Regress governed short-separation nuisance signals — shortSeparationRegress • PhysioNIRS

Fits a centred short-channel regression separately for each long channel. Short channels are copied unchanged. Positive lag means that the short signal at t - lag_seconds predicts the long signal at t.

Usage

shortSeparationRegress(
  x,
  assay_name = "OD",
  short = NULL,
  method = c("nearest", "all"),
  lag_seconds = 0,
  output_assay = paste0(assay_name, "_ssr"),
  condition_limit = 1e+10
)

Arguments

x: A governed OD or pair-collapsed haemoglobin PhysioExperiment.
assay_name: Exact source assay name.
short: A compatible nirs_short_channels map returned by identifyShortChannels(), or NULL to use the strict 10-mm default.
method: Exactly "nearest" or "all".
lag_seconds: A finite lag that corresponds to an exact sample count.
output_assay: Exact name for the added corrected assay.
condition_limit: Finite regression condition-number limit greater than one.

Value

A clone of x with one short-separation-corrected assay.

Details

For a target y and selected short signals X, the fitted intercept is retained and only (X - colMeans(X)) beta is subtracted. Regression uses a deterministic SVD solve and rejects rank-deficient or over-limit designs. Short columns and lag-edge rows remain unchanged.

References

Saager and Berger (2005), DOI: 10.1364/JOSAA.22.001874.

Examples

od <- intensityToOD(PhysioNIRS:::.nirs_demo_object())
short <- identifyShortChannels(od, threshold_m = 0.015)
corrected <- shortSeparationRegress(od, short = short)
SummarizedExperiment::assayNames(corrected)
#> [1] "raw"    "OD"     "OD_ssr"