DICOM extract cleanup, spinestats v3 (endplate Cobb), point registration, stitching refactor by robert-graf · Pull Request #160 · Hendrik-code/TPTBox

robert-graf · 2026-10-28T18:51:15Z

Summary

Supersedes #154. That branch's scope grew well beyond the Cobb endplate work and had conflicts with main; this PR carries all of #154's commits on a fresh merge of main, plus the DICOM / stitching / registration work that accumulated alongside.

Grouped highlights

DICOM: 2D support, more 4D edge-case handling, cleanup of the extraction pipeline.
spinestats v3 (originally spinestats: version 3 — endplate-plane Cobb + apex + tightened multi-Cobb #154): endplate-plane Cobb + apex + tightened multi-Cobb; S1 endplate; Veridah lordosis; skip T13 for apex; skip missing plane/source POIs; wider sacrum-endplate crop; bumped `run_all` defaults.
POI: auto-transform of direction-vector fields in `POI.info`; `label_name` integrated into `map_labels` remap.
Registration: Deepali point registration + cleanup.
Rib assign: `_rib_assign` bug fixes (NII `&` TypeError on bool-backed NIIs; misplaced bracket in `extract_label`); tests added.
Stitching: refactored to use TPTBox functions directly; smaller overlap NII to save RAM; `c_val` option; `min_value` fix.
Internal: smauglab support in the internal trainer; parallel prewarm of `get_grid_info`.

38 commits ahead of main; +2277 / -174 across 23 tracked file groups.

Test plan

`pytest unit_tests/` passes locally
CI green on 3.9 and 3.10+ legs
`ruff check .` / `ruff format --check .` clean
Spot-check Cobb endplate output on a known case
Spot-check DICOM 2D extraction on a sample
Spot-check stitching on a multi-station case

Adds a small registration mechanism so producers can attach per-vertebra direction vectors (endplate PCA normals, orientation vectors, ...) or label-keyed scalars (endplate_internal_angle, curvature_*) to poi.info and have them stay aligned with the POI points automatically: - New TPTBox/core/poi_fun/vector_fields.py with the constants POI_INFO_VECTOR_FIELDS_KEY / POI_INFO_LABEL_KEYED_FIELDS_KEY and three helpers (_transform_direction_vectors_inplace, _rotate_direction_vectors_inplace, _remap_vector_field_keys_inplace). - POI.reorient signed-permutes vectors; POI.resample_from_to applies the real R_ref.T @ R_self rotation; POI.rescale is a no-op. - POI_Global.to_cord_system flips vector components on RAS<->LPS. - POI.map_labels remaps top-level keys of both registries. - endplates.py registers angle_superior/inferior_endplate, endplate_internal_angle, and curvature_* when they're produced. - Fix: vertebra_pois_non_centroids.py now (re)runs calc_endplate_points_ when S1 is present but its Sacrum_Endplate- derived superior-endplate landmark is missing, so the sacrum endplate direction actually lands in the buffer. - Angles: _endplate_ap_direction preferred over Vertebra_Direction_ Posterior for MoveTo.TOP/BOTTOM in _get_norm, yielding endplate- aligned Cobb-style lordosis/kyphosis lines. Docs: new section in docs/api/poi_fun.md. Tests: unit_tests/test_poi_vector_fields.py (13 tests). Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

Merges the poi-label-names branch (LABEL_NAME + nested {region: {subregion: name, "name": group}} format in info["label_name"], new label_name / set_label_name / level_one_name / set_level_one_name accessors, migration helpers, and level_*_info propagation in copy / _set_inplace / calc_poi_average / POI_Global.to_other). Then wires label_name into the auto-transform mapping scheme: - New vector_fields._remap_label_name_inplace handles the nested format: remaps top-level region keys via label_map_region and inner subregion keys via label_map_subregion, preserving the "name" group entry. Colliding regions merge with last-write-wins on inner keys. - map_labels calls both _remap_vector_field_keys_inplace (for the registered flat vector / label-keyed fields) and the new _remap_label_name_inplace on all three return paths. - Also triggers the legacy flat-format migration on remap. Docs updated with a "Special-cased field" section for label_name. Tests: 5 new cases in Test_LabelName_MapLabels covering region-only, subregion-only, combined, region collision, and legacy-format migration paths (18/18 passing). Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

