Skip to content

Latest commit

 

History

History
53 lines (39 loc) · 3.85 KB

README.md

File metadata and controls

53 lines (39 loc) · 3.85 KB
Raw

Labels2Rois

1. Overview

This is an OMERO.web script to convert grey scale label images into OMERO ROIs.
Depending on the python packages installed in the OMERO server virtual environment one can choose between Poylgon or Mask as a Shape for the ROIs.

2. Naming convention

For now the script relies on the label images having the same exact name as the original image with "-label" added to the end.
For example Larvae_3w_400l_trRL.ome.tiff will need a label image Larvae_3w_400l_trRL-label.ome.tif for the script to recognize it.
This label image needs to be in the same dataset as the its "original" image.
The created Shapes and ROIs will be named according to the corresponding grey-value their original label had.

3. Package dependencies

For the creation of the Polygon ROIs I rely on ezomero, a fantastic toolbox to make life easier for OMERO devs.
Additionally, the script tries to import omero_rois and scikit-image and will give you feedback if those are not installed. Please consult your OMERO system-admin for installation.
The reason two packages are needed, is to give you the option what type of omero.model.Shape you want to have for your ROI.

What is a Shape?:
An OMERO ROI (omero.model.Roi) is a container object consisting of one or multiple Shapes (e.g. a Line, Polyline, Polygon, Mask, etc.) which constitute the actual forms/shapes you see in the OMERO.iviewer.

If you want to avoid additional packages and the related dependency-bloat you can still create Polygon Shapes as is, with just a small increase in script runtime.
To achieve that I refactored the underlying Cython code from the relevant scikit-image function into "pure" Python, therefore having only numpy as dependency.

3.1 Mask Shape

Creating Mask Shapes for the ROIs relies on the package omero_rois created by the OME team.

3.2 Polygon Shape

Creating Polygon Shapes for the ROIs relies on the find_contours() function from scikit-image.
In short this functions relies on the "marching squares" algorithm. For more details read the comments in the source code linked above.
To install scikit-image on our OMERO instance we had to install the following previously uninstalled packages:

PyWavelets, cycler, decorator, imageion, kiwisolver, matplotlib, networkx, scipy, tifffile

Here is the dependency tree (made with pipdeptree) for the scikit-image installation we performed on our server (green: existing packages, blue: newly installed packages):

3. Caveats

  • At the moment the input Data type is limited to Datasets and Images.
    This can easily expanded in the future if the need arises, just contact the author via mail or image.sc.
  • The script has not been tested on really complex ROI patterns. There might be situations where the underlying find_contours() function from scikit-image will fail to produce an accurate Polygon ROI.
    The underlying function to create Mask ROIs though is independent of shape complexity and might provide a good fallback option in this case.

4. Outlook

If you use the script and see room for improvement or have a special use case that is not covered by the generic code I wrote, please write an issue here at Github or contact me via mail or Image.sc.
I might implement some logic to artificially create a Polygon Shape from the Mask Shape the is created with omero_rois to better deal with complex ROI forms.
To make it more generic, there might also be the option to put in a regex pattern to determine the label images from the selection.