Threshold-Free Cluster Enhancement (TFCE)

Enhances a statistic map by integrating, at every point, the extent of the suprathreshold cluster containing that point raised to the power E times the threshold height raised to the power H, over all thresholds (Smith & Nichols 2009). This rewards signal that is either spatially broad or locally high without committing to a single cluster-forming threshold. Works on a time-by-channel (or time-by-frequency) matrix, or a 1-D vector, with time (and channel) adjacency.

Usage

tfce(stat_map, E = 0.5, H = 2, dh = NULL, adjacency = NULL, tail = 0)

Arguments

stat_map

A numeric matrix (time x channel) or a numeric vector.

E

Extent exponent (default: 0.5).

H

Height exponent (default: 2).

dh

Threshold step. NULL (default) integrates exactly over the distinct statistic levels; a numeric value uses a regular grid.

adjacency

Optional channel-adjacency matrix (columns of stat_map). NULL uses neighbouring-column (4-connectivity) adjacency.

tail

1 for positive effects, -1 for negative, or 0 (default) for a signed two-sided map (positive TFCE minus negative TFCE).

Value

A TFCE map with the same shape as stat_map.

References

Smith, S. M., & Nichols, T. E. (2009). Threshold-free cluster enhancement: addressing problems of smoothing, threshold dependence and localisation in cluster inference. NeuroImage, 44(1), 83-98.

See also

clusterPermutationTest()

Examples

m <- matrix(0, 100, 1); m[20:80, 1] <- 2; m[90:93, 1] <- 6
tf <- tfce(m)
c(broad = tf[50, 1], narrow = tf[91, 1])
#>     broad    narrow 
#>  20.82733 144.00000