ANTsPy: document using an existing transform as the initial transform

[muratmaga]

Documents the "Use existing transform" initial-transform mode, which the tutorial never covered, and the displacement field downsampling control added in SlicerMorph/SlicerANTsPy#35 (fixing SlicerANTsPy issues #33 and #34).

What is added

Advanced Topics H - Using an Existing Transform as the Initial Transform

• how to select the mode in the Pair-wise tab
• why the last node of a chain is the one to pick - FastModelAlign leaves <name>_scaling -> <name>_rigid -> <name>_deformable and only the leaf carries the complete alignment. Selecting a middle node initializes the registration with part of the alignment, which is the trap behind SlicerANTsPy#34
• how linear and non-linear transforms are serialized differently, and why a non-linear one is flattened onto the fixed image grid
• what the downsampling factor does and when to raise it, with the measured trade-off (a landmark warp at factor 4 reproduced the original to about 0.002 mm, far below a voxel, with a field 64 times smaller)
• that the inverse output is unavailable with a non-linear initial transform, and what is still correct in that case

Two troubleshooting entries: the Registration failed with error code 1 failure from SlicerANTsPy#33, including a manual workaround for anyone on an older build, and "aligns less than expected" pointing at the parent-chain trap.

Table of contents fix (pre-existing): it listed five steps while the body has seven, so three of its links pointed at headings that do not exist. All internal anchors now resolve.

Note on the duplicate

SlicerANTsPy also carries a copy of this tutorial at its repo root (ANTsRegistration_Tutorial.md). The two had drifted about 66 lines apart: this copy is the more current one - it has the ANTsPyRegistration module rename, the Extension Catalogue install instructions, and the ScriptEditor pointer, none of which the extension copy has.

So the new content was merged into this copy rather than the other way around, and the two files are now identical except for the parts where this one is ahead. A follow-up in SlicerANTsPy removes its stale duplicate and links here instead, so there is one tutorial to maintain.

[claude]

Claude finished @muratmaga's task in 1m 8s —— View job

---

Review: ANTsPy — document using an existing transform as the initial transform

• Gather context (diff, repo guidelines — no CLAUDE.md present)
• Read changed file in full
• Review for correctness, edge cases, regressions
• Post findings

Summary: This is a documentation-only change to ANTsPy/README.md (+51/-7). I verified every new/changed piece:

• ToC fix: all 11 anchor links (including the three renumbered ones for Steps 3–7) resolve to real headings in the file — checked with a full header listing.
• New anchor #h-using-an-existing-transform-as-the-initial-transform: matches GitHub's slug for ### H. Using an Existing Transform as the Initial Transform and is used correctly by the new troubleshooting entry's "See Advanced Topics H" link.
• Internal consistency: the FastModelAlign chain description (<name>_scaling → <name>_rigid → <name>_deformable) is stated the same way in both the new Advanced Topics H section and the "aligns less than expected" troubleshooting entry.
• Downsampling math checks out: a factor-2 reduction in a 3D field is 2³ = 8× smaller, factor-4 is 4³ = 64× smaller, matching the text.
• UI symbol usage (☑ for checkboxes, ◉ for radio buttons) is consistent with the rest of the document.

No correctness bugs, broken links, or regressions found. This is a low-risk, well-scoped docs PR — no issues to flag.