edges.edge_strength

edges.edge_strength(
    pcs,
    valid,
    res_um,
    sigma_um=DEFAULT_SIGMA_UM,
    erode_um=DEFAULT_ERODE_UM,
    aggregate=DEFAULT_AGGREGATE,
)

Gradient magnitude of whitened PCs, aggregated across PCs, per mm.

Each PC is Gaussian-smoothed (NaN-aware: filtered data divided by the filtered mask, which avoids a bias at the mask border), then its gradient magnitude (root of the summed squared derivatives over the three axes) is taken. The smoothing removes the blocky grid of the interpolated volume. The per-PC magnitudes are then aggregated across PCs: "median" keeps voxels where many PCs change together and "mean" averages them.

Parameters

Name	Type	Description	Default
pcs	np.ndarray	(AP, ML, DV, k) whitened PCs, e.g. load_whitened_pcs(...)["pcs"].	required
valid	np.ndarray	(AP, ML, DV) bool mask of voxels with data.	required
res_um	float	Voxel size, sets the sigma in voxels and the gradient spacing.	required
sigma_um	float	Gaussian sigma in um (0 = unsmoothed).	DEFAULT_SIGMA_UM
erode_um	float	Voxels this close to the mask border are dropped: their gradient is a mask artefact.	DEFAULT_ERODE_UM
aggregate	(median, mean)	How the per-PC gradient magnitudes are combined across PCs.	"median"

Returns

Name	Type	Description
	np.ndarray	(AP, ML, DV) float32 edge strength (units: EDGES_UNITS [aggregate]), NaN outside the eroded mask.