Histoflow

Segmentation of histological images through machine learning.

Available commands:

• -d or --data to create a new data folder or define an existing one
• --scale to scale the images by a defined factor
• -m or --mask to create a mask for the images
  • add option input if you only want to create mask input images
  • add option mask if you only want to create masks from existing input images
• -n or --normalize to normalize the images
• -t or --tiles to create tiles from the images, this takes two int values
  • the first is the <tile size>
  • the second is the <overlap size>
• -c or --classifier to train a classifier. This requires two images in the data/training folder. One must be RGB and one a mask.
• -s or --segmentation to segment all created tiles. This requires a trained classifier.
• -o or --output to stitch the segmented tiles back together.
• -v or --visualize to visualize the segmentation process, this takes two int values
  • the first is the <tile size>
  • the second is the <overlap size>

Suggested workflow:

1. Create a new data folder with the -d or --data command.
2. Copy your images into the data/input folder.
3. Scale the images with the --scale command if necessary.
4. Create a mask for the images with the -m or --mask command.
5. Normalize the images with the -n or --normalize command.
6. Create training data out of the normalized images. It consists of two images, one RGB and one binary mask.
7. Copy the training data into the data/training folder. There should be two images, one RGB and one binary mask.
8. Train a classifier with the -c or --classifier command.
9. Segment the images with the -s or --segmentation command.
10. Stitch the segmented tiles back together with the -o or --output command.
11. Optionally visualize the segmentation process with the -v or --visualize command.
12. See the results in the data/output folder.

You can combin multiple commands in one call, e.g.:

python histoflow.py -d -m -n -t 512 64 -c -s -o -v 512 64

Please cite the author if you use this code:

@software{JakobVoerkelius.2023,
  author = {Jakob Voerkelius},
  title = {Histoflow},
  titleaddon = {Segmentation of histological images through machine learning},
  version = {1.0.0},
  year = {2023},
  url = {https://github.com/Asklios/histoflow},
  orcid = {0009-0003-1630-2265},
  license = {MIT}
}