…Cobb Extends the endplate-plane direction machinery used for v2 lordosis/kyphosis to the coronal Cobb angle, and bumps ``CURRENT_VERSION`` to 3 so cached stat files recompute the angle keys (POI buffers stay valid — no new landmarks). Angles (``TPTBox/spine/spinestats/angles.py``): - ``_get_norm`` now routes ``Vertebra_Direction_Right`` through a new ``_endplate_r_direction`` helper (analogue of ``_endplate_ap_direction``), so Cobb lines are true endplate lines averaged across the shared disc instead of the two flanking vertebrae disagreeing on the same boundary. - ``plot_cobb_angle`` always uses that endplate right direction — the ``use_ivd_direction`` branch used ``Vertebra_Disc_Inferior`` from each vertebra separately, giving two different lines at one anatomic disc. - ``compute_angel_between_two_points_`` orders ``id1``/``id2`` by ``Vertebra_Instance.order_dict()`` rather than by numeric label, so T13 (label 28, anatomically between T12 and L1) pairs to the right endplates and ``MoveTo.TOP``/``BOTTOM`` are not swapped. - ``compute_max_cobb_angle_multi`` uses an exclusive split (``+ 1`` on the ``below`` slice) so sibling curves cannot share an endpoint vertebra. - ``VERT_START_COBB`` is C3 (C2 excluded from Cobb search). - ``compute_lordosis_and_kyphosis`` emits ``{curve}_apex`` keys, computed by the new ``_find_curve_apex`` helper (same bisector heuristic as ``compute_max_cobb_angle``). ``plot_compute_lordosis_and_kyphosis`` renders a ``*apex`` marker at that vertebra. - ``plot_cobb_angle`` anchors the label at the apex disc and widens the coronal pad from 50 to 80 mm so labels don't get clipped. Runner (``TPTBox/spine/spinestats/_run_all.py``): - Introduces ``POI_INVALIDATING_VERSION = 2`` so the ``_poi_is_stale_wrt_stat`` threshold is a named constant rather than a magic ``2``, and its docstring points at a symbol that actually exists (previously referenced a nonexistent ``ANGLES_VERSION``). - Documents the v3 bump in the ``CURRENT_VERSION`` history comment. Veridah (``TPTBox/spine/spinestats/veridah_angles.py``): - ``_compute_k_variant`` also returns ``lumbar_lordosis_apex`` so the per-k variants carry the new apex key. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

…cobb-endplate-v3

Manual fixes in ``TPTBox/spine/spinestats/_load_nako_wh.py``: - D205 (verify_missing_images, verify_hardlink): reword docstrings so the summary line stands alone from the description body. - PLW2901 (``v = preferred`` at line 787): add ``# noqa: PLW2901`` to match the existing noqa on the sibling ``k = mapping.get(k, k)`` rebinding — the loop deliberately overrides both bindings. - PERF102 (line 1385): iterate the ``t2w_chunk`` mapping via ``.values()`` since only the value is used. - TRY300 (``_grid_worker``): move the success ``return`` out of the ``try`` block; only ``_add_grid_info_to_json`` may raise. Everything else is ``ruff check --fix`` and ``ruff format`` output. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

…bb-endplate-v3

…tching-refactor # Conflicts: # TPTBox/stitching/stitching.py

…all defaults Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

…robert-develop

…ct-cleanup

github-actions · 2026-10-28T18:55:45Z

Benchmark comparison

Speed (wall time per call)

baseline 9c15f60 vs head 9c15f60 · python 3.11.17 · 5 repeats + 1 warmup

Showing the 5 most-changed measurements per case; the rest are collapsed. Rows with a baseline below 3 ms are tagged (noise) — runner jitter on those swamps any real change.

`ct_3d` — shape (73, 47, 73)

Measurement	baseline ms (median ±½·range)	head ms (median ±½·range)	Δ %	p
`nii_rescale_seg`	3.71 ±0.08	3.23 ±0.13	-13.1%	0.001
`nii_save`	12.14 ±0.12	11.50 ±0.25	-5.2%	0.012
`poi_calc_centroids_nocrop`	4.33 ±0.08	4.16 ±0.10	-3.9%	0.122
`nii_erode_msk`	6.30 ±0.08	6.53 ±0.19	+3.6%	0.024
`nii_resample_from_to`	44.09 ±0.23	42.87 ±0.25	-2.8%	0.000

