WSIPreprocessing

The whole slide image (WSI) pre-processing repository includes the patching process and the Vahadane colour normalisation method. These pre-processing steps are strongly recommended before applying a deep learning method.

Tiling

• Description: Divide WSI to image patches called tiles.
• See: Tiling folder
• 🎯 Usage: python Tiling/Tiling.py --WSIFolder /path/to/WSIs --outputdir /path/where/tiles/will/saved --PatientID PatientID|WSI_basename
• Help: python Tiling/Tiling.py --help
• Note:
  • This program handles WSI in svs format (Leica scanner) and mrxs format (3DHistech scanner)
  • It automatically handles x40 and x20 magnification
  • PatientID argument is expected to be the basename of the WSI file eg: if the WSI filename is patient_id1.svs then PatientID must be patient_id1
• Configuration and examples:
  • The command lines used to pre-process the Ki-67, PHH3 and HE/HES WSIs from the ESMO Open slides are available in the Tiling/config

Vahadane color normalization method

• Description: Python implementation of the color deconvolution methods from Vahadane et al (IEEE Trans Med Imaging, 2016 - https://github.com/abhishekvahadane/CodeRelease_ColorNormalization). This method can be used to artificially removed saffron stain on HES tiles extracted from WSI, and more generally to normalised color between tiles of WSIs coming from different hospitals.
• See: VahadaneColorNorm folder
• 🎯 Usage: python VahadaneColorNorm/ApplyVahadaneNormalization.py --TargetTile /path/to/reference/tile.jpg --inputdir /path/to/tiles/directory --PatientID FolderNameOfTilesToNormalized --outputdir /path/where/tiles/Normalised/will_be_saved
• Help: python VahadaneColorNorm/ApplyVahadaneNormalization.py --help
• Note:
  • The normalised tiles will be of the same sized as the non-normalised tiles
• Structure input directory:

- Non-normalized tiles directory
    - Patient ID1
            - accept (non background tiles)
                    - patient_id1_x_y.jpg
            - reject (baackground tiles)

• Structure output directory::

- Normalized tiles directory
    - Patient ID1
            - accept (non background tiles)
                    - patient_id1_x_y.jpg
            - reject (baackground tiles if ApplyVahadaneOnBackgroundTiles specified)