Multi-modal whole slide image registration

wsireg performs multi-modal or mono-modal whole slide image registration in a graph structure for complex registration tasks using elastix. For detailed introduction and installation and usage instructions see the docs.

Features

• Graph based approach to defining modalities and arbitrary transformation paths between associated images
• Use of elastix (through ITKElastix) to perform registration
• Support for linear and non-linear transformation models
• Transform associated data (masks, shape data) along the same path as the images.
• Supports images converted to OME-TIFF using bioformats2raw -> raw2ometiff pipeline as well as array_like images from memory (np.ndarray, zarr.Array, da.core.Array from numpy, zarr, and dask, respectively)
• All registered images exported as pyramidal OME-TIFF or OME-zarr that can be viewed in software such as Vitessce,vizarr, QuPath, OMERO or any platform that supports these formats.
• All transforms for complex registration paths are internally composited and only 1 interpolation step is performed, avoiding accumulation of interpolation error from many registrations
• Shape data (polygons, point sets, etc.) in GeoJSON format (portable format for QuPath detection/annotation data since v0.3.0) can be imported and transformations applied producing a modified GeoJSON
• Some support for reading native WSI formats: currently reads .czi and .scn but could be expanded to other formats supported by tifffile

Installation

Install cross-platform Python packages with pip:

pip install wsireg

Usage

For detailed usage, see usage sections in documents, see documentation.

Example registering two images.

from wsireg.wsireg2d import WsiReg2D

# initialize registration graph
reg_graph = WsiReg2D("my_reg_project", "./project_folder")

# add registration images (modalities)
reg_graph.add_modality(
    "modality_fluo",
    "./data/im1.tiff",
    image_res=0.65,
    channel_names=["DAPI", "eGFP", "DsRed"],
    channel_colors=["blue", "green", "red"],
    preprocessing={
        "image_type": "FL",
        "ch_indices": [1],
        "as_uint8": True,
        "contrast_enhance": True,
    },
)

reg_graph.add_modality(
    "modality_brightfield",
    "./data/im2.tiff",
    image_res=1,
    preprocessing={"image_type": "BF", "as_uint8": True, "invert_intensity": True},
)

reg_graph.add_reg_path(
    "modality_fluo",
    "modality_brightfield",
    thru_modality=None,
    reg_params=["rigid", "affine"],
)

reg_graph.register_images()
reg_graph.save_transformations()
reg_graph.transform_images(file_writer="ome.tiff")

This will register the images and save them in the project output directory as tiled, pyramidal OME-TIFF. Original data types are respected for images.

Future support

• (More) Memory-efficient transformation
• 3D registration
• Automatic image conversion using bioformats
• Use of registration libraries besides elastix
• Deeper OME metadata