Internship: Statistics methods for neuroimaging data reuse in the presence of sample overlap

Assignment
10 years ago, a series of publications pointed to the difficulty of reproducing scientific findings. This reproducibility crisis was a wake-up call for scientific communities to rethink how we practise research, and an important driver towards greater transparency and robust results. In the field of brain imaging (or neuroimaging), this momentum highlighted the need to increase sample sizes and gave an impetus to data sharing. Several large datasets have emerged with the goal of sharing thematic data such as the ABCD dataset for brain development. Neuroimaging has thus moved from results from small studies (N < 30 participants) to large-scale analyses (N > 1000).
Data sharing has been identified as one of the keys to maximize those two types of variability while making the best use of the available data. Secondary use of neuroimaging data involves combining multiple datasets and re-purposing them to answer a new research question. In this process – in departure from primary use – researchers typically combine pseudonymized datasets in which only minimal information is retained about each participant. While this is an important feature to protect participant’s privacy, in practice it also means that data from the same participant may be included twice without the researcher being aware of it. This induces a dependence in the input data, a phenomenon known as sample overlap.
The goal of this internship will be to measure, quantify and mitigate sample overlap in neuroimaging data reuse.
Specifically, the successful applicant will focus on task functional Magnetic Resonance Imaging (task-fMRI) an in vivo imaging technique for the study of brain function.
This work will be supervised by Dr. Camille Maumet and is part of the ANR JCJC VICUNA.
Main activities
To reach this goal you will :
Familiarise yourself with OpenNeuro and Neurovault, two repositories for brain data sharing.
Learn about neuroimaging statistics with dependant data in Magnetic Resonance Imaging (see for example (Germani et al. 2025; Lefort-Besnard et al. 2025));
Read scientific papers on sample overlap (Bom and Rachinger 2020), neuroimaging data analysis (Mumford and Nichols 2009)
Develop code to extract information from OpenNeuro and NeuroVault, build simulations
Références
Bom, Pedro R. D., and Heiko Rachinger. 2020. “A Generalized-Weights Solution to Sample Overlap in Meta-Analysis.” Research Synthesis Methods 11 (6): 812–32.
https://doi.org/10.1002/jrsm.1441.
Germani, Elodie, Xavier Rolland, Pierre Maurel, and Camille Maumet. 2025. “On the Validity of fMRI Mega-Analyses Using Data Processed with Different Pipelines.” Imaging Neuroscience 3 (April): imag_a_00522.
https://doi.org/10.1162/imag_a_00522.
Lefort-Besnard, Jeremy, Thomas E. Nichols, and Camille Maumet. 2025. “Statistical Inference for Same Data Meta-Analysis in Neuroimaging Multiverse Analyzes.” Imaging Neuroscience 3 (March): imag_a_00513.
https://doi.org/10.1162/imag_a_00513.
Mumford, Jeanette A., and Thomas Nichols. 2009. “Simple Group fMRI Modeling and Inference.” NeuroImage 47 (4): 1469–75. https://doi.org/10.1016/j.neuroimage.2009.05.034.
Skills
Required skills
Masters in statistics or data science
Strong interest in medical imaging
Good communication skills
Good programming skills
Excellent understanding of English