(a) PCA analysis identified linear combinations of hctsa time-series features with maximum variance across the cortex. Collectively, the first two components (PC1 and PC2) account for 75% of the total variance in time-series features of BOLD dynamics. To estimate the extent to which cortical regions display the patterns of intrinsic dynamics captured by each component, hctsa matrices were projected back onto the PC weights (eigenvectors), yielding spatial maps of brain scores for each component. Spatial maps are depicted based on the standard deviation of their respective brain score distributions. (b) To understand the feature composition of the intrinsic dynamic patterns captured by PC1 and PC2, feature loadings were computed by correlating individual hctsa feature vectors with the PC score maps. PC loadings thus estimate the shared spatial variance between an individual time-series feature and the composite intrinsic dynamic map captured by a PC. time-series features are ordered by their individual loadings. Grey indicates non-significance based on 10,000 spatial permutation tests (FDR correction, ). Features corresponding to the top and bottom 5% of PC1 and PC2 are visualized using word clouds. The complete list of features (ranked by loading), their definitions, correlations and p-values for both components is presented in machine-readable format in Supplementary files 3 and 4. Feature nomenclature in hctsa is organized such that the term prefix indicates the broad class of measures (e.g. AC = autocorrelation, DN = distribution) and the term suffix indicates the specific measure (for a complete list, see https://hctsa-users.gitbook.io/hctsa-manual/list-of-included-code-files). (c) The spatial distributions of two high-loading representative time-series features are depicted for each component, including lag-1 linear autocorrelation (AC_1) and lag-[0,2,3] nonlinear autocorrelation (AC_nl_023, estimated as average across time-series x) for PC1; and kurtosis (DN_Moments_4) and entropy (EN_DistributionEntropy_ks__02) of the time-series points distribution for PC2. To build intuition about what each component reflects about regional signals, three regional time-series from one participant are selected based on their lag-1 autocorrelation and kurtosis (circles on the brain surface: pink = 5th percentile, green = 50th percentile, purple = 95th percentile). Going from low-ranked to high-ranked regions results in a slowing down of BOLD fluctuations for the former and increasingly heavier symmetric tails of the signal amplitude distributions for the latter. Note that normalized feature values are shown in the first row, whereas the raw feature values are shown in the second row.