Fusion searchlight (FuSL) is a framework that allows to integrate multiple types of data in a searchlight decoding analysis.
-
Different data sources, related to different metrics or neuroimaging modalties, are combined in a searchight analysis to decode a brain state.
-
Searchlight decoding yields a spatial map of local decoding accuracies that highlight informative brain regions.
-
In a next step explainable AI (SHAP) is used to reconstruct the differential impact of each source on the decoding.
The framework is based on our recent paper:
Wein, S., Riebl, M., Brunner, L., Nothdurfter, C., Rupprecht, R., Schwarzbach, J.: Data Integration with Fusion Searchlight: Classifying Brain States from Resting-state fMRI (2024).
A demo version is provided, which can be directly run in Google Colab:
https://colab.research.google.com/github/simonvino/FuSL/blob/main/FuSL_demo_colab_surf.ipynb
This demo version uses an artificial dataset, which can serve as a template for your own data. This jupyter notebook is based on surface files in cifti format, and an alternative version using volumetric files in nifti format is provided in:
https://colab.research.google.com/github/simonvino/FuSL/blob/main/FuSL_demo_colab_vol.ipynb
Both demo versions contain a step-by-step manual, how your data can be structured to enter the FuSL workflow.
Demo jupyter notebooks you can run locally after installing the requirements are provided:
- numpy
- scikit-learn
- nilearn
- shap
- hcp-utils
- mne
Also a conda FuSL.yml file is provided. The environment can be installed with:
conda env create -f FuSL.yml
To run FuSL you can either name your files in analogy to the example files in the data folder, or you can directly format your data arrays in python as described in the demos in more detail. The MRI preprocessing pipline fMRIPrep provides one possibility to generate surface files in the cifti format.
If you find our framework useful, please reference our paper:
Wein, S., Riebl, M., Brunner, L., Nothdurfter, C., Rupprecht, R., Schwarzbach, J.: Data Integration with Fusion Searchlight: Classifying Brain States from Resting-state fMRI (2024).