An experimental plugin to lazily load multiscale whole-slide tiff images with openslide and dask.

This napari plugin was generated with Cookiecutter using with @napari's cookiecutter-napari-plugin template.

Step 1.) Make sure you have OpenSlide installed. Download instructions here.

NOTE: Installation on macOS is easiest via Homebrew: brew install openslide. Up-to-date and multiplatform binaries for openslide are also avaiable via conda: conda install -c sdvillal openslide-python

Step 2.) Install napari-lazy-openslide via pip:

pip install napari-lazy-openslide

$ napari tumor_004.tif

By installing this package via pip, the plugin should be recognized by napari. The plugin attempts to read image formats recognized by openslide that are multiscale (openslide.OpenSlide.level_count > 1).

It should be noted that napari-lazy-openslide is experimental and has primarily been tested with CAMELYON16 and CAMELYON17 datasets, which can be downloaded here.

The OpenSlideStore class wraps an openslide.OpenSlide object as a valid Zarr store. The underlying openslide image pyramid is translated to the Zarr multiscales extension, where each level of the pyramid is a separate 3D zarr.Array with shape (y, x, 4).

import dask.array as da
import zarr

from napari_lazy_openslide import OpenSlideStore

store = OpenSlideStore('tumor_004.tif')
grp = zarr.open(store, mode="r")

# The OpenSlideStore implements the multiscales extension
# https://forum.image.sc/t/multiscale-arrays-v0-1/37930
datasets = grp.attrs["multiscales"][0]["datasets"]

pyramid = [grp.get(d["path"]) for d in datasets]
print(pyramid)
# [
#   <zarr.core.Array '/0' (23705, 29879, 4) uint8 read-only>,
#   <zarr.core.Array '/1' (5926, 7469, 4) uint8 read-only>,
#   <zarr.core.Array '/2' (2963, 3734, 4) uint8 read-only>,
# ]

pyramid = [da.from_zarr(store, component=d["path"]) for d in datasets]
print(pyramid)
# [
#   dask.array<from-zarr, shape=(23705, 29879, 4), dtype=uint8, chunksize=(512, 512, 4), chunktype=numpy.ndarray>,
#   dask.array<from-zarr, shape=(5926, 7469, 4), dtype=uint8, chunksize=(512, 512, 4), chunktype=numpy.ndarray>,
#   dask.array<from-zarr, shape=(2963, 3734, 4), dtype=uint8, chunksize=(512, 512, 4), chunktype=numpy.ndarray>,
# ]

# Now you can use numpy-like indexing with openslide, reading data into memory lazily!
low_res = pyramid[-1][:]
region = pyramid[0][y_start:y_end, x_start:x_end]

Contributions are very welcome. Tests can be run with tox, please ensure the coverage at least stays the same before you submit a pull request.

If you encounter any problems, please file an issue along with a detailed description.