SlideSegmenter

SlideSegmenter is a Python 3.9+ package for tissue and pen marking segmentation in low-magnification (1.25x) whole slide images (WSIs). We also developed a custom bitmap annotation tool, which was used in the development of the method. More details can be found in the corresponding paper. [SPIE][arXiv]

Installing SlideSegmenter

SlideSegmenter can be installed from GitHub:

$ pip install git+https://github.com/RTLucassen/slidesegmenter

Example

A minimal example of how SlideSegmenter can be used for segmenting WSIs.

from slidesegmenter import SlideSegmenter

# initialize SlideSegmenter instance
segmenter = SlideSegmenter(device='cpu', model_folder='latest')

# segment the tissue and pen markings
cross_sections, pen_markings = segmenter.segment(low_magnification_image)

Versions

Multiple versions of the model are available, which can be selected by specifying the name of the model folder name (see options below) as argument when initializing a SlideSegmenter instance. By default, the latest version of the model is used.

• '2024-01-10': Model from the corresponding paper, trained on 140 WSIs of H&E stained skin biopsies and excisions.
• '2023-08-13': Model from a prior version of the corresponding paper, trained on 100 WSIs of H&E stained skin biopsies and excisions.

Output

The output depends on the configuration of the SlideSegmenter instance (see cases below). At initialization, there is the option to disable: (1) tissue segmentation, (2) pen marking segmentation, and (3) cross-section separation. If any of these are disabled, the corresponding decoders of the model will not be initialized, making inference faster. By default, tissue and pen marking segmentation are enabled and cross-section separation is disabled.

Case 1. All enabled

segmenter = SlideSegmenter(tissue_segmentation=True, pen_marking_segmentation=True, separate_cross_sections=True)
cross_sections, pen_markings = segmenter.segment(low_magnification_image)
# cross_sections: [H, W, S], pen_markings: [H, W, 1], where H = image height, W = image width, S = number of cross-sections

Case 2. Cross-section separation disabled

segmenter = SlideSegmenter(tissue_segmentation=True, pen_marking_segmentation=True, separate_cross_sections=False)
tissue, pen_markings = segmenter.segment(low_magnification_image)
# tissue: [H, W, 1], pen_markings: [H, W, 1], where H = image height, W = image width

Case 3. Pen marking segmentation disabled

segmenter = SlideSegmenter(tissue_segmentation=True, pen_marking_segmentation=False, separate_cross_sections=True)
cross_sections = segmenter.segment(low_magnification_image)
# cross_sections: [H, W, S], where H = image height, W = image width, S = number of cross-sections

Case 4. Cross-section separation and pen marking segmentation disabled

segmenter = SlideSegmenter(tissue_segmentation=True, pen_marking_segmentation=False, separate_cross_sections=False)
tissue = segmenter.segment(low_magnification_image)
# tissue: [H, W, 1], where H = image height, W = image width

Case 5. Tissue segmentation disabled

segmenter = SlideSegmenter(tissue_segmentation=False, pen_marking_segmentation=True, separate_cross_sections=False)
pen_markings = segmenter.segment(low_magnification_image)
# pen_markings: [H, W, 1], where H = image height, W = image width

Citing

If you found SlideSegmenter useful in your research, please consider citing our paper:

@inproceedings{lucassen2024tissue,
  author = {Ruben T Lucassen and Willeke A M Blokx and Mitko Veta},
  title = {{Tissue cross-section and pen marking segmentation in whole slide images}},
  volume = {12933},
  booktitle = {Medical Imaging 2024: Digital and Computational Pathology},
  organization = {International Society for Optics and Photonics},
  publisher = {SPIE},
  pages = {129330B},
  year = {2024},
  doi = {10.1117/12.3004683},
  URL = {https://doi.org/10.1117/12.3004683}
}