting:
"In the DMN results, it looks like the seeded regions are connected with themselves. This seems like circular reasoning. Perhaps, if MPFC and PCC were averaged together and then correlated with all other voxels in the brain this would be less concerning. However, the statement that MPFC is connected to MPFC gives the appearance that the timeseries were not averaged between MPFC and PCC before correlating across that brain, in which case I really can't understand the meaning of a region that is correlated to itself."
My understanding of this remark is that, when doing a seed-to-voxel analysis, a seed (which is the mean timeserie of the region's voxels) is always going to correlate with the voxels in the region of the seed (eg, PCC seed is always going to show a positive correlation with the PCC region), and thus the question: why is this meaningless autocorrelation not suppressed by CONN?
I thought about this and my guess is that, although autocorrelation is indeed meaningless in the case of average connectivity of a group/session, it can become meaningful when doing a difference (comparison between 2 groups or 2 sessions), because then the autocorrelation effect size can change, and thus give us an indication about the changes in internal connectivity of this region.
Could you please help us get some insights about this question? Is this the purpose of keeping seeds autocorrelation, or is there another one?
Thank you very much in advance, and happy new year!
Best regards,
Stephen
I transmit here a question we got from a reviewer, which I found quite interesting:
"In the DMN results, it looks like the seeded regions are connected with themselves. This seems like circular reasoning. Perhaps, if MPFC and PCC were averaged together and then correlated with all other voxels in the brain this would be less concerning. However, the statement that MPFC is connected to MPFC gives the appearance that the timeseries were not averaged between MPFC and PCC before correlating across that brain, in which case I really can't understand the meaning of a region that is correlated to itself."
My understanding of this remark is that, when doing a seed-to-voxel analysis, a seed (which is the mean timeserie of the region's voxels) is always going to correlate with the voxels in the region of the seed (eg, PCC seed is always going to show a positive correlation with the PCC region), and thus the question: why is this meaningless autocorrelation not suppressed by CONN?
I thought about this and my guess is that, although autocorrelation is indeed meaningless in the case of average connectivity of a group/session, it can become meaningful when doing a difference (comparison between 2 groups or 2 sessions), because then the autocorrelation effect size can change, and thus give us an indication about the changes in internal connectivity of this region.
Could you please help us get some insights about this question? Is this the purpose of keeping seeds autocorrelation, or is there another one?
Thank you very much in advance, and happy new year!
Best regards,
Stephen