… 31 more measurements

Measurement	baseline ms (median ±½·range)	head ms (median ±½·range)	Δ %	p
`nii_fill_holes`	4.52 ±0.08	4.60 ±0.06	+1.8%	0.121
`nii_rescale`	41.47 ±0.29	40.84 ±0.17	-1.5%	0.001
`nii_dilate_msk`	122.90 ±0.34	124.74 ±0.90	+1.5%	0.005
`poi_calc_poi_from_subreg_vert`	12.86 ±0.16	12.96 ±0.31	+0.8%	0.343
`nii_filter_connected_components`	4.36 ±0.12	4.38 ±0.10	+0.6%	0.727
`nii_load_img`	10.79 ±0.23	10.78 ±0.13	-0.1%	0.578
`metric_voxels`	250463.00 ±0.00	250463.00 ±0.00	+0.0%	—
`metric_labels`	3.00 ±0.00	3.00 ±0.00	+0.0%	—
`metric_foreground_pct`	14.51 ±0.00	14.51 ±0.00	+0.0%	—
`nii_map_labels`	0.98 ±0.06	1.20 ±0.12	+21.6% (noise)	0.004
`nii_extract_label`	0.34 ±0.05	0.40 ±0.05	+18.0% (noise)	0.140
`poi_reorient`	0.39 ±0.02	0.44 ±0.03	+13.6% (noise)	0.018
`nii_apply_crop`	0.73 ±0.06	0.64 ±0.08	-11.8% (noise)	0.430
`nii_load_seg`	2.94 ±0.16	2.64 ±0.07	-10.4% (noise)	0.005
`nii_get_array`	0.05 ±0.01	0.05 ±0.01	-10.2% (noise)	0.648
`poi_local_to_global_arr`	0.31 ±0.02	0.28 ±0.02	-9.9% (noise)	0.076
`nii_unique`	0.67 ±0.04	0.73 ±0.04	+9.2% (noise)	0.031
`poi_load`	1.26 ±0.11	1.16 ±0.10	-8.6% (noise)	0.051
`nii_pad_to`	0.63 ±0.07	0.69 ±0.06	+8.4% (noise)	0.303
`poi_save`	0.33 ±0.03	0.35 ±0.04	+5.4% (noise)	0.319
`nii_volumes`	1.93 ±0.08	2.02 ±0.04	+4.7% (noise)	0.052
`poi_map_labels`	0.15 ±0.02	0.16 ±0.02	+4.1% (noise)	0.306
`nii_get_connected_components`	2.29 ±0.06	2.37 ±0.13	+3.7% (noise)	0.084
`nii_reorient`	0.87 ±0.08	0.91 ±0.06	+3.7% (noise)	0.144
`poi_calc_centroids`	2.53 ±0.09	2.44 ±0.11	-3.4% (noise)	0.139
`nii_center_of_masses`	1.91 ±0.08	1.96 ±0.10	+2.8% (noise)	0.127
`poi_to_global`	0.17 ±0.02	0.17 ±0.03	-1.6% (noise)	0.815
`nii_set_dtype`	0.46 ±0.05	0.47 ±0.05	+1.5% (noise)	0.866
`nii_compute_crop`	0.35 ±0.03	0.36 ±0.02	+0.5% (noise)	0.572
`poi_rescale`	0.19 ±0.03	0.19 ±0.03	+0.3% (noise)	0.896
`poi_resample_from_to`	0.68 ±0.04	0.69 ±0.03	+0.2% (noise)	0.677

`ct_2d` — shape (73, 47, 1)

Measurement	baseline ms (median ±½·range)	head ms (median ±½·range)	Δ %	p
`nii_rescale`	6.01 ±0.14	5.80 ±0.05	-3.4%	0.007
`nii_resample_from_to`	7.90 ±0.10	7.66 ±0.13	-2.9%	0.021
`nii_dilate_msk`	14.81 ±0.10	14.90 ±0.19	+0.6%	0.439
`metric_voxels`	3431.00 ±0.00	3431.00 ±0.00	+0.0%	—
`metric_labels`	3.00 ±0.00	3.00 ±0.00	+0.0%	—

… 30 more measurements

Measurement	baseline ms (median ±½·range)	head ms (median ±½·range)	Δ %	p
`metric_foreground_pct`	37.60 ±0.00	37.60 ±0.00	+0.0%	—
`nii_extract_label`	0.48 ±0.03	0.55 ±0.04	+14.7% (noise)	0.006
`nii_get_connected_components`	0.73 ±0.04	0.66 ±0.05	-10.7% (noise)	0.041
`nii_pad_to`	0.60 ±0.04	0.66 ±0.10	+10.4% (noise)	0.079
`nii_unique`	0.13 ±0.00	0.14 ±0.00	+10.1% (noise)	0.001
`nii_map_labels`	0.51 ±0.02	0.55 ±0.05	+8.1% (noise)	0.070
`poi_save`	0.50 ±0.03	0.46 ±0.02	-8.0% (noise)	0.122
`nii_get_array`	0.02 ±0.01	0.02 ±0.00	-7.9% (noise)	0.506
`nii_fill_holes`	1.01 ±0.07	0.93 ±0.05	-7.6% (noise)	0.165
`poi_reorient`	0.49 ±0.02	0.47 ±0.01	-4.8% (noise)	0.079
`poi_load`	1.52 ±0.06	1.45 ±0.03	-4.6% (noise)	0.078
`nii_save`	1.52 ±0.07	1.58 ±0.09	+3.8% (noise)	0.162
`nii_filter_connected_components`	0.94 ±0.04	0.91 ±0.02	-3.7% (noise)	0.033
`nii_erode_msk`	1.13 ±0.03	1.09 ±0.07	-3.6% (noise)	0.597
`nii_rescale_seg`	0.89 ±0.02	0.86 ±0.04	-3.3% (noise)	0.083
`nii_reorient`	0.90 ±0.11	0.92 ±0.04	+3.0% (noise)	0.359
`nii_compute_crop`	0.16 ±0.02	0.16 ±0.01	-2.8% (noise)	0.604
`poi_resample_from_to`	0.83 ±0.02	0.81 ±0.03	-2.5% (noise)	0.138
`nii_load_img`	1.78 ±0.03	1.74 ±0.07	-2.4% (noise)	0.301
`poi_rescale`	0.28 ±0.01	0.28 ±0.02	-2.4% (noise)	0.297
`nii_set_dtype`	0.47 ±0.02	0.45 ±0.03	-2.3% (noise)	0.202
`nii_volumes`	0.34 ±0.03	0.33 ±0.02	-1.9% (noise)	0.658
`poi_calc_centroids`	1.03 ±0.03	1.04 ±0.04	+1.3% (noise)	0.643
`poi_local_to_global_arr`	0.42 ±0.02	0.42 ±0.01	+1.2% (noise)	0.270
`poi_map_labels`	0.23 ±0.02	0.23 ±0.04	+1.0% (noise)	0.289
`poi_to_global`	0.24 ±0.02	0.24 ±0.01	-0.9% (noise)	0.217
`nii_center_of_masses`	0.25 ±0.01	0.25 ±0.01	+0.9% (noise)	0.330
`poi_calc_centroids_nocrop`	1.29 ±0.05	1.29 ±0.07	-0.2% (noise)	0.380
`nii_load_seg`	1.84 ±0.05	1.84 ±0.03	+0.2% (noise)	0.821
`nii_apply_crop`	0.79 ±0.06	0.79 ±0.04	+0.1% (noise)	0.650

