View OME-NGFF collection of Images

We are making progress with defining a specification for a collection of Images at https://github.com/ome/ngff/issues/31 and with generating a collection of OME-NGFF Images (export from OMERO) at https://github.com/ome/omero-cli-zarr/pull/83 so I'm starting to think how we could visualise/browse these. Although there has been discussion of grouping subsets of a collection by various criteria, in the simplest case we'd just want to be able to see a thumbnail and or name for each image in the collection.

In some ways this is similar to the HCS view, but it is more complex because we have to handle the case where all the images have different number of channels, different pixel type and rendering settings. There will be some image collections where you might want the HCS-type view, with the same adjustable rendering settings applied to all images. I don't know if we'd want to try and handle this use-case or just assume that all collections are un-related images.

I don't know if this should even be built-in to the existing vizarr viewer, or if this should be an entirely different app that could link to vizarr viewer when an image is clicked?

This will be really useful! I have a HCS dataset we published and was wondering exactly the feature you are describing. Have some other use cases too, we want to build a viewer for machine learning datasets, which would also fall into the collection case.

@oeway Viewing of a HCS Plate of images in OME-NGFF HCS layout is already supported - See https://www.openmicroscopy.org/2020/12/01/zarr-hcs.html for a few original examples and https://hms-dbmi.github.io/vizarr/?source=https://uk1s3.embassy.ebi.ac.uk/idr/zarr/v0.3/idr0094A/7751.zarr to view a more recent version. Clicking on the Well thumbnails opens the Well images in a new Tab.

Nice! This is the dataset I mentioned earlier https://www.covid19dataportal.se/highlights/immunofluorescence/

Do you think we can use this HCS format?

Yes, and I would expect bioformats2raw should be able to convert each plate.ome.tiff into an NGFF Plate? cc @sbesson

In general yes bioformats2raw should work on any HCS imaging file formats supported by Bio-Formats and preserve the HCS layout (rows, columns, wells, fields of views) in the converted OME-NGFF representation. Input format support alos includes OME-TIFF as long as the metadata has appropriate Plate hierarchy.

Note the currently 0.4.0 version of bioformats2raw still writes Zarr datasets according the OME-NGFF 0.2 specification. The upcoming 0.5.0 release should update this to use the 0.4 specification, with improvements/clarifications on the HCS specification. A release candidate was just made available yesterday in case you are interested in testing /cc @melissalinkert

In general yes bioformats2raw should work on any HCS imaging file formats supported by Bio-Formats and preserve the HCS layout (rows, columns, wells, fields of views) in the converted OME-NGFF representation. Input format support alos includes OME-TIFF as long as the metadata has appropriate Plate hierarchy.

Note the currently 0.4.0 version of bioformats2raw still writes Zarr datasets according the OME-NGFF 0.2 specification. The upcoming 0.5.0 release should update this to use the 0.4 specification, with improvements/clarifications on the HCS specification. A release candidate was just made available yesterday in case you are interested in testing /cc @melissalinkert

Thank you for the information, I wonder how you handle irregular HCS collections, e.g. for each well, we may have different number of field of views and maybe with different size, would that work? Try to see the limitations here.

Thank you for the information, I wonder how you handle irregular HCS collections, e.g. for each well, we may have different number of field of views and maybe with different size, would that work? Try to see the limitations here.

There should be no limitation to both scenarios, at least at the specification level. Implementations might choose to be more restrictive in their support. The number of fields of views for each well is defined at the well metadata level and the dimensions of each field of views are defined at the level of the datasets for each multiscales group.

I think that vizarr GridView for a Plate is restricted to all Wells being the same dimensions. And I think that the Plate view in napari also assumes that all Wells are the same size.