HRF-overlap temporal confound RDM

Builds a temporal confound RDM based on predicted HRF overlap between events. Each event is convolved with a canonical HRF and pairwise similarity is computed as either normalized overlap (cosine similarity) or Pearson correlation. The result can be returned as a distance-like quantity suitable for regression.

Usage

temporal_hrf_overlap(
  onsets,
  durations = NULL,
  run = NULL,
  TR,
  hrf = c("spm", "glover", "gamma"),
  oversampling = 16L,
  length_s = 32,
  similarity = c("overlap", "corr"),
  metric = c("distance", "similarity"),
  within_blocks_only = TRUE,
  normalize = c("z", "rank", "none"),
  as_dist = TRUE
)

Arguments

onsets

numeric vector of event onsets in seconds

durations

optional numeric vector of event durations in seconds (scalar or length N). Default 0 (impulse).

run

optional vector of run/block identifiers

TR

repetition time in seconds (required)

hrf

character: one of "spm", "glover", or "gamma"

oversampling

integer oversampling factor relative to TR (default 16)

length_s

numeric length of HRF kernel in seconds (default 32)

similarity

one of "overlap" (cosine similarity) or "corr"

metric

return type: "distance" or "similarity" (default "distance")

within_blocks_only

logical; if TRUE zero-out cross-run entries (default TRUE)

normalize

one of "rank", "z", or "none"

as_dist

logical; if TRUE return a dist object (default TRUE)

Value

A dist object or symmetric matrix (N x N)

Examples

if (FALSE) { # \dontrun{
  onsets <- c(0, 5, 10, 15, 20)
  hrf_rdm <- temporal_hrf_overlap(onsets, TR=2, hrf="spm")
} # }