`mri_3d` — shape (68, 52, 67)

Measurement	baseline ms (median ±½·range)	head ms (median ±½·range)	Δ %	p
`poi_load`	3.07 ±0.26	2.75 ±0.07	-10.6%	0.021
`poi_calc_centroids`	3.02 ±0.08	2.85 ±0.11	-5.6%	0.046
`nii_filter_connected_components`	5.19 ±0.07	5.02 ±0.16	-3.3%	0.048
`nii_rescale_seg`	3.73 ±0.08	3.62 ±0.06	-3.1%	0.020
`poi_calc_centroids_nocrop`	9.05 ±0.30	8.79 ±0.40	-2.9%	0.609

… 31 more measurements

Measurement	baseline ms (median ±½·range)	head ms (median ±½·range)	Δ %	p
`nii_fill_holes`	5.77 ±0.14	5.61 ±0.09	-2.9%	0.058
`poi_calc_poi_from_subreg_vert`	12.46 ±0.17	12.19 ±0.25	-2.2%	0.164
`nii_resample_from_to`	41.40 ±0.38	40.63 ±0.32	-1.9%	0.030
`nii_save`	9.78 ±0.15	9.61 ±0.04	-1.7%	0.029
`nii_rescale`	38.24 ±0.25	38.67 ±0.33	+1.1%	0.146
`nii_erode_msk`	8.07 ±0.13	7.98 ±0.13	-1.1%	0.102
`nii_load_img`	9.18 ±0.15	9.15 ±0.25	-0.3%	0.258
`nii_load_seg`	3.23 ±0.06	3.23 ±0.16	-0.2%	0.480
`nii_dilate_msk`	160.62 ±0.47	160.41 ±0.21	-0.1%	0.607
`metric_voxels`	236912.00 ±0.00	236912.00 ±0.00	+0.0%	—
`metric_labels`	8.00 ±0.00	8.00 ±0.00	+0.0%	—
`metric_foreground_pct`	18.21 ±0.00	18.21 ±0.00	+0.0%	—
`nii_apply_crop`	0.83 ±0.06	0.74 ±0.03	-11.4% (noise)	0.069
`poi_local_to_global_arr`	0.43 ±0.06	0.39 ±0.03	-10.5% (noise)	0.196
`poi_to_global`	0.31 ±0.02	0.28 ±0.02	-9.0% (noise)	0.050
`nii_pad_to`	0.83 ±0.02	0.76 ±0.05	-7.5% (noise)	0.212
`nii_extract_label`	0.52 ±0.02	0.48 ±0.04	-7.4% (noise)	0.090
`nii_get_connected_components`	2.64 ±0.05	2.45 ±0.04	-7.2% (noise)	0.000
`nii_reorient`	0.91 ±0.04	0.98 ±0.03	+7.1% (noise)	0.143
`poi_resample_from_to`	0.88 ±0.02	0.94 ±0.11	+7.1% (noise)	0.162
`poi_rescale`	0.31 ±0.01	0.33 ±0.02	+6.3% (noise)	0.209
`nii_center_of_masses`	2.29 ±0.05	2.14 ±0.10	-6.2% (noise)	0.045
`poi_reorient`	0.54 ±0.01	0.51 ±0.03	-5.3% (noise)	0.306
`nii_map_labels`	1.31 ±0.09	1.24 ±0.07	-5.3% (noise)	0.120
`poi_map_labels`	0.29 ±0.00	0.30 ±0.01	+5.2% (noise)	0.008
`nii_unique`	0.73 ±0.03	0.69 ±0.01	-4.9% (noise)	0.035
`nii_compute_crop`	0.38 ±0.01	0.36 ±0.02	-4.8% (noise)	0.311
`nii_get_array`	0.07 ±0.00	0.06 ±0.01	-3.8% (noise)	0.696
`nii_set_dtype`	0.58 ±0.02	0.59 ±0.01	+2.8% (noise)	0.023
`nii_volumes`	2.23 ±0.10	2.19 ±0.51	-1.8% (noise)	0.240
`poi_save`	0.54 ±0.03	0.55 ±0.04	+1.7% (noise)	0.757

