Draft: proof-of-concept for streamline to precomputed annotation, include CORS server for rendering in local neuroglancer

[aaronkanzer]

@balbasty @ayendiki -- starting a branch here for work on trk to precomputed annotation conversion -- Note: I am testing this against google's current neuroglancer, not our fork

To replicate locally/work with if you'd like:

python -m venv venv
source venv/bin/activate
pip install -r requirements.txt
python trk_to_annotation.py

This will produce a few thousand small (1KB-ish) coordinates with the JSON properties necesssary for rendering in neuroglancer -- my current obstacles is getting the precomputed annotation layer, and the nifti layer (sub-I58_sample-hemi_desc-preproc_dwi_FA.nii.gz) in this case to be in the same voxel space

Will keep you posted as things progress -- wanted to get code on GitHub to start the conversation

[aaronkanzer]

@balbasty -- a couple mini updates here with the script (things I am trying to tackle)

• It seems that the NIFTI and the precomputed annotations are still offset in neuroglancer -- e.g. they dont overlap when added as layers --

backend.ts:456 Error retrieving chunk [object Object]:3,3,3: RangeError: Offset is outside the bounds of the DataView

• It seems that the empty bytes are failing to be recognized by the precomputed backend for neuroglancer (whether or not we should have "empty" bytes -- I'm not sure?)

backend.ts:891 Uint8Array(8) [0, 0, 0, 0, 0, 0, 0, 0, buffer: ArrayBuffer(8), byteLength: 8, byteOffset: 0, length: 8, Symbol(Symbol.toStringTag): 'Uint8Array']
backend.ts:456 Error retrieving chunk [object Object]:5,5,2: Error: Expected at least 8 bytes

[balbasty]

I think you want the tracts coordinate to be in mm RAS, not in voxels. So don't apply the inverse affine, set the axes as 1 mm (not voxel_size) and compute the upper and lower bounds in RAS mm space. There's an example in the Google doc.

[aaronkanzer]

I think you want the tracts coordinate to be in mm RAS, not in voxels. So don't apply the inverse affine, set the axes as 1 mm (not voxel_size) and compute the upper and lower bounds in RAS mm space. There's an example in the Google doc.

@balbasty thanks for the guidance -- just so I fully understand, you mean this Google doc? https://docs.google.com/document/d/1TnDNvDigOxpNSuZBjHkve_epp7tbpswcxUUcBOaRV94/edit?tab=t.0

If so, do you mean:

 "dimensions": {"x": [1, "mm"], "y": [1, "mm"], "z": [1, "mm"]},
  "lower_bound": vertices.min(axis=0).tolist(), 
  "upper_bound": vertices.max(axis=0).tolist(), 

in your info JSON example?

[balbasty]

Yes! Although it should really be a "double loop" minimum/maximum (over tracts and edge per tract)

[aaronkanzer]

Yes! Although it should really be a "double loop" minimum/maximum (over tracts and edge per tract)

What exactly is accomplished in the "double loop" that wouldn't be accomplished via a "single loop"? I tried to search online to see why, but couldn't really make 100% sense.