On the validity of fMRI mega-analyses using data processed with different pipelines

Elodie Germani ¹, Xavier Rolland ¹, Pierre Maurel ¹

⁰Univ Rennes, Inria, CNRS, Inserm, Rennes, France.

Imaging Neuroscience (cambridge, Mass.) +

|

August 13, 2025

View abstract on PubMed

Summary

Analytical variability in functional magnetic resonance imaging (fMRI) data processing can lead to false positives. Combining fMRI data from different analysis pipelines without accounting for variations inflates detection rates.

Area Of Science

Neuroimaging
Functional Magnetic Resonance Imaging (fMRI)
Data Science

Background

Publicly available neuroimaging data, particularly from functional magnetic resonance imaging (fMRI), are valuable for increasing study sample sizes and improving robustness.
Functional magnetic resonance imaging (fMRI) studies typically preprocess raw data using a consistent analysis pipeline to generate subject-level contrast maps for group analysis.

Purpose Of The Study

To investigate the impact of analytical variability, arising from different data processing pipelines, on the reliability of mega-analyses combining functional magnetic resonance imaging (fMRI) data.
To determine if unaddressed differences in fMRI analysis workflows can induce false positive findings in large-scale data aggregation.

Main Methods

Utilized the Human Connectome Project (HCP) multi-pipeline dataset, comprising contrast maps from 1,080 participants.
Processed the HCP dataset using 24 distinct analysis pipelines.
Conducted between-groups analyses comparing contrast maps generated by different pipelines to estimate pipeline-induced detection rates.

Main Results

Analytical variability stemming from diverse fMRI processing pipelines can significantly impact results when data are combined.
Failure to account for differences in analysis workflows leads to an inflation of false positive rates in mega-analyses.
The study demonstrated that using data processed with different pipelines without harmonization inflates false positive detections.

Conclusions

Analytical variability in functional magnetic resonance imaging (fMRI) data processing is a critical issue that must be addressed.
Unaccounted-for differences in fMRI analysis pipelines can compromise the validity of findings in large-scale data integration efforts.
Future mega-analyses should implement strategies to harmonize or account for variations in data processing pipelines to ensure accurate and reliable results.

Keywords:

analytical variability data re-use neuroimaging pipelines validity