Towards Efficient FMRI Data Re-Use: Can We Run Between-Group Analyses With Datasets Processed Differently?

The increased amount of shared data creates an opportunity to reuse existing data to reach larger sample sizes and hence increase statistical power in neuroimaging studies. However, doing so may require to perform analyses using subject data processed differently. Here, we performed between-group analyses under the null hypothesis (making any detection a false positive), with data from the Human Connectome Project (HCP) (n=1080) processed with different pipelines. We compared the estimated false positive rates obtained to the theoretical false positive rate, to assess whether the variability in processing pipelines (called analytical variability) impacts the validity of the analyses. We found that some differences in parameter values caused invalidity, suggesting that analytical variability has to be taken into account before combining subject data processed with different pipelines.