Ilastic-like segmentation

Note

New for version 0.4.4. More documentation planned.

Deconwolf is capable of segmenting 2D images using texture features and a random forest classifier.

Workflow

  1. Create training data

dw nuclei --init image1.tif image2.tif

This will create image1.tif.a.png and image2.tif.a.png.

Open these files with your favorite image editor and draw the nuclei (or any objects of interest) in green, and non-nuclei (background etc) in red.

  1. Create a classifier

dw nuclei --fit model.trf *.tif

This command will tell dw to read the annotated images that you created above and create a random forest classifier which will be saved to disk as model.trf. For each image it will look of an associated png image (with file extension .a.png). The tif file will simply be ignored if there is no associated png file.

  1. Classify

dw nuclei --predict model.trf file1.tif file2.tif