About

Work in progress

This repository is the code associated with the engineering thesis "Learning-Based Staining Normalization for Improving Registration of Histology Images".

This presents our try to evaluate deep learning approaches in stain normalization as a preprocessing step in image registration.

How to use

Keep in mind we are currently streamlining this process

For affine alignment configure and run AffineClicker/main.py. Click through all the images and save the list. Then run AffineClicker/affine_trans.py to generate a summary table of the dataset with additional affine alignment matrices.

For trainig the deep learning model make your configuration in HistNet/config.py and run HistNet/train.py

For HPC (SLURM) we provide run_hpc.slurm file that lets you configure and easily run training with live tensorboard tunneling through ssh.

Dataset

Dataset provided by the ANHIR Grand Challenge organizers and is available here

Dependencies

• PyTorch
• NumPy
• Matplotlib
• Torchio
• Kornia
• Pandas
• OpenCV