Hemodynamic Response Functions for fMRI Data Analysis

Creates, manipulates, and evaluates hemodynamic response functions and event-related regressors for functional magnetic resonance imaging data analysis. Supports multiple basis sets including Canonical, Gamma, Gaussian, B-spline, and Fourier bases. Features decorators for time-shifting and blocking, and efficient convolution algorithms for regressor construction. Methods are based on standard fMRI analysis techniques as described in Jezzard et al. (2001, ISBN:9780192630711).

fmrihrf

Hemodynamic response functions and event-related regressors for fMRI analysis in R.

fmrihrf provides a composable toolkit for constructing, modifying, and convolving HRFs with experimental designs. It ships with standard basis sets (SPM canonical, B-spline, Fourier, FIR, Gamma, Gaussian, and more), decorators for time-shifting and blocking, and a fast C++ convolution backend.

Installation

# From CRAN
install.packages("fmrihrf")

# Development version
remotes::install_github("bbuchsbaum/fmrihrf")

Upgrading from 0.3.0 to 0.4.0

Rebuild design matrices and refit analyses when upgrading. This release corrects SPMG response shape and the SPMG3 dispersion derivative, block integration and off-grid event timing, B-spline/tent endpoints, constant weighted windows, query-dependent block normalization, and integration of blocks shorter than the numerical step. Old coefficients are not interchangeable with new ones; neither model fits nor t-statistics are guaranteed to remain unchanged. See NEWS for details.

Cubic B-spline bases require at least two functions. When a downstream wrapper defaults to one, specify a valid basis count explicitly (for example, fmridesign::hrf(condition, basis = "bspline", nbasis = 5)). Requests for an undersized basis now error; older versions could warn and return zero columns.

Record the package version, HRF parameters, span, precision, and normalization policy with each analysis. summate = FALSE means a temporal average, not a constant peak; use normalize = TRUE when unit-peak scaling is intended. The default 24-second SPMG support remains unchanged and truncates some of the undershoot. Choose an explicit longer support when needed, for example gen_hrf(hrf_spmg1, span = 32). The kernels are continuous and unorthogonalized; matching SPM/Nilearn sampled designs requires matching sampling, support, normalization, and basis processing as well.

Quick start

library(fmrihrf)

# Evaluate the SPM canonical HRF over 0-30 seconds
t <- seq(0, 30, by = 0.1)
y <- evaluate(HRF_SPMG1, t)
plot(t, y, type = "l", xlab = "Time (s)", ylab = "Response")

# Build a regressor from event onsets
reg <- regressor(onsets = c(2, 10, 18), hrf = HRF_SPMG1,
                 duration = 0, amplitude = 1,
                 span = 24, sampling_frame = sampling_frame(blocklens = 100, TR = 1))
plot(reg)

Key features

Multiple basis sets --- Use a single canonical HRF or a flexible basis set to capture response variability.

HRF_SPMG1                                    # SPM canonical (double gamma)
HRF_SPMG3                                    # canonical + temporal & dispersion derivatives
hrf_bspline(t, N = 6)                        # B-spline basis
hrf_fourier(t, nbasis = 5)                        # Fourier basis

Decorators --- Modify any HRF through functional composition.

lag_hrf(HRF_SPMG1, lag = 2)                  # shift peak by 2 s
block_hrf(HRF_SPMG1, width = 15)             # sustained/blocked response
normalise_hrf(HRF_SPMG1)                     # unit peak-normalised

Custom HRFs --- Wrap any f(t) into the HRF system.

my_hrf <- as_hrf(function(t) exp(-t / 5), name = "exponential", span = 20)
evaluate(my_hrf, seq(0, 20, by = 1))

Regressor construction --- Convolve events with HRFs to produce design-matrix columns, with support for variable durations, amplitudes, and multi-basis expansion.

sf <- sampling_frame(blocklens = c(200, 200), TR = 2)
reg <- regressor(onsets = c(10, 30, 50), hrf = HRF_SPMG1,
                 duration = c(0, 5, 0), amplitude = c(1, 1.5, 1),
                 sampling_frame = sf)
evaluate(reg)

Fast convolution --- Core routines are implemented in C++ (Rcpp / RcppArmadillo) for efficient large-scale design matrix generation.

Documentation

Full documentation is available at https://bbuchsbaum.github.io/fmrihrf/.

Vignettes cover the main workflows:

Hemodynamic Response Functions --- overview of built-in HRFs and the basis-set system
Building Regressors --- constructing design-matrix regressors from event timing
HRF Generators --- programmatically generating families of HRFs
Advanced Modeling and Design --- multi-basis designs, trial-varying HRFs, and more

Command Line

Install the package:

install.packages("fmrihrf")

Install the command wrapper:

fmrihrf::install_cli("~/.local/bin", overwrite = TRUE)

If needed, add the directory to PATH:

export PATH="$HOME/.local/bin:$PATH"

Check the command:

fmrihrf --help

List available HRFs:

fmrihrf list --details

Evaluate a canonical HRF:

fmrihrf eval --hrf spmg1 --from 0 --to 30 --by 0.5

Evaluate a flexible basis set as JSON:

fmrihrf eval --hrf bspline --nbasis 6 --span 24 --json

Evaluate one event regressor on an acquisition grid:

fmrihrf regressor \
  --onsets 0,12,24 \
  --blocklens 200 \
  --tr 2 \
  --hrf spmg1 \
  --output regressor.csv

Build a design matrix from an event table:

fmrihrf design \
  --events events.csv \
  --blocklens 200,200 \
  --tr 2 \
  --hrf spmg1 \
  --output design.csv

The default event columns are onset, condition, block, duration, and amplitude. Use --onset, --condition, --block, --duration, and --amplitude when a table uses different column names.

License

MIT

Albers theme

This package uses the albersdown theme. Its vignettes use the albersdown::albers_vignette() output format (family = 'red', preset = 'interaction', set in each vignette's YAML), which embeds the stylesheet, script and fonts when a vignette renders. The pkgdown site uses template: package: albersdown.