Stop reading VisuCorePosition's leading size as a slice count - #210
Merged
headmeister merged 3 commits intoAug 10, 2026
Merged
Conversation
`ruff format --check` reported twelve files as unformatted, so any change touching one of them dragged unrelated reflow into its diff. Run the formatter once, on its own, so subsequent changes show only what they actually change. Formatting only: no behaviour changes, and `ruff check` is clean before and after. The suite is unchanged at 2140 passed, 12 skipped. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_017SNrm3jFhpTPShP8xGkePC
The extent recipe built its third component from abs(VisuCorePosition[1,2] - VisuCorePosition[0,2]) -- the patient-z component of the step between two slice centres -- and multiplied it by the number of positions plus one. Spec 7.2 makes VisuCorePosition a 3-vector in patient coordinates, and spec 5.4/12 let the slice normal point anywhere: ACQ_slice_orient ranges over sagittal, coronal and Arbitrary_Oblique, and the normal is the third row of VisuCoreOrientation. Projecting onto z alone therefore reports exactly zero for every coronal, sagittal or oblique stack, and the +1 overstates the axial ones by one slice. Across the pv5, pv6, pv7 and new-orientation corpora all 77 multi-slice 2-D reconstructions were wrong: 14 reported 0.0, the other 63 were long by one slice thickness. Both now report n * step, the length of the whole difference -- which is what the sibling resolution recipe and Dataset.slice_distance already use, and which keeps extent = size * resolution as it is in plane. The corpus property references move with it, 54 extent values in four files. Closes isi-nmr#194 Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_017SNrm3jFhpTPShP8xGkePC
is_single_slice, extent and resolution branched on #VisuCorePosition.size[0] > 1 as though it were the number of slices, then took position[1] - position[0] as the slice step. Spec 7.4's cardinality rule forbids that: a parameter that is not tied to a frame group may carry no value, exactly one value, or VisuCoreFrameCount values, and "a reader must not assume 'one per slice' for any of these". An FG_ISA parameter-map PROCNO writes one position per map, all identical, and is not in VisuGroupDepVals at all -- so the >1 branch fired and the "slice step" was the difference between two copies of one position, exactly zero. Ten reconstructions in the corpus reported a slice resolution of 0.0. What makes a dataset multi-slice is the positions differing, so test that instead, and let extent and resolution follow is_single_slice rather than re-deriving it. VisuCoreFrameThickness is read per frame for the same reason: it is written once per frame on these datasets, and np.array([e0, e1, <six values>]) does not build a vector. The ten affected reconstructions now report the recorded 0.8 or 1.0 mm slice thickness, and their maps sit on the singleton slice axis every other single-slice dataset already has. No other reconstruction of the 514 in the corpus changes. The d3proc fixture in test_latent_conformance gave three FG_SLICE frames one repeated position, which no ParaVision file does; it now steps them. Closes isi-nmr#195 Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_017SNrm3jFhpTPShP8xGkePC
gdevenyi
force-pushed
the
fix/visucore-position-cardinality
branch
from
August 8, 2026 20:51
b1b4e79 to
a51ac1c
Compare
Contributor
Author
|
Rebased onto #213 ( Integration check across all eleven conformance PRs merged together: 2164 passed, 12 skipped; corpus load test 3202 → 3207 of 3478 with zero newly broken datasets. |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Fixes #195.
Spec 7.4:
is_single_slice,extentandresolutionall branched on#VisuCorePosition.size[0] > 1as though it were a slice count, then tookposition[1] - position[0]as the step. An FG_ISA parameter-map PROCNO writes one position per map, all identical — so the step was the difference between two copies of one position, exactly zero.Change
is_single_slicetests whether the positions actually differ, not how many there are.extentandresolutionfollowis_single_sliceinstead of re-deriving it.VisuCoreFrameThicknessis read per frame — it is written once per frame on these datasets, andnp.array([e0, e1, <six values>])does not build a vector.Ten reconstructions change, all of them the FG_ISA parameter maps; they now report the recorded 0.8 or 1.0 mm thickness and their maps sit on the singleton slice axis every other single-slice dataset already has. No other reconstruction of the 514 in the corpus changes.
The
d3procfixture intest_latent_conformancegave threeFG_SLICEframes one repeated position, which no ParaVision file does; it now steps them.Test
test_repeated_positions_are_one_slice_not_a_stack— a synthetic six-map FG_ISA PROCNO with repeated positions and per-frame thickness. Fails on master.Suite: 2144 passed, 12 skipped.