PathFlow-MixMatch

Improved giga-pixel WSI registration through automated segment-based registration. Proof-of-concept pipeline, would love your contributions!

Currently under review and additional validation, Biorxiv: https://www.biorxiv.org/content/10.1101/2020.03.22.002402v1

UPDATE: We are in the process of updating this repository with the necessary level of documentation and adding a Wiki page, stay tuned!

This package can be installed for Python 3.6+ using the following command:

pip install pathflow_mixmatch

Our latest package build can be installed using:

pip install git+https://github.com/jlevy44/PathFlow-MixMatch.git   

Minimal working example:

pathflow-mixmatch register_images --im1 A.png --im2 B.png --fix_rotation False --output_dir output_registered_images/ --gpu_device 0 --transform_type similarity --lr 0.01 --iterations 1000 --min_object_size 50000

To run without the segment based analysis, if images have black background (eg. using HistoQC) and training on the CPU:

pathflow-mixmatch register_images --im1 A.png --im2 B.png --fix_rotation False --output_dir output_registered_images/ --gpu_device -1 --transform_type similarity --lr 0.01 --iterations 1000 --min_object_size 50000 --no_segment_analysis True --black_background True

See https://airlab.readthedocs.io/ for further description of available transformations and loss functions.

Currently available loss functions:

• mse
• ncc
• lcc
• mi
• mgf
• ssim

Currently available transformations:

• similarity
• affine
• rigid