`mri_2d` — shape (68, 52, 1)

Measurement	baseline ms (median ±½·range)	head ms (median ±½·range)	Δ %	p
`nii_rescale`	6.11 ±0.11	5.79 ±0.21	-5.2%	0.009
`nii_resample_from_to`	7.96 ±0.21	7.91 ±0.07	-0.6%	0.423
`nii_dilate_msk`	19.75 ±0.26	19.74 ±0.35	-0.1%	0.765
`metric_voxels`	3536.00 ±0.00	3536.00 ±0.00	+0.0%	—
`metric_labels`	7.00 ±0.00	7.00 ±0.00	+0.0%	—

… 30 more measurements

Measurement	baseline ms (median ±½·range)	head ms (median ±½·range)	Δ %	p
`metric_foreground_pct`	50.17 ±0.00	50.17 ±0.00	+0.0%	—
`poi_resample_from_to`	0.82 ±0.03	0.95 ±0.08	+15.8% (noise)	0.008
`poi_rescale`	0.34 ±0.02	0.29 ±0.01	-14.7% (noise)	0.011
`nii_set_dtype`	0.44 ±0.03	0.48 ±0.03	+10.9% (noise)	0.065
`nii_unique`	0.13 ±0.01	0.14 ±0.01	+9.7% (noise)	0.367
`nii_fill_holes`	1.26 ±0.05	1.15 ±0.04	-8.6% (noise)	0.020
`nii_extract_label`	0.47 ±0.03	0.50 ±0.03	+7.0% (noise)	0.248
`nii_get_array`	0.02 ±0.00	0.02 ±0.01	-6.6% (noise)	0.502
`poi_save`	0.56 ±0.05	0.53 ±0.04	-6.1% (noise)	0.653
`nii_get_connected_components`	0.72 ±0.02	0.69 ±0.03	-5.3% (noise)	0.099
`poi_calc_centroids_nocrop`	1.64 ±0.08	1.56 ±0.03	-4.9% (noise)	0.059
`nii_center_of_masses`	0.54 ±0.03	0.51 ±0.02	-4.8% (noise)	0.055
`nii_map_labels`	0.58 ±0.04	0.55 ±0.03	-4.8% (noise)	0.302
`poi_map_labels`	0.27 ±0.01	0.28 ±0.02	+4.1% (noise)	0.249
`nii_pad_to`	0.62 ±0.03	0.64 ±0.02	+3.7% (noise)	0.228
`poi_local_to_global_arr`	0.42 ±0.01	0.41 ±0.02	-2.9% (noise)	0.293
`nii_reorient`	0.94 ±0.06	0.97 ±0.08	+2.9% (noise)	0.455
`poi_to_global`	0.29 ±0.02	0.29 ±0.03	-2.7% (noise)	0.765
`nii_rescale_seg`	0.83 ±0.04	0.85 ±0.07	+2.5% (noise)	0.288
`nii_load_seg`	1.91 ±0.06	1.87 ±0.07	-2.3% (noise)	0.548
`nii_erode_msk`	1.41 ±0.04	1.44 ±0.10	+2.1% (noise)	0.466
`nii_compute_crop`	0.16 ±0.01	0.16 ±0.01	+1.8% (noise)	0.936
`nii_volumes`	0.58 ±0.02	0.57 ±0.02	-1.3% (noise)	0.231
`nii_save`	1.55 ±0.16	1.53 ±0.11	-1.2% (noise)	0.427
`nii_apply_crop`	0.78 ±0.05	0.79 ±0.03	+1.0% (noise)	0.465
`nii_filter_connected_components`	0.92 ±0.04	0.92 ±0.04	-0.8% (noise)	0.130
`poi_load`	2.55 ±0.10	2.53 ±0.06	-0.8% (noise)	0.171
`poi_calc_centroids`	1.35 ±0.07	1.34 ±0.08	-0.6% (noise)	0.506
`nii_load_img`	1.66 ±0.09	1.65 ±0.14	-0.6% (noise)	0.630
`poi_reorient`	0.52 ±0.02	0.51 ±0.01	-0.0% (noise)	0.640

Gate: a measurement fails when the baseline is ≥ 1 ms, the median grows by ≥ 50%, and Welch's t-test gives p < 0.05. metric_* rows are context only.

