This repository provides scripts to generate annotation masks for tissue/cell segmentation using immunofluorescence restaining.

Python 3.7 or newer

• numpy
• matplotlib
• seaborn
• joblib
• pandas
• scipy
• openslide
• pillow
• tqdm
• cellpose
• mlflow
• opencv
• pytorch
• torchvision
• pytorch lightning
• torchmetrics
• segmentation_models_pytorch
• albumentations
• scikit-image
• kornia
• optuna
• dali
• SimpleITK
• imreg_dft

This script extracts patches from Whole Slide Images (.ndpi) of tissue microarray sections after rigid and non-rigid registration between H&E-stained and immunofluorescence (IF)-restained sections.

usage:

python 1.registration_patch_extraction.py targetdir outdir [option] 

targetdir must contain subdirectories, each of which have the following three .ndpi files.

1. HE-stained WSI file the file must contain either 'DAPI' or 'Opal' in its name.
2. DAPI-stained WSI file (the file must contain 'DAPI' in its name.)
3. IF-stained WSI file (the file must contain 'Opal' in its name.)

The slides must be scanned at 40x magnification.

The last directory name of targetdir is used in other scripts as antibody.

output:

• HE-stained patch file, which ends with _HE.png
• IF-stained patch file, which ends with _IHC_nonrigid.png

The HE and IF-image pair files have the same prefix in their name.

This script runs Cellpose to the extrated patches (for cell segmentation).

usage:

python 2_CELL.run_cellpose.py input_dir [option] 

input_dir is the output directory created by 1.registration_patch_extraction.py

output:

• image of IF-stained regions detected by Cellpose, which ends with _IHC_cellpose_nonrigid.png
• cell mask, which ends with _mask.pkl

This script generates the segmentation masks based on the patches from IF-restained sections and the Cellpose output.

usage:

python 3_CELL.mask_generation.py input_dir 

input_dir is the output directory containig files created by 2_CELL.run_cellpose.pyy

output:

• mask of IF-stained regions detected by Cellpose, which ends with _IHC_cellpose_mask.png

This script generates the segmentation masks for red blood cells based on the patches from IF-restained sections.

usage:

python 3_RBC.mask_generation.py input_file [option] 

input_file is the output HE-stained patch file created by 1.registration_patch_extraction.py

output:

• RBC mask file, which ends with _IHC_nonrigid_mask2.png

This script generates the segmentation masks for tissues based on the patches from IF-restained sections (requires MLFlow).

usage:

python 3_REGION.mask_generation.py input_dir [option] 

Note: Please modify the source code so that the RBC segmentation model can be loaded from MLFlow server (l.22-25, l.100).

input_dir is the output directory created by 1.registration_patch_extraction.py

output:

• mask file, which ends with _IHC_nonrigid_mask2.png

This script calculates blur level and the correlation between DAPI and Hematoxylin signal.

usage:

python 4.QC_make_summary.py input_dir 

input_dir is the output directory created by 1.registration_patch_extraction.py

output:

• csv file which contains blur level and the correlation between DAPI and Hematoxylin signal for each patch file.

This script filters out patches based blur level and the correlation between DAPI and Hematoxylin signal.

usage:

python 5.filter_QC.py input_dir antibody 

input_dir is the output directory created by 1.registration_patch_extraction.py antibody is used in the output csv.

output:

• csv file which contains blur level and the correlation between DAPI and Hematoxylin signal for each patch file.

This script trains the segmentation models (requires MLFlow). usage:

python 6.train_segmentation_model.py antibody [option]

Note: Please modify the source code so that the RBC segmentation model can be loaded from MLFlow server (l.192-195).

output:

• Segmentation model (stored in MLFlow).