Hi everyone,
I have spent some time analysing some PET images from a clinic project, and there are some questions that I couldn’t solve by myself or find someone within my institution with expertise to ask. First of all, our study is composed of two cohorts (groups) and is longitudinal (two scans at different times to analyze the condition of each group). At the moment, I can only analyze the first session of the studies, as the acquisition is still ongoing. I have read a lot in different places to execute a voxel-based morphometry analysis based on statistical parameters mapping (SPM). First, I started with the SPM12 toolbox, as it was the proposed framework. I organized everything according to BIDS’ specifications. After exploring other tools like FSL, AFNI, and FREESURFER, I came back to Nilearn and Python scripting (where I find myself more comfortable).
My questions arise from the bias of only fMR images in the tutorials and explanations on using general linear models in neuroimages. As my studies are static, there is no time series signal across the images to extract, and from PET images, there is no HRF model to parameterise.
The first-level analysis, or within-subject analysis, does not make any sense to me to apply, as there is only one image per subject in each session. I have been told to do this kind of voxel-based morphometry on each subject/sesison first. Nilearn’s level model (group comparison) expects statistical maps from the first-level analysis, but I can’t understand how to get or organize the model. And second-level analysis makes more sense to me as I want to see the pattern differences due to the different conditions in both groups.
I have a reference set of images as a control of both groups/conditions. Should I use first-level analysis on each subject’s first image alongside this reference to get contrasted statistical maps? Or should I apply the first-level analysis, taking the full group’s first session set of images as a single subject for the study and then comparing them to the reference?
In SPM12, there is no first or second level of analysis for the PET&VBM option.
Additionally, I have insisted that the acquisition has not adhered to the protocol, and therefore, the images differ due to the parameters used in the scan. I am working on harmonizing this, as I understand it affects the VBM analysis. Is this assumption correct? Or are the differences due to the acquisition not significant once the images are preprocessed? I have also suggested another kind of analysis, such as connectivity, that doesn’t appear to be as affected by this as VBM could. Am I wrong with this assumption?
This could be such concept confusion from my part. I’ll appreciate every help I can get!!! 
.