REACT: Receptor-Enriched Analysis of functional Connectivity by Targets

The react-fmri software package allows to estimate target-enriched functional connectivity maps from functional MRI data using Positron Emission Tomography (PET) templates as spatial priors of the density distribution of neurotransmitters in the brain.

This software is coded in pure Python, its code is available on Github, it can be installed via Pypi and it is released under MIT License.

It is implemented as a two-step multivariate regression analysis:

• Step 1: the PET maps are used as a set of spatial regressors to estimate the functional connectivity in terms of fitting the BOLD fluctuations across voxels with respect to the dominant fluctuation within each of these maps. The same resolution is required for the PET maps (i.e., spatial regressors) and the fMRI images (i.e., input data). Both fMRI data and the PET maps will be demeaned at this stage, before estimating the BOLD fluctuations, in order to obtain a good fit. This step requires a mask that restricts the analysis to the voxels for which the neurotransmitter density information is available. This mask can either be estimated independently or by using the react_masks command provided in this package.
• Step 2: The BOLD fluctuations estimated in Step 1 are used as a set of temporal regressors to estimate the subject-specific target-enriched functional connectivity maps. At this stage, the fMRI data (i.e., input data) and the BOLD fluctuations (i.e., design matrix) will be demeaned before estimating the functional maps; the design matrix columns will also be normalised to unit standard deviation. This stage requires a binary grey matter mask, which can be estimated independently or by using the react_masks command provided in this package.

Usage

To normalize the PET atlases: react_normalize

react_normalize original_pet_atlas.nii.gz pet_atlas.nii.gz

• Input data (required to be in standard space):
  • original_pet_atlas.nii.gz is a 3D or 4D volume including one or more PET atlases to be used for the estimation of the target-enriched functional maps. Of note, the regions used as reference regions in the kinetic model for the quantification of the PET data should be masked out from the respective PET volume. An example is the pet_atlas.nii.gz file provided in the data directory, where the cerebellum was masked out as it was used as reference region. All values < 0 will be set at 0.
• Output: The command will normalize each PET atlas included in the original_pet_atlas.nii.gz file by shifting the minimum value to zero and rescaling the resulting values by the span between the minimum and the maximum values of the image. Then, it will generate:
  • pet_atlas.nii.gz: normalised version of original_pet_atlas.nii.gz. The order of the PET atlases is the same as the one in the original file. The optional argument -v will output the min and max values for each PET map.

To create the masks: react_masks

react_masks subject_list.txt pet_atlas.nii.gz gm_mask.nii.gz out_masks

• Input data (required to be all in the same standard space and with the same resolution):
  • subject_list.txt is a text file that specifies which functional MRI data are to be used for the creation of the masks. Typically, it includes all the subjects of the dataset. An example is the subject_list.txt file provided in the data directory.
  • pet_atlas.nii.gz is a 3D or 4D volume including one or more PET atlases to be used for the estimation of the target-enriched functional maps. Of note, the regions used as reference regions in the kinetic model for the quantification of the PET data should be masked out from the respective PET volume. An example is the pet_atlas.nii.gz file provided in the data directory, where the cerebellum was masked out as it was used as reference region.
  • gm_mask.nii.gz is a grey matter mask. A grey matter image, gm_mask.nii.gz, is provided in the data directory and can be used as input in this command. It was estimated by thesholding the probabilistic grey matter image provided by FSL at the intensity value of 77 in order to retain all voxels with a probability of at least 30% of being grey matter. The resulting thresholded image was then binarised.
• Output: The command will create two masks to be used as input in the react command. First, the script estimates a dataset-specific mask by intersecting all the subject-specific masks specified by the user in the subject_list.txt file and a PET-specific mask by intersecting all the PET atlases provided by the user in the pet_atlas.nii.gz file. Then, it will generate two masks:
  • mask_stage1.nii.gz: intersection of the dataset-specific mask, PET-specific mask and grey matter mask gm_mask.nii.gz;
  • mask_stage2.nii.gz: intersection of the dataset-specific mask and the grey matter mask gm_mask.nii.gz.

Once the two masks are generated, they can be used in react for the estimation of the subject-specific target-enriched functional maps.

To run the target-enriched fMRI analysis: react

react subject001_fmri.nii.gz mask_stage1.nii.gz mask_stage2.nii.gz pet_atlas.nii.gz REACT/subject001

• Input data (required to be all in the same standard space and with the same resolution):

  • fmri_subject001.nii.gz is the subject-specific 4D fMRI data set.
  • mask_stage1.nii.gz is the mask used in the step 1 of the multivariate regression analysis.
  • mask_stage2.nii.gz is the mask used in the step 2 of the multivariate regression analysis.
  • pet_atlas.nii.gz is a 3D or 4D volume including one or more PET atlases to be used for the estimation of the target-enriched functional maps. Of note, the regions used as reference regions in the kinetic model for the quantification of the PET data should be masked out from the respective PET volume. An example is the pet_atlas.nii.gz file provided in the data directory, where the cerebellum was masked out as it was used as reference region. We recommend to rescale each PET image in the [0,1] range after removing the reference region. This step can be done using the react_normalize command.

