3D image support

[pr4deepr]

Hi
Thanks for making this package available publicly. It is really powerful.
Is there any plan to support feature extraction for 3D images in the near future?

Cheers
Pradeep

[sameeul]

Hi, yes we are currently actively working adding 3D image support. See #65 . The plan is to get this done in the next 2-3 months to get something out there. One key challenge we see is the lack to data to test things out. There are many ways to represent 3D images. For example, a stack of 2D images in a directory, a zarr file using OME NGFF spec, a single Tiff file with OME TIFF metadata . Currently, we are planning to support only the first one (2D images in a directory) and then gradually add the rest. Please, feel free to share what 3D features you are interested in :-)

[pr4deepr]

Thanks @sameeul.

I think within the python API, if you allow the flexibility of passing a 3D array and process it so that 3D features are extracted, that would be great. Currently, it extracts 2D features for every slice.
With reading files in a directory, is there the flexibility for the user to define a custom image reader?

For datasets, I can see an Arabidopsis one here: https://bioimage.io/#/?tags=3d&type=dataset&id=ilastik%2Fplantseg_ovules
Within the Broad Bioimage benchmark challenge, you can find some datasets here if you search for 3D: https://bbbc.broadinstitute.org/search/3d?
Within this resource, you can see many datasets are organized as tiff stacks, instead of a 2D slice in a directory. It would be worth supporting this format.

To be honest, with features, I am interested in all the features currently listed on the nyxus website. Would be great to have as many as possible. Cellprofiler measurements are usually a good reference, which you probably know.
I'm definitely interested in object intensity distribution, i.e. radial features.
Another option is extracting features after application of a convolutional filter, such as a gaussian filter. Is this something of interest?

[sameeul]

@pr4deepr : Thanks for the inputs. We will definitely keep these in mind as we keep developing the features.

[pr4deepr]

Tagging @haesleinhuepf for ideas