StarDist cell segmentation — sg_segment

Segments cells (typically nuclei) using the StarDist deep learning model via reticulate. Requires a working Python installation with stardist and tensorflow installed. Use sg_setup_python() to configure the environment.

Usage

sg_segment_stardist(
  image,
  model = c("2D_versatile_fluo", "2D_versatile_he", "2D_paper_dsb2018", "custom"),
  channel = 1L,
  prob_thresh = 0.5,
  nms_thresh = 0.4,
  scale = NULL,
  custom_model_path = NULL,
  n_tiles = NULL
)

Arguments

image: An sg_image object.
model: Character string specifying the StarDist model. One of "2D_versatile_fluo" (default), "2D_versatile_he", "2D_paper_dsb2018", or "custom".
channel: Integer index of the image channel to segment. Default is 1L.
prob_thresh: Numeric; object probability threshold. Objects with probability below this value are discarded. Default is 0.5.
nms_thresh: Numeric; non-maximum suppression overlap threshold. Overlapping detections above this threshold are merged. Default is 0.4.
scale: Numeric vector of length 1 or 2, or NULL. Scaling factor applied to the image before prediction. If NULL, no scaling is applied.
custom_model_path: Character string or NULL. Path to a custom-trained StarDist model directory. Required when model = "custom".
n_tiles: Integer vector of length 2 or NULL. Number of tiles in each dimension for processing large images. If NULL, StarDist determines tiling automatically.

Value

An sg_mask object with labelled cell regions.

Examples

# \donttest{
pixels <- matrix(runif(400), nrow = 20, ncol = 20)
img <- new_sg_image(pixels)
# Requires Python + stardist:
# mask <- sg_segment_stardist(img, model = "2D_versatile_fluo")
# }