project_coordinates with thickness_type="normalized_layers" does not respect layer_thicknesses

[ablot]

The main issue is described in #9
Repeating here for dicussion, in case it's a feature that I didn't understand

If thickness_type is normalized_layers, a list of layer thicknesses (layer_thicknesses kwarg) must be provided to the projector and each layer is normalised to have this thickness across the whole cortex. However, depth is rescaled later to have the size of the deepest cortical layer in the volume. This seems confusing as one must know this size to the downscale again to the thicknesses that were provided.
The implementation of the micron seems also buggy as the normalisation happens twice (but I don't use it so I haven't tested).

[gouwens]

PR #9 that referenced this issue got closed, so maybe this isn't applicable anymore, but just wanted to check.

Both "normalized" options produce outputs where every part of cortex covers the same depth range (e.g. from 0 (pia) to 1 (white matter)). The "normalized_full" option doesn't care about layers - a coordinate that is found halfway between pia and white matter will end up with the same normalized depth value no matter which part of cortex it's in. The "normalized_layers" option does care about layers, so if a coordinate halfway between pia and white matter in one part of cortex lands in layer 4, while in another part it lands in layer 5, they will end up with different normalized depth values. The ratios of layer thicknesses in the provided parameter determine how much of the normalized depth each layer takes up (and that proportion ends up the same everywhere).

I can try to update the documentation to make that clearer. I'm not seeing a double normalization with the microns side, but let me know if you're seeing something else. Thanks!

[ablot]

I think you can close but the doc might need clarification on the meaning of voxels. If thickness_type="normalized_layers", the output "voxels" means rows of the normalized-layers slab (paths.shape[1] from pia to white matter), not 10 µm CCF voxels. And that value is independent of whatever I put in IsocortexCoordinateProjector.layer_thicknesses.

I was expecting voxels to be the values in microns divided by the atlas size, but that's not true in this specific case.

[gouwens]

That makes sense, and explanation is very helpful. I'll work on clarifying in the documentation (and potentially rename the parameter/option). Thank you!