PET
PET is central to our appetite and food reward line,
where we image neurotransmitter systems — dopamine, opioid and endocannabinoid — that fMRI
cannot address. The pet/ folder holds the PET-specific workflows, including kinetic modelling.
Where PET differs from the fMRI pipeline
PET data need their own handling before they reach anything resembling the second-level stage:
- Kinetic modelling to derive the parameter of interest — binding potential, volume of distribution — from the dynamic time-activity data, rather than fitting a GLM to a BOLD time series
- Input functions, whether arterial or reference-region based
- Attenuation correction and motion correction over acquisitions long enough that subject movement is a certainty
- Partial volume effects, which matter more at PET’s resolution than at fMRI’s
The output is a parametric image per subject. From that point the second-level machinery
applies as it does for fMRI: the images become fmri_data objects and go through the same
group inference and multivariate pipelines.
Simultaneous PET-MR
For simultaneous PET-MR acquisitions, the two modalities are preprocessed on their own terms and then brought into a common space, allowing the receptor measure and the BOLD response to be related within the same session and the same participant.
Multivariate analysis of PET data
The PLSR pipeline explicitly supports PET feature matrices — its usage guide names PET alongside fMRI, cortical thickness and connectivity as input types. Parametric PET images are well suited to it: strongly correlated features, modest sample sizes.
For relating spatial patterns to receptor distributions, see JuSpace.
This page is an orientation rather than a manual. Scripts, exact model implementations and their options are in the repository, which is the source of truth.
Next: MR spectroscopy
Source: pet/