Fig. 9: Clustering pipelines based on performance across criteria and datasets.
From: Systematic evaluation of fMRI data-processing pipelines for consistent functional connectomics
Left: hierarchical clustering of pipelines, in terms of similarity (correlation) of their performance across datasets and criteria. The clustering solution highlights the difference between pipelines producing binary versus weighted networks. For the PDiv criterion, best performance refers to the smallest PDiv; for the propofol criterion, best performance is the greatest t-score in the correct direction; for the within-between criterion, best performance means the greatest proportion of participants for whom the within-subjects PDiv is smaller than between-subjects PDiv; for the motion correlation criterion, best performance is identified as the smallest magnitude of correlation with motion. The empty networks criterion is not included, since it is not continuous. Overall rank is the mean across all columns. Right: correlation between each pair of pipelines in terms of performance, arranged by the same hierarchical clustering. See Fig. S37 for the same figure, but sorted by overall rank.
