Robust space-time smoothing (Huber or Tukey) + anisotropic TV

Applies a robust data term (Huber or Tukey) with spatial/temporal TV priors, optionally with motion-aware temporal weights.

Usage

stv_robust4d(
  vec,
  lambda_s = 0.8,
  lambda_t = 0.2,
  loss = c("huber", "tukey"),
  delta = NULL,
  cthresh = NULL,
  alpha = 1,
  temporal_weights = NULL,
  motion_params = NULL,
  sigma_m = 0.5,
  iters = 35L,
  mask = NULL,
  tau = NULL,
  sigma = NULL,
  theta = 1
)

Arguments

vec: 4D fMRI array or `NeuroVec`.
lambda_s: Spatial TV weight.
lambda_t: Temporal TV weight.
loss: Robust loss: `'huber'` or `'tukey'`.
delta: Huber threshold; if `NULL`, derived from noise.
cthresh: Tukey threshold; if `NULL`, derived from noise.
alpha: Tukey shape parameter.
temporal_weights: Optional length-(T or T-1) temporal edge weights.
motion_params: Optional motion regressors to derive temporal weights.
sigma_m: Motion weight scale.
iters: Iterations for the robust solver.
mask: Optional 3D logical/0-1 mask.
tau: Algorithm primal step size; if `NULL`, a sensible default is used.
sigma: Algorithm dual step size; if `NULL`, a sensible default is used.
theta: Over-relaxation parameter.

Value

Smoothed 4D data, wrapped like `vec` when possible.