We are working on a project to investigate contrasting changes in regional brain blood flow during propofol-induced sedation and unconsciousness. To this aim, we use PLS to analyze PET rCBF images of 20 participants during four periods: baseline, mild sedation, deep sedation, unconsciousness. The results revealed two significant latent variables shown  in the following:

https://www.dropbox.com/s/zpv3x50o6j4tbc6/ALL_LV1.png?dl=0

https://www.dropbox.com/s/4m49la5hqcb0kc5/ALL_LV2.png?dl=0

After using PLS to identify the cerebral structures that are affected by propofol we plan to use a linear mixed modelling regression approach to identify structures showing a step change  between deep sedation to unconsciousness from those that follow a uniform linear change from baseline to unconsciousness.

To address the issue of "double dipping" (N. Kriegeskorte et al. 2009 (doi:10.1038/nn.2303))  we divided our dataset in two group of ten participants (ODD and EVEN enrolment numbers) and re-did PLS analysis for these two groups.

The results for the  ODD group follow same trends as the original dataset (with 20 participants):

https://www.dropbox.com/s/rwo9zgxh0yxg52l/ODD_LV1.png?dl=0

https://www.dropbox.com/s/xwdafli4wc8xx6e/ODD_LV2.png?dl=0

The results for LV1 of the  EVEN group follows same trends as the original dataset (with 20 participants). But LV2 of the EVEN group is a  mirror image of  LV2 of the original dataset (with 20 participants) i.e. as if the the design scores had been multiplied by -1:

https://www.dropbox.com/s/1g7xtat1wdk03qu/EVEN_LV1.png?dl=0

https://www.dropbox.com/s/2h4ktogfud58br8/EVEN_LV2.png?dl=0

How can this mirror image finding be explained?

Thank you