Authors: A.I. Luppi, H.M. Gellersen, Z-Q. Liu, A.R.D. Peattie, A.E. Manktelow, R. Adapa, A.M. Owen, L. Naci, D.K. Menon, S.I. Dimitriadis & E.A. Stamatakis
This repository provides the Interactive Pipeline Selection Tool from Luppi et al., "Systematic evaluation of fMRI data-processing pipelines for consistent functional connectomics" Nature Communications (2024) (published article).
The Pipeline Selection Tool provides the results of systematically assessing 768 pipelines for obtaining a functional connectivity network from (preprocessed) functional MRI data. The pipelines consist of all possible combinations of the following steps: (i) Global signal regression (GSR or no GSR); (ii) Definition of network nodes: AAL, Brainnetome, Glasser, Lausanne multi-scale anatomical (129, 234 or 463 nodes), Schaefer-Tian multi-scale functional (116, 232, or 454 nodes), Independent Components Analysis (100, 200, or 300 components). (iii) Number of nodes: approximately 100, 200 or 300-400. (iv) Edge definition: Pearson correlation or Mutual Information. (v) Edge filtering: pre-specified density (retaining 5%, 10%, or 20% of total edges, or matching the density of the structural connectome); pre-specified minimum weight (0.3 or 0.5); or data-driven methods (Efficiency Cost Optimisation, and Orthogonal Minimum Spanning Trees). (vi) Binary or weighted networks.
Each pipeline is benchmarked against the following criteria for trustworthy functional connectomics: Criterion (I): Avoiding random test-retest differences. Criterion (II): Detecting true experimental differences (reconfiguration induced by anaesthesia). Criterion (III): Detecting inter-individual differences. Criterion (IV): Avoiding systematic motion-induced differences. Criterion (V): Producing non-empty networks.
To further ensure that results and recommendations are generalisable, each criterion is assessed for each pipeline in four test-retest datasets (spanning ~1h, 2-4 weeks, 1-11 and 5-16 months) including different acquisition parameters and preprocessing/denoising (aCompCor and FIX-ICA).
As reported in the peper, the majority of pipelines do not meet all criteria consistently. We identified a number of pipelines that satisfy every criterion in every dataset, whose use we recommend. However, we also provide the Pipeline Selection Tool so that users can choose pipelines based on their own combination of criteria and requirements. To make this repository self-contained, a PDF Guide to the Pipeline Selection Tool is also provided. The paper (available Open Access) provides a full description of each criterion and dataset.
For questions, please email: al857@cam.ac.uk.