Data Pre-processing for ICIAR 2018 Challenge

ICIAR 2018 Grand Challenge on Breast Cancer Histology images provides data in form of .svs and the corresponding .xml annotation files.

The main.py script cut the whole slide image into patches based on the region of interest. (Because a large region of the whole slide images are considered normal and thus are not relevant for performance evaluation.)

Environment

• Python 3.6.8.
• OpenSlide

Usage

# Install requirements
pip install -r requirements.txt

# Run the main script
python main.py

Results

After runing the main script, results of image patches will be generated in splited_svs, splited_xml, splited_svs_little, splited_xml_little, splited_xml_little_P and splited_svs_resize, inside the data/ folder.

• splited_svs: 5000 x 5000 RGB images.
• splited_xml: 5000 x 5000 RGB mask of interest.
• splited_svs_little: 1000 x 1000 RGB images cropped from splited_svs.
• splited_xml_little: 1000 x 1000 RGB images cropped from splited_xml.
• splited_xml_little_P: Gray sclae version of 1000 x 1000 RGB images cropped from splited_xml, annotate type as the pixel value.
• splited_svs_resize: Resized 500 x 500 RGB images patches of splited_svs.

Original .svs whole slide image preview

splited_svs and splited_xml samples

splited_svs_little and splited_xml_little samples

Acknowledgments

The script xml_to_mask.py is borrowed from brendonlutnick/extract_xml_region, and the whole project is inspired by it.