Reproducibility can be one of the bigger challenges in neuroimaging, particularly in PD, where multi-site data, heterogeneous acquisition protocols, and small cohorts are common. Below, a quick overview of pipelines worth knowing about if you’re working with multimodal MRI data.
fMRIPrep → resting-state & task fMRI The go-to for functional MRI preprocessing. It adapts automatically to your dataset’s acquisition protocol, minimises manual intervention, and outputs detailed quality reports so you can catch problem scans early. Works well with PD data, though head motion in PD patients is worth monitoring closely in your QC step.
FreeSurfer → structural MRI The standard for cortical and subcortical structural analysis. Widely used for parcellation, cortical thickness measurement, and subcortical volume extraction, including structures relevant to PD like the putamen and caudate. Its longitudinal stream is particularly useful for tracking structural change over time, which is central to most PD progression studies.
ANTsX / ANTsPyMM A strong option when you need a unified framework across modalities. Handles T1, DWI, and resting-state fMRI and integrates well with multi-site data. ANTs complements FreeSurfer well, and some workflows use both.
A 2025 replication study in PLoS ONE by Germani et al. (Predicting Parkinson’s disease trajectory using clinical and functional MRI features: A reproduction and replication study) reported that fMRI-based PD biomarkers varied meaningfully depending on which preprocessing pipeline was used, including segmentation tool choice and whether structural priors were included. Pipeline choice isn’t just a technical detail; it can affect your results. Documenting and justifying your choices matters.
Let us know what are your go-to pipelines for data processing and what is your experience using these?
Hello Amgad,
Thanks for the post! I do have a question for you. From my understanding, at an early stage, structural differences in cortical areas among PD patients or in comparison to controls may not be as significant as when you do the same for Alzheimer’s, for instance. This has led me to think we should explore additional structures as well (such as the subcortical volumes FreeSurfer extracts and brainstem and cerebellar volumes)
I was introduced by a neuroimaging expert to a module called FastSurfer, and he mentioned that this module can extract cerebellar features. So my questions are:
- Is it correct what I said about structural MRI cortical volumes at an early-PD stage being not so differentiating between patients?
- Is any one of those tools able to extract cerebellar features? As far as I was aware, FreeSurfer wasn’t able, but I’m not an expert on the field.
- What are your thoughts on FastSurfer?
Hi @danieltds, these are really important points you raise.
1. Cortical volumes in early-stage PD
Early-stage PD is predominantly characterized by subcortical pathology, particularly in the substantia nigra and basal ganglia, and subtle or absent at that stage. That said, some studies have reported mild cortical thinning in specific regions (e.g., frontal and parietal areas) even in early PD, though the effect sizes are generally smaller and more variable compared to what’s seen in other diseases like Alzheimer’s disease. I tend to agree that relying on it alone would likely limit our sensitivity, especially in early cohorts. Incorporating subcortical, brainstem, and cerebellar structures seems like a well-motivated direction.
2. Cerebellar segmentation
Standard FreeSurfer (recon-all) does perform a basic cerebellar segmentation as part of its subcortical pipeline, but it’s quite coarse. The output labels the cerebellum as a few broad regions rather than providing detailed lobular parcellation. If you are interested in more granular cerebellar features, you might want to consider dedicated tools like CERES (volBrain : CERES). Another option is the Spatially Unbiased Infratentorial Template (SUIT) toolbox (SUIT - A spatially unbiased atlas for the cerebellum and brainstem), or the cerebellar segmentation pipeline included in CAT12 (CAT Manual).
3. FastSurfer is a widely used pipeline. Its main advantage is speed, which matters a lot for large datasets. Validation studies suggest its morphometric outputs are highly comparable. Of course, the quality of the used data plays a key role in the accuracy and re-producubility any segmentatin method.
Thanks a lot for your answer, it was really deatiled!
Great explanation there @AmgadDroby
Am learning