Feature/niftizarr

[MikeSchutzman]

Created Nifti Zarr file

Used Volume Logic from Zarr but modelTransform logic from Nifti

[balbasty]

Cool!

Did you check that when you load the demo file with the niftizarr:// protocol on top of this state, both layers are aligned?

[aaronkanzer]

@balbasty @MikeSchutzman could you share a valid, converted nifti-zarr? I just wanted to stress-test this on my own locally with a valid asset -- thanks!

[balbasty]

@aaronkanzer See the last post in the brainhack issue: brainhack-boston/brainhack-boston.github.io#50

There are links to two nifti-zarr assets (the first one should be used, as it has a single channel), and links to two neuroglancer scenes: one that just uses the default zarr reader, and one where I manually applied the correct "nifti" transform on top of it. The asset read with the niftizarr protocol should aligned with the second scene.

[aaronkanzer]

@MikeSchutzman were you able to confirm this working on your local dev environment? e.g. in my local dev server, there is no data source for niftizarr://

There seems to be additional setup/files that neuroglancer is expecting in order to fully facilitate an end-to-end support for a file type -- e.g. see here for how zarr is supported: https://github.com/google/neuroglancer/blob/master/package.json#L356-L373 -- thus, I'm curious how you were able to validate the code here successfully

Wanted to double-check first with you with these details to ensure that I'm not missing something in regards to setup...

[aaronkanzer]

@MikeSchutzman could you also merge main back in here when you have the chance? The maintainers of neuroglancer recently changed their frontend buildpack from vite to webpack, so wanted to ensure we don't miss any inconsistencies with that -- thanks again for your work here!

[aaronkanzer]

@MikeSchutzman @balbasty @kabilar Just another thought here, as it seems that there is a decent amount of redundancy of code here vs. the zarr equivalent in https://github.com/lincbrain/neuroglancer/tree/master/src/datasource/zarr

Where are we validating confirmation that pixel-to-pixel (e.g. what we see is truly accurate) with the nifti -> zarr conversion? My estimation would be that we need to confirm this on both the computational side, but also, as it pertains to this PR, in a rendered environment.

Let me know your thoughts -- happy to brainstorm here, but without proper validation, this presents risk

[MikeSchutzman]

@MikeSchutzman were you able to confirm this working on your local dev environment? e.g. in my local dev server, there is no data source for niftizarr://

There seems to be additional setup/files that neuroglancer is expecting in order to fully facilitate an end-to-end support for a file type -- e.g. see here for how zarr is supported: https://github.com/google/neuroglancer/blob/master/package.json#L356-L373 -- thus, I'm curious how you were able to validate the code here successfully

Wanted to double-check first with you with these details to ensure that I'm not missing something in regards to setup...

haven't tested yet. once i test, will change pr from draft to published

[MikeSchutzman]

@MikeSchutzman @balbasty @kabilar Just another thought here, as it seems that there is a decent amount of redundancy of code here vs. the zarr equivalent in https://github.com/lincbrain/neuroglancer/tree/master/src/datasource/zarr

Where are we validating confirmation that pixel-to-pixel (e.g. what we see is truly accurate) with the nifti -> zarr conversion? My estimation would be that we need to confirm this on both the computational side, but also, as it pertains to this PR, in a rendered environment.

Let me know your thoughts -- happy to brainstorm here, but without proper validation, this presents risk

yes great point... not sure how to confirm this is actually working logically as it is supposed to

[MikeSchutzman]

just to update with status:
the issue lines are caused by these 2 statements:
staticAnnotations: makeDataBoundsBoundingBoxAnnotationSet(box),
It wants volumeZarr.modelSpace.bounds,

and

modelTransform: {
            sourceRank: info.rank,
            rank: info.rank,
            inputSpace,
            outputSpace,
            transform: info.transform,
          },

it wants modelTransform: makeIdentityTransform(volumeNifti.modelSpace),

[MikeSchutzman]

still not sure how to make this work. Been playing around with it. It breaks when i use getNiftiVolumeInfo() instead of MultiscaleVolumeChunkSource but that's the entire point of this combo niftizarr file...

[MikeSchutzman]

@aaronkanzer @balbasty any ideas on last 2 comments?

[MikeSchutzman]

i didn't create a combo niftizarr backend that's the issue...

[MikeSchutzman]

still not sure how to make this work. Been playing around with it. It breaks when i use getNiftiVolumeInfo() instead of MultiscaleVolumeChunkSource but that's the entire point of this combo niftizarr file...

update here - it's the call to this that breaks it. Originally, i thought the issue was saving this result to a var and passing that var into later transform broke it but this was wrong. Just calling this function breaks the code, even if i pass the zarr info into the transform later on

[MikeSchutzman]

brainhack-boston/brainhack-boston.github.io#50