Dear Renzo,

Rather than absolute thresholds or any form of one-size-fits-all solution for denoising I would advocate for adapting the denoising strategy to the needs of each individual dataset, and that includes choosing the specific scrubbing thresholds that work best in your data (see for example Wang et al. 2024 for discussing the relatively heterogeneous performance of individual denoising strategies across datasets). It is from that perspective where I am hoping that the new additions to the Denoising & QC tab in CONN (providing several measures that jointly quantify the quality of your data after denoising) will help researchers  evaluate different denoising thresholds/approaches and find those "optimal" strategies for each dataset, moving beyond the limitations of fixed/normative recommendations. 

Hope this helps

Alfonso

Wang, H. T., Meisler, S. L., Sharmarke, H., Clarke, N., Gensollen, N., Markiewicz, C. J., ... & Bellec, P. (2024). Continuous evaluation of denoising strategies in resting-state fMRI connectivity using fMRIPrep and Nilearn. PLOS Computational Biology, 20(3), e1011942.

Originally posted by Renzo Torrecuso:

Dear Alfonso and or CONN experts,

Given that Power 2012 is the consensus on the Art settings for framewise displacement (0.5 + 3 std), does anyone know a solid literature advocating for the reliability of results found with intermediate  setting (0.9 + 6 std)?

Any input is highly appreciated.

All best

Renzo