• Output: The command will run the two-step multivariate regression analysis and generate:

  • REACT/subject001_react_stage1.txt: subject-specific time series associated to the PET atlases used as spatial regressors;
  • REACT/subject001_react_stage2.nii.gz: subject-specific functional connectivity maps associated to the PET atlases provided as regressors. If the number of PET atlases is higher than 1, the 4D file subject001_react_stage2.nii.gz will be split into 3D files, e.g.:
    • subject001_react_stage2_map0.nii.gz
    • subject001_react_stage2_map1.nii.gz
    • subject001_react_stage2_map2.nii.gz

Requirements

• python3
• numpy >= 1.15
• scipy >= 0.19.1
• nibabel >= 3.0.0
• scikit-learn >= 0.22

Installation

You can install the react-fmri software package using Pypi by typing in your terminal:

pip install react-fmri

However, we recommend to use REACT in a dedicated environment. If you are familiar with the Anaconda Python distribution, here's how you can safely install REACT without interfering with other local software.

conda create -n react-fmri python=3
conda activate react-fmri
pip install react-fmri

Then you will need to activate the react-fmri environment for using REACT.

conda activate react-fmri
react_masks ...
react ...

Getting help

For help requests and bug reporting please use the Github issues system.

Comparison with FSL

The react-fmri package is entirely based on Python and does not need FSL to run. If you want to check that the results matches with the FSL output (i.e., using the scripts from the original REACT paper) you just need to run the following commands:

fsl_glm -i subject001_fmri.nii.gz -d pet_atlas.nii.gz -o REACT/subject001_stage1.txt -m mask_stage1.nii.gz --demean
fsl_glm -i subject001_fmri.nii.gz -d REACT/subject001_stage1.txt -o REACT/subject001_stage2 -m mask_stage2.nii.gz --demean --des_norm

How to cite REACT

• Primary Reference: O. Dipasquale, P. Selvaggi, M. Veronese, A.S. Gabay, F. Turkheimer, M.A. Mehta, "Receptor-Enriched Analysis of functional connectivity by targets (REACT): A novel, multimodal analytical approach informed by PET to study the pharmacodynamic response of the brain under MDMA", Neuroimage 2019, 195, 252-260, https://doi.org/10.1016/j.neuroimage.2019.04.007.

• Other References:

  • O. Dipasquale, D. Martins, A. Sethi, M. Veronese, S. Hesse, M. Rullmann, O. Sabri, F. Turkheimer, N.A. Harrison, M.A. Mehta, M. Cercignani, "Unravelling the effects of methylphenidate on the dopaminergic and noradrenergic functional circuits", Neuropsychopharmacology 2020, 45 (9), 1482-1489, https://doi.org/10.1038/s41386-020-0724-x

  • M. Cercignani, O. Dipasquale, I. Bogdan, T. Carandini, J. Scott, W. Rashid, O. Sabri, S. Hesse, M. Rullmann, L. Lopiano, M. Veronese, D. Martins, M. Bozzali, "Cognitive fatigue in multiple sclerosis is associated with alterations in the functional connectivity of monoamine circuits", Brain Communications 2021, https://doi.org/10.1093/braincomms/fcab023

  • D. Martins, M. Veronese, F.E. Turkheimer, M.A. Howard, S.C.R. Williams, O. Dipasquale, "A candidate neuroimaging biomarker for detection of neurotransmission-related functional alterations and prediction of pharmacological analgesic response in chronic pain", Brain Communications 2022, https://doi.org/10.1093/braincomms/fcab302

  • T. Lawn, M.A. Howard, F. Turkheimer, B. Misic, G. Deco, D. Martins, O. Dipasquale, "From neurotransmitters to networks: Transcending organisational hierarchies with molecular-informed functional imaging", Neuroscience & Biobehavioral Reviews 2023, https://doi.org/10.1016/j.neubiorev.2023.105193

• Github Repository: Ottavia Dipasquale and Matteo Frigo, "REACT-fMRI Python package", 2021, https://github.com/ottaviadipasquale/react-fmri/, DOI:10.5281/zenodo.4730559.

Developers

• Ottavia Dipasquale - original REACT implementation in FSL
• Matteo Frigo - porting to Python