Registration of part-to-whole with same to same resolution

[CAguilarNeuro]

Describe the problem or limitation
Hi! I am a Ph.D. student in Thomas Frank's lab. We do 2P calcium imaging in larval zebrafish, and usually image a fish twice, once to obtain our "functional" stacks, usually 9 planes, and a second time to get an "anatomical" stack, around 200 planes.

Describe the solution you'd like
The idea would be to register the functional stacks onto the anatomical stacks, so a part-to-whole registration. Both of them are in the same resolution. If there is anything I could provide please let know.

Describe alternatives you've considered
We could use AntsPy and other libraries, but it takes too long.

[bjudkewitz]

Was about to answer. Did you close it because you got it to work?

[CAguilarNeuro]

Hi Benjamin, sorry! I closed it by accident. No, I haven't got it to work yet.

[bjudkewitz]

No problem. I transferred to a discussion, but I'm happy to convert it back to an issue if you feel it isn't resolved.

First I should point out that reference and moving volumes are allowed to have different array shapes (as of version 0.1.0; previously, the workaround was using the utility function warpfield.ndimage.zoom_chop_pad). Volumes are now only expected to be of roughly the same resolution. I just saw that there was a remnant misleading sentence in the readme about warpfield requiring same array shape, which I just fixed, thanks!

Even though part to whole registration is now technically possible, it can be more tricky to get satisfactory results. For example, edge effects can bias the cross-correlation and it can help to smoothen the edges of the smaller volume. Why don't you give it a try and post the outcome here?

[bjudkewitz]

@CAguilarNeuro should I leave this discussion open or has your question been addressed?

[CAguilarNeuro]

Hi Benjamin, I haven't had a chance to try it out but this week I'll sit down and do it and show you some of the results, sorry for the delay!