Skip to contents

Fit generalized PCA at the parcel level (ClusteredNeuroVec)

fit_subject_genpca_parcel.Rd

Builds a parcel Laplacian and parcel-level A/M metrics then fits genpca on stacked runs. See voxel-level fit_subject_genpca() for an overview.

Usage

fit_subject_genpca_parcel(
  cnv_list,
  parc_vol = NULL,
  parcel_coords = NULL,
  adjacency = NULL,
  gm_vol = NULL,
  wm_vol = NULL,
  csf_vol = NULL,
  gm_p = NULL,
  wm_p = NULL,
  csf_p = NULL,
  wm_parcels = NULL,
  FD_list = NULL,
  DVARS_list = NULL,
  k = 20,
  p_ar = 1L,
  lambda_s = 0.5,
  lambda_t = 0.3,
  knn_k = 6,
  knn_sigma = 2.5
)

Arguments

cnv_list

List of ClusteredNeuroVec runs.

parc_vol, parcel_coords, adjacency

Ways to define the parcel graph: provide exactly one.

gm_vol, wm_vol, csf_vol

Optional tissue maps used if gm_p/wm_p/csf_p are not provided.

gm_p, wm_p, csf_p

Optional parcel-level tissue summaries (avoids recomputation).

wm_parcels

Optional WM parcel indices for AR whitener estimation.

FD_list, DVARS_list

Optional lists of per-run motion metrics. Each list should contain numeric vectors (one per run in cnv_list). FD is framewise displacement (mm), DVARS is temporal derivative of signal variance. Used to down-weight high-motion frames. See make_frame_weights for detailed definitions of these metrics.

k

Number of components to extract (default 20).

p_ar

AR order for temporal whitening (default 1).

lambda_s

Spatial regularization weight (default 0.5).

lambda_t

Temporal regularization weight (default 0.3).

knn_k

Number of nearest neighbors for parcel graph (default 6).

knn_sigma

Gaussian kernel bandwidth for parcel graph (default 2.5).

Value

A genpca object with parcel-level loadings.

Details

This is the parcel-level analogue of fit_subject_genpca(), operating on ClusteredNeuroVec data where voxels have been aggregated into parcels. The workflow is:

  1. Build parcel graph Laplacian from adjacency, coordinates, or labels

  2. Compute parcel tSNR across runs

  3. Aggregate tissue maps to parcels (if not provided)

  4. Construct column metric A with spatial and tissue weights

  5. Build per-run row metrics M with AR whitening and motion weights

  6. Stack runs temporally and fit generalized PCA

Parcellation reduces dimensionality and computation while preserving spatial structure through the parcel graph.

See also

fit_subject_genpca() for voxel-level analysis

Other main fitting: fit_subject_genpca(), fit_subject_metapca()

Examples

# \donttest{
# Fit genpca on parcellated data with tissue weights
fit <- fit_subject_genpca_parcel(cnv_list, parc_vol = parcellation,
                                 gm_vol = gm, wm_vol = wm, csf_vol = csf,
                                 k = 20)
#> Error: object 'cnv_list' not found
# }