stainwarpy

stainwarpy is a command-line tool and a Python package for registering H&E stained and multiplexed tissue images. It provides a feature based registration pipeline, saving registered images, transformation maps and evaluation metrics.

Features

• Register H&E images and multiplexed images (after extracting DAPI channel) using transformations.
• Supports feature-based registration.
• Outputs registered images, transformation maps and evaluation metrics (TRE and Mutual Information).
• Transforms segmentation masks based on the computed transformations

Recommendations

• For most cases, it is recommended to register H&E images onto multiplexed images (H&E as moving image).
• The default similarity transformation usually works well and stable, therefore recommended.

Installation

You can install stainwarpy using pip:

pip install stainwarpy

or

Bioconda (for Linux/macOS):

conda install -c bioconda stainwarpy

---

Usage as a command-line tool

Register Images

stainwarpy register <fixed_path> <moving_path> <output_folder> <final_img_sz> [options]

Examples:

stainwarpy register data/fixed_img.ome.tiff data/moving_img.ome.tiff ../output multiplexed multiplexed

stainwarpy register data/fixed_img.tif data/moving_img.tif ../output multiplexed multiplexed --multiplexed-px-sz 0.21 --hne-px-sz 0.52

Arguments:

• multiplexed_path: Path to the multiplexed image (.tif/.tiff./.ome.tif/.ome.tiff)
• hne_path: Path to the H&E image (.tif/.tiff./.ome.tif/.ome.tiff)
• output_folder: Folder to save the registered images and metrics
• fixed_img: Which one to be taken as fixed image multiplexed or hne
• final_img_sz: Final moving image pixel size to be kept in the size of multiplexeed or hne image pixel size

Options:

• --multiplexed-px-sz : Pixel size of the multiplexed image (no need to provide for ome.tiff, so default: None)
• --hne-px-sz : Pixel size of the H&E image (no need to provide for ome.tiff, so default: None)
• --feature-tform : Feature transformation method: similarity oraffine or projective (default: similarity)
• --channel-idx : Channel index (DAPI) to extract if channel extraction not done beforehand for multiplexed image (default: 0), not used if already extracted.

Output

After running registration, the following files/folders will be generated and saved in the specified output folder:

• registration_metrics_tform_map.json — TRE and Mutual Information and transformation map in an user friendly file format
• 0_final_channel_image.ome.tif — Registered image (in the pixel size of moving image)
• feature_based_transformation_map.npy — Transformation map

Extract a Channel (DAPI can be extracted for registration)

stainwarpy extract-channel <file_path> <output_folder_path> [--channel-idx N]

Arguments

• file-path : Path to multichannel image (.tif/.tiff/.ome.tif/.ome.tiff)
• output-folder-path : Folder to save the extracted channel image

Options

• --channel-idx: Channel index to extract (default: 0 for DAPI)

Output

• multiplexed_single_channel_img.ome.tif - Image with the extracted channel saved in the specified output folder

Transform segmentation Masks

Transform segmentation masks based on the transformation maps produced with the command register.

stainwarpy transform-seg-mask <mask_path> <fixed_path> <moving_path> <output_folder_path> <tform_map_path> <multiplexed/hne> <multiplexed/hne> [options]

Arguments

• mask_path : Path to the segmentation mask of the moving image (.ome.tif/.ome.tiff/.tif/.tiff/.npy)
• multiplexed_path : Path to the multiplexed image (.tif/.tiff/.ome.tif/.ome.tiff)
• hne_path : Path to the H&E image (.tif/.tiff/.ome.tif/.ome.tiff)
• output_folder_path : Folder to save the transformed segmentation mask
• tform_map_path : Path to the transformation map
• fixed_img : Which image taken as fixed image multiplexed or hne
• final_mask_sz : Pixel size for final mask: multiplexed or hne

Options

• --multiplexed-px-sz : Pixel size of the multiplexed image (no need to provide for ome.tiff, so default: None)
• --hne-px-sz : Pixel size of the H&E image (no need to provide for ome.tiff, so default: None)

Output

• transformed_segmentation_mask.ome.tif : The segmentation mask transformed to the fixed image coordinate space saved in the specified output folder

---

Usage as a Python Library

Although stainwarpy is mainly a command-line tool, its functions can also be used directly in Python for scripting.

Example: Running the Registration Pipeline

from stainwarpy.regPipeline import registration_pipeline

# run registration pipeline
tform_map, final_img, tre, mi = registration_pipeline(
    fixed_path="fixed_image.tif",
    moving_path="moving_image.tif",
    fixed_px_sz=0.5,
    moving_px_sz=0.5,
    fixed_img="multiplexed",
    final_img_sz="fixed",
    feature_tform="affine"        # to use a transformation other than default "similarity"
)

print("TRE:", tre)
print("Mutual Information:", mi)

---

License

This project is licensed under the MIT License.

This project includes portions of code in stainwarpy/preprocess.py adapted from HistomicsTK (https://github.com/DigitalSlideArchive/HistomicsTK/), which is licensed under Apache License 2.0. See LICENSE_HISTOMICSTK.txt for the full license text.