Memory (peak RSS growth per call)

baseline 9c15f60 vs head 9c15f60 · python 3.11.17 · 5 repeats + 1 warmup · sampler proc-status, isolation fork · measurement floor ≈ 0.74 MiB

Showing the 5 most-changed measurements per case; the rest are collapsed. Rows with a baseline below 3 MiB are tagged (noise) — runner jitter on those swamps any real change.

`ct_3d` — shape (73, 47, 73)

Measurement	baseline MiB (median ±½·range)	head MiB (median ±½·range)	Δ %	p
`nii_rescale`	11.48 ±0.00	11.39 ±0.00	-0.8%	—
`nii_rescale_seg`	11.48 ±0.00	11.39 ±0.00	-0.8%	—
`nii_resample_from_to`	11.54 ±0.00	11.45 ±0.00	-0.8%	—
`poi_map_labels`	3.07 ±0.00	3.05 ±0.00	-0.8%	—
`poi_load`	3.10 ±0.00	3.08 ±0.00	-0.8%	—

… 31 more measurements

Measurement	baseline MiB (median ±½·range)	head MiB (median ±½·range)	Δ %	p
`nii_set_dtype`	11.11 ±0.00	11.03 ±0.00	-0.7%	—
`nii_extract_label`	11.11 ±0.00	11.03 ±0.00	-0.7%	—
`nii_pad_to`	11.17 ±0.00	11.09 ±0.00	-0.7%	—
`poi_calc_centroids_nocrop`	11.29 ±0.00	11.21 ±0.00	-0.7%	—
`nii_map_labels`	11.30 ±0.00	11.21 ±0.00	-0.7%	—
`nii_apply_crop`	11.40 ±0.00	11.32 ±0.00	-0.7%	—
`poi_calc_centroids`	11.48 ±0.00	11.40 ±0.00	-0.7%	—
`nii_load_seg`	11.51 ±0.00	11.43 ±0.00	-0.7%	—
`nii_fill_holes`	11.71 ±0.00	11.63 ±0.00	-0.7%	—
`nii_reorient`	12.11 ±0.00	12.02 ±0.00	-0.7%	—
`poi_calc_poi_from_subreg_vert`	12.15 ±0.00	12.07 ±0.00	-0.7%	—
`nii_get_connected_components`	12.92 ±0.00	12.84 ±0.00	-0.6%	—
`nii_filter_connected_components`	13.11 ±0.00	13.03 ±0.00	-0.6%	—
`poi_rescale`	5.28 ±0.00	5.26 ±0.00	-0.4%	—
`nii_unique`	5.39 ±0.00	5.36 ±0.00	-0.4%	—
`nii_compute_crop`	5.51 ±0.00	5.49 ±0.00	-0.4%	—
`nii_dilate_msk`	11.30 ±0.00	11.25 ±0.00	-0.4%	—
`nii_erode_msk`	11.38 ±0.00	11.33 ±0.00	-0.4%	—
`nii_center_of_masses`	5.82 ±0.00	5.80 ±0.00	-0.4%	—
`nii_volumes`	5.89 ±0.00	5.86 ±0.00	-0.4%	—
`nii_save`	5.98 ±0.00	5.96 ±0.00	-0.4%	—
`nii_load_img`	6.00 ±0.00	5.97 ±0.00	-0.4%	—
`poi_local_to_global_arr`	7.45 ±0.00	7.42 ±0.00	-0.3%	—
`poi_to_global`	7.51 ±0.00	7.48 ±0.00	-0.3%	—
`poi_reorient`	7.95 ±0.00	7.92 ±0.00	-0.3%	—
`poi_resample_from_to`	10.92 ±0.00	10.90 ±0.00	-0.2%	—
`metric_voxels`	250463.00 ±0.00	250463.00 ±0.00	+0.0%	—
`metric_labels`	3.00 ±0.00	3.00 ±0.00	+0.0%	—
`metric_foreground_pct`	14.51 ±0.00	14.51 ±0.00	+0.0%	—
`poi_save`	2.05 ±0.00	2.03 ±0.03	-1.1% (noise)	0.035
`nii_get_array`	2.95 ±0.00	2.93 ±0.00	-0.8% (noise)	—

`ct_2d` — shape (73, 47, 1)

Measurement	baseline MiB (median ±½·range)	head MiB (median ±½·range)	Δ %	p
`nii_rescale`	11.48 ±0.00	11.39 ±0.00	-0.8%	—
`nii_rescale_seg`	11.48 ±0.00	11.39 ±0.00	-0.8%	—
`nii_resample_from_to`	11.54 ±0.00	11.45 ±0.00	-0.8%	—
`poi_map_labels`	3.07 ±0.03	3.05 ±0.00	-0.8%	0.317
`poi_load`	3.10 ±0.00	3.08 ±0.00	-0.8%	—

