Comparing the carbon footprint of fMRI data processing and analysis approaches

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Environmental Science

Background:

• Functional magnetic resonance imaging (fMRI) analysis involves computationally intensive preprocessing and statistical analysis.
• Software packages like FSL, SPM, and fMRIPrep are commonly used for fMRI data processing.
• The environmental impact, specifically carbon emissions, of these software packages is not well understood.

Purpose of the Study:

• To compare the carbon emissions associated with preprocessing and statistical analysis of fMRI data across FSL, SPM, and fMRIPrep.
• To evaluate the scientific performance, in terms of statistical sensitivity, of each software package.
• To provide guidance for researchers in selecting fMRI software based on both environmental impact and analytical performance.

Main Methods:

• Utilized an existing open-access fMRI dataset for comparative analysis.
• Quantified and compared the carbon emissions generated by FSL, SPM, and fMRIPrep during data processing.
• Assessed the statistical sensitivity of each package to detect brain activation, considering regional variations.

Main Results:

• fMRIPrep exhibited substantially higher carbon emissions, approximately 30x that of FSL and 23x that of SPM.
• fMRIPrep demonstrated slightly superior statistical sensitivity compared to FSL and SPM.
• FSL outperformed SPM in statistical sensitivity, though performance patterns varied across different brain regions.

Conclusions:

• Researchers must consider the substantial carbon footprint of fMRIPrep when choosing neuroimaging software.
• A trade-off exists between the computational cost (carbon emissions) and the scientific performance of fMRI analysis tools.
• Future development and usage of fMRI software should prioritize energy efficiency and minimize unnecessary computational demands while ensuring high-quality results.