Fig. 1. Calculating functional brain connectivity from resting-state fMRI measurements.

Raw brain images of N = 116 participants, in total, underwent automated artefact removal, involving despiking, nuisance regression, bandpass filtering and censoring of motion-contaminated time-frames. The effects of these procedures on the BOLD signal are exemplified on the carpet-plots (a, x: time, y: voxel, colour: intensity). Subsequently, a multi-stage, high-precision brain atlas individualisation was performed to obtain regional grey-matter signals for M = 122 functionally defined brain regions (b). Partial correlation between all possible region pairs was computed to asses functional connectivity and ordered based on large-scale modularity to form individual connectivity matrices. Partial correlations of all regions with the global grey-matter signal was retained to account for, but not completely discard the effect of the global signal, a component of brain activity often regarded as a confound but also related to e.g. vigilance⁸³. c Subject-level connectivity matrices (depicted by the group-mean connectivity matrix) from Study 1 served as an input for machine-learning-based prediction of behavioural pain sensitivity.