WIP: Mirror transforms

[dokato]

Introducing symmetrisation based on landmarks to FANCr.

Done:

• mirror_fanc
• added simplified mesh FANC.surf
• mirror_fanc_reglist
• symmetric_fanc

TBD:

• fill the parameters of FANCsym

[dokato]

About FANCsym template, I was tempted to just copy https://github.com/navis-org/navis-flybrains/blob/a8c24e44dd1a692661b0add93dc74829977a6105/flybrains/data/template_meta.json#L651 but we need a symmetrised version of these, right @jefferis ?

[jefferis]

Thanks a lot @dokato. So actually the symmetric FANC space defined by the registration is just the symmetric MANC space. So I think you could use the dimensions of that.

> dput(MANCsym)
structure(list(name = "mancsym-4-symmetric", regName = "MANCsym", 
    type = "Synthetic average brain from T-bar predictions based on MANC FIB sem data", 
    sex = "M", dims = c(672L, 830L, 1280L), voxdims = c(0.512, 
    0.512, 0.512), origin = c(0, 0, 0), BoundingBox = structure(c(0, 
    343.552, 0, 424.448, 0, 654.848), .Dim = 2:3, class = "boundingbox"), 
    units = NULL, description = NULL, doi = NULL), class = "templatebrain")

> MANCsym
=== Template Brain === 
Name: mancsym-4-symmetric 
Short Name: MANCsym 
Type: Synthetic average brain from T-bar predictions based on MANC FIB sem data 
Sex:  M 
Dimensions:672 x 830 x 1280 voxels
Voxel size:
  x = 0.512 
  y = 0.512 
  z = 0.512 
Bounding box ():
  x = 0, y = 0, z = 0,
  x = 343.552, y = 424.448, z = 654.848.
Description: 
DOI:

[dokato]

Are you sure @jefferis? All my landmark are in nm so my transformation expects nm too; the above is microns. When I plot FANC and MANC surfaces meshes in um I got sth like that ⬇️ . Do I miss some scaling factor?

[schlegelp]

If you are using the FANC_mirror_landmarks.csv from navis-flybrains then this is not a bridging but a mirror transform - i.e. it compensates for left right asymmetries after flipping within the FANC template space.

[dokato]

No, I get that and I use landmarks file just for symmetrization, above is supposed to show that I don't quite get why: "symmetric FANC space defined by the registration is just the symmetric MANC space" when they are dissimilar in um.

[jefferis]

Hi both, I guess I didn't look closely enough at what @dokato did. I gave @schlegelp a bridging registration from FANC (µm) -> MANCsym (µm). Philipp then used that to generate landmark pairs that would define FANC (nm) -> FANC-mirrored (nm). Now I guess what Dominik has done is use the LR to mirroring field to generate a symmetrised FANC template version. How did you do that in detail, Dominik? Is it actually affine aligned to the YZ plane, something we normally do as final step when making an image-based symmetric template?

[dokato]

@jefferis I did:

  mirror_reg_f <- tpsreg(
    mirror_landmarks[c("x_flip", "y_flip", "z_flip")],
    mirror_landmarks[c("x_mirr", "y_mirr", "z_mirr")]
  )

but TBH I got confused a little, I can see from these points that they just shift the voxels a little

I.e. they symmetrize the mesh in FANC nm space (at least what I thought it's doing) - hence why the name mirror landmarks or "FANC-mirrored (nm)" as I don't see any mirroring done?
Yet... as after symmetrisation (as done above) I imagined that I could use the same landmarks, invert the x axis to perform mirroring, does that sound like it?