… 30 more measurements

Measurement	baseline MiB (median ±½·range)	head MiB (median ±½·range)	Δ %	p
`nii_set_dtype`	11.11 ±0.00	11.03 ±0.00	-0.7%	—
`nii_extract_label`	11.11 ±0.00	11.03 ±0.00	-0.7%	—
`nii_pad_to`	11.17 ±0.00	11.09 ±0.00	-0.7%	—
`nii_map_labels`	11.30 ±0.00	11.21 ±0.00	-0.7%	—
`poi_calc_centroids_nocrop`	11.36 ±0.00	11.27 ±0.00	-0.7%	—
`nii_apply_crop`	11.40 ±0.00	11.32 ±0.00	-0.7%	—
`poi_calc_centroids`	11.48 ±0.00	11.40 ±0.00	-0.7%	—
`nii_load_seg`	11.51 ±0.00	11.43 ±0.00	-0.7%	—
`nii_fill_holes`	11.71 ±0.00	11.63 ±0.00	-0.7%	—
`nii_reorient`	12.11 ±0.00	12.02 ±0.00	-0.7%	—
`nii_get_connected_components`	12.92 ±0.00	12.84 ±0.00	-0.6%	—
`nii_filter_connected_components`	13.11 ±0.00	13.03 ±0.00	-0.6%	—
`poi_rescale`	5.28 ±0.00	5.26 ±0.00	-0.4%	—
`nii_unique`	5.39 ±0.00	5.36 ±0.00	-0.4%	—
`nii_compute_crop`	5.51 ±0.00	5.49 ±0.00	-0.4%	—
`nii_dilate_msk`	11.30 ±0.00	11.25 ±0.00	-0.4%	—
`nii_erode_msk`	11.38 ±0.00	11.33 ±0.00	-0.4%	—
`nii_center_of_masses`	5.82 ±0.00	5.80 ±0.00	-0.4%	—
`nii_volumes`	5.89 ±0.00	5.86 ±0.00	-0.4%	—
`nii_save`	5.98 ±0.00	5.96 ±0.00	-0.4%	—
`nii_load_img`	6.00 ±0.00	5.97 ±0.00	-0.4%	—
`poi_local_to_global_arr`	7.45 ±0.00	7.42 ±0.00	-0.3%	—
`poi_to_global`	7.51 ±0.00	7.48 ±0.00	-0.3%	—
`poi_reorient`	7.95 ±0.00	7.92 ±0.00	-0.3%	—
`poi_resample_from_to`	10.92 ±0.00	10.90 ±0.00	-0.2%	—
`metric_voxels`	3431.00 ±0.00	3431.00 ±0.00	+0.0%	—
`metric_labels`	3.00 ±0.00	3.00 ±0.00	+0.0%	—
`metric_foreground_pct`	37.60 ±0.00	37.60 ±0.00	+0.0%	—
`poi_save`	2.05 ±0.00	2.03 ±0.00	-1.1% (noise)	—
`nii_get_array`	2.95 ±0.00	2.93 ±0.00	-0.8% (noise)	—

`mri_3d` — shape (68, 52, 67)

Measurement	baseline MiB (median ±½·range)	head MiB (median ±½·range)	Δ %	p
`nii_rescale`	11.48 ±0.00	11.39 ±0.00	-0.8%	—
`nii_rescale_seg`	11.48 ±0.00	11.39 ±0.00	-0.8%	—
`nii_resample_from_to`	11.54 ±0.00	11.45 ±0.00	-0.8%	—
`poi_map_labels`	3.07 ±0.00	3.05 ±0.00	-0.8%	—
`poi_load`	3.10 ±0.00	3.08 ±0.00	-0.8%	—

