Sliding window correlation

Compute functional connectivity matrices over sliding windows of a BOLD timeseries (Hansen et al., 2015).

Usage

corr_slide(timeseries, window, step = NULL)

Arguments

timeseries

Numeric matrix [N x Tmax]. BOLD signal with N parcels as rows and Tmax timepoints as columns.

window

Integer. Window size in timepoints.

step

Integer. Step between window onsets. Defaults to window (non-overlapping windows).

Value

A list with:

corr_mats

Array [N, N, n_windows] of Pearson correlation matrices.

idx

Integer vector of 1-indexed window onset positions.

Details

The correlation matrices are computed by a compiled C++ backend (corr_slide_cpp()) that calculates Pearson correlation directly on the input matrix. No per-window submatrix allocation or transposition is performed; the t-outer loop ordering exploits the column-major memory layout of the input for sequential cache access.

References

Hansen, E. C. A. et al. (2015). Functional connectivity dynamics: Modeling the switching behavior of the resting state. NeuroImage, 105, 525-535. doi:10.1016/j.neuroimage.2014.11.001

Examples

set.seed(1)
ts <- matrix(rnorm(10 * 200), nrow = 10, ncol = 200)
res <- corr_slide(ts, window = 20)
dim(res$corr_mats)  # 10 x 10 x 10
#> [1] 10 10 10