Neuroflux is a fully CPU-based preprocessing pipeline for paired sMRI and PET data, built on top of powerful open-source neuroimaging tools. Given a T1w sMRI scan and its corresponding PET acquisition from the same visit, together with a reference space such as MNI152, it produces a set of standardized and analysis-ready outputs, each saved as an independent derivative.

For the sMRI scan, Neuroflux generates a registered, skull-stripped and intensity-normalized image, a binary brain mask, a multi-ROI segmentation obtained with SynthSeg 2.0, and a dedicated CSV containing ROI-wise volumetric measures. This segmentation comprises around 100 ROIs based on the Desikan–Killiany atlas. For the PET scan, Neuroflux produces a T1-aligned and template-registered image, a skull-stripped version, a PVC-corrected reconstruction and an SUVR-normalized map. Each of these outputs is saved independently, and an additional CSV reports ROI-level uptake features using the same atlas.

Installation

The use of a dedicated virtual environment, like Anaconda, is recommended to avoid dependency conflicts.
To run the entire pipeline, Docker is needed and must be downloaded.
Clone the repository and install the package in editable mode:

git clone https://github.com/antonioscardace/Neuroflux.git
cd neuroflux/
pip install -e .

Usage

The n_threads parameter is optional. By default, Neuroflux automatically selects the maximum number of available physical CPU cores (excluding logical ones) to avoid oversubscription and ensure optimal performance without counterproductive overhead.

python3 scripts/neuroflux.py \
  --mri_path PATH \
  --pet_path PATH \
  --template_path PATH \
  --mri_output_dir PATH \
  --path_output_dir PATH \
  --n_threads INT \
  --verbose

Citations

[1] Avants et al. (2009): Advanced normalization tools (ANTs). The Insight Journal, 2(365), 1–35.
[2] Marcoux et al. (2018): An automated pipeline for the analysis of PET data on the cortical surface. Frontiers in Neuroinformatics.
[3] Isensee et al. (2019): Automated brain extraction of multi‑sequence MRI using artificial neural networks. Human Brain Mapping.
[4] López‑González et al. (2020): Intensity normalization methods in brain FDG‑PET quantification. NeuroImage, 222, 117229.
[5] Tustison et al. (2021): The ANTsX ecosystem for quantitative biological and medical imaging: ANTsR, ANTsPy, and deep learning. Nature Communications.
[6] Billot et al. (2023): SynthSeg: Segmentation of brain MRI scans of any contrast and resolution without retraining. Medical Image Analysis, 83, 102789.