Mesmer (DeepCell) cell segmentation — sg_segment

Segments cells using the Mesmer deep learning model from the DeepCell library via reticulate. Mesmer is designed for multiplexed tissue imaging and can segment both nuclear and whole-cell compartments.

Usage

sg_segment_mesmer(
  image,
  compartment = c("whole-cell", "nuclear", "both"),
  image_mpp = NULL
)

Arguments

image: An sg_image object. Must have at least two channels: a nuclear channel (first) and a membrane/cytoplasm channel (second).
compartment: Character string specifying which compartment to segment. One of "whole-cell" (default), "nuclear", or "both".
image_mpp: Numeric scalar or NULL. Microns per pixel of the input image. If NULL, the value is taken from the image resolution metadata; if that is also unavailable, 0.5 is used as a default.

Value

An sg_mask object (for "whole-cell" or "nuclear") or a named list of two sg_mask objects (for "both").

Details

Requires a working Python installation with deepcell installed.

Examples

# \donttest{
pixels <- array(runif(200), dim = c(10, 10, 2))
img <- new_sg_image(pixels, channels = c("nuclear", "membrane"))
# Requires Python + deepcell:
# mask <- sg_segment_mesmer(img, compartment = "whole-cell")
# }