Support in dask_image for different image formats

[ruaridhg]

Check which formats dask_image supports/implement unsupported but common formats (DICOM, nifti)
Has an imread function
Chunks like slices

[K-Meech]

dask seems to provide two main ways of reading image data to arrays:

• dask_image.imread.imread
• dask.array.image.imread

There is an old ongoing issue that lists potential issues with these imread functions + advocates to merge these two functions into one, but that doesn't seem to have happened.

dask_image.imread.imread

• seems to have no documentation of supported file formats
• underneath seems to use pims, so would be worth at least linking to their documentation?
• Trying it out locally, managed to load a multi-page tiff + dcm series:

from dask_image.imread import imread

img = imread("path/to/image.tif")
img = imread("path/to/dir_of_dicoms/*.dcm")

• The dicom series gives a warning, that is already reported on an issue
• Didn't seem to be able to read .nii files directly: errored with 'Could not autodetect how to load a file of type nii'

dask.array.image.imread

• In docs, says that skimage.io.imread is used as the default reader, but a custom reader can also be provided.
• Looking up the skimage.io.imread docs, this provides no info on supported file formats (except that jpg is supported 😅 )
• I managed to open a multi-page tiff, dcm series and nifti file locally:

from dask.array.image import imread

img = imread("path/to/image.tif")

img = imread("path/to/image.nii")

img = imread("path/to/dir_of_dicoms/*.dcm")

Bioio

Another potential option is bioio - this supports readers for a wide variety of file formats + you can retrieve a dask array with .dask_data

• Has extensive documentation of supported file formats, linking to the relevant external libraries where appropriate.
• Manged to read multi-page tiff locally, but had issues with nifti / dicom

from bioio import BioImage

img = BioImage("path/to/image.tif")
print(img.dask_data.shape)
print(img.dask_data.dtype)

• Nifti was loaded with BioImage, but failed on calling .dask_data: 'the itk plugin does not support meta data currently.'
• The dicom series errored - potentially you have to loop over individual images in the directory + load them.

[K-Meech]

What to do next?

• make an issue on scikit-image to expand the imread docs. I believe it is a thin wrapper around imageio but this should be stated, so people know what formats are supported.

• make an issue on dask_image to expand the dask_image.imread.imread docs. If it's using pims this should be mentioned + linked to, so it's known which formats are supported. Might also be worth linking to the other dask imread as an alternative option.

• Could look at swapping dask.array.image.imread from skimage.io to imageio, as discussed on the relevant issue. Not urgent though - skimage aren't planning to remove their thin wrappers around imageio any time soon.

• Anything else? I could try to fix some of the errors above, but depends on which file formats are most important... dask.array.image.imread seemed to handle these simple use cases fine.

[K-Meech]

Open issues:

• Document supported file formats for skimage.io scikit-image/scikit-image#7879
• Document supported file formats for dask_image.imread dask/dask-image#407