Skip to main content
. 2020 May 1;211:116604. doi: 10.1016/j.neuroimage.2020.116604

Fig. 3.

Fig. 3

The impact of projecting to tangent space and applying shrinkage in tangent space on prediction power after deconfounding. [A-F] (HCP Data), [G,H] (UKB): The y-axis depicts the prediction accuracy/correlation for different behavioural measures. “Tangent Space” means that tangent space projection was applied on functional connectivity estimates (originally in the “Ambient Space”). The “Shrinkage” strategy means that non-isotropic PoSCE shrinkage was applied to connectivity estimates in tangent space before feeding to the predictor/classifier. “No Shrinkage” means that projected functional connectivity estimates in tangent space were directly fed to the predictor/classifier, and did not undergo PoSCE shrinkage. The violin plots show the prediction variability over 4 different parcellation schemes and 5 measures of functional connectivity estimates.