… 31 more measurements

Measurement	baseline MiB (median ±½·range)	head MiB (median ±½·range)	Δ %	p
`nii_extract_label`	11.11 ±0.00	11.03 ±0.00	-0.7%	—
`nii_pad_to`	11.17 ±0.00	11.09 ±0.00	-0.7%	—
`nii_set_dtype`	11.17 ±0.00	11.09 ±0.00	-0.7%	—
`poi_calc_centroids_nocrop`	11.29 ±0.00	11.21 ±0.00	-0.7%	—
`nii_map_labels`	11.30 ±0.00	11.21 ±0.00	-0.7%	—
`nii_apply_crop`	11.40 ±0.00	11.32 ±0.00	-0.7%	—
`poi_calc_centroids`	11.48 ±0.00	11.40 ±0.00	-0.7%	—
`nii_load_seg`	11.51 ±0.00	11.43 ±0.00	-0.7%	—
`nii_fill_holes`	11.71 ±0.00	11.63 ±0.00	-0.7%	—
`nii_reorient`	12.11 ±0.00	12.02 ±0.00	-0.7%	—
`poi_calc_poi_from_subreg_vert`	12.15 ±0.00	12.07 ±0.00	-0.7%	—
`nii_get_connected_components`	12.92 ±0.00	12.84 ±0.00	-0.6%	—
`nii_filter_connected_components`	13.11 ±0.00	13.03 ±0.00	-0.6%	—
`poi_rescale`	5.28 ±0.00	5.26 ±0.00	-0.4%	—
`nii_unique`	5.39 ±0.00	5.36 ±0.00	-0.4%	—
`nii_compute_crop`	5.51 ±0.00	5.49 ±0.00	-0.4%	—
`nii_dilate_msk`	11.30 ±0.00	11.25 ±0.00	-0.4%	—
`nii_erode_msk`	11.38 ±0.00	11.33 ±0.00	-0.4%	—
`nii_center_of_masses`	5.82 ±0.00	5.80 ±0.00	-0.4%	—
`nii_volumes`	5.89 ±0.00	5.86 ±0.00	-0.4%	—
`nii_save`	5.98 ±0.00	5.96 ±0.00	-0.4%	—
`nii_load_img`	6.00 ±0.00	5.97 ±0.00	-0.4%	—
`poi_local_to_global_arr`	7.45 ±0.00	7.42 ±0.00	-0.3%	—
`poi_to_global`	7.51 ±0.00	7.48 ±0.00	-0.3%	—
`poi_reorient`	7.95 ±0.00	7.92 ±0.00	-0.3%	—
`poi_resample_from_to`	10.92 ±0.00	10.90 ±0.00	-0.2%	—
`metric_voxels`	236912.00 ±0.00	236912.00 ±0.00	+0.0%	—
`metric_labels`	8.00 ±0.00	8.00 ±0.00	+0.0%	—
`metric_foreground_pct`	18.21 ±0.00	18.21 ±0.00	+0.0%	—
`poi_save`	2.12 ±0.00	2.09 ±0.00	-1.1% (noise)	0.000
`nii_get_array`	2.95 ±0.00	2.93 ±0.00	-0.8% (noise)	—

`mri_2d` — shape (68, 52, 1)

Measurement	baseline MiB (median ±½·range)	head MiB (median ±½·range)	Δ %	p
`nii_rescale`	11.48 ±0.00	11.39 ±0.00	-0.8%	—
`nii_rescale_seg`	11.48 ±0.00	11.39 ±0.00	-0.8%	—
`nii_resample_from_to`	11.54 ±0.03	11.45 ±0.00	-0.8%	0.002
`poi_map_labels`	3.07 ±0.00	3.05 ±0.00	-0.8%	—
`poi_load`	3.10 ±0.00	3.08 ±0.00	-0.8%	—

… 30 more measurements

Measurement	baseline MiB (median ±½·range)	head MiB (median ±½·range)	Δ %	p
`nii_extract_label`	11.11 ±0.00	11.03 ±0.00	-0.7%	—
`nii_pad_to`	11.17 ±0.00	11.09 ±0.00	-0.7%	—
`nii_set_dtype`	11.17 ±0.00	11.09 ±0.00	-0.7%	—
`nii_map_labels`	11.30 ±0.00	11.21 ±0.00	-0.7%	—
`poi_calc_centroids_nocrop`	11.36 ±0.00	11.27 ±0.00	-0.7%	—
`nii_apply_crop`	11.40 ±0.00	11.32 ±0.00	-0.7%	—
`poi_calc_centroids`	11.48 ±0.00	11.40 ±0.00	-0.7%	0.000
`nii_load_seg`	11.51 ±0.00	11.43 ±0.00	-0.7%	—
`nii_fill_holes`	11.71 ±0.00	11.63 ±0.00	-0.7%	—
`nii_reorient`	12.11 ±0.00	12.02 ±0.00	-0.7%	—
`nii_get_connected_components`	12.92 ±0.00	12.84 ±0.00	-0.6%	—
`nii_filter_connected_components`	13.11 ±0.00	13.03 ±0.00	-0.6%	0.000
`poi_rescale`	5.28 ±0.00	5.26 ±0.00	-0.4%	—
`nii_unique`	5.39 ±0.00	5.36 ±0.00	-0.4%	—
`nii_compute_crop`	5.51 ±0.00	5.49 ±0.00	-0.4%	—
`nii_dilate_msk`	11.30 ±0.00	11.25 ±0.00	-0.4%	—
`nii_erode_msk`	11.38 ±0.00	11.33 ±0.00	-0.4%	0.000
`nii_center_of_masses`	5.82 ±0.00	5.80 ±0.00	-0.4%	—
`nii_volumes`	5.89 ±0.00	5.86 ±0.00	-0.4%	—
`nii_save`	5.98 ±0.00	5.96 ±0.00	-0.4%	—
`nii_load_img`	6.00 ±0.00	5.97 ±0.00	-0.4%	—
`poi_local_to_global_arr`	7.45 ±0.00	7.42 ±0.00	-0.3%	—
`poi_to_global`	7.51 ±0.00	7.48 ±0.00	-0.3%	—
`poi_reorient`	7.95 ±0.00	7.92 ±0.00	-0.3%	—
`poi_resample_from_to`	10.92 ±0.00	10.90 ±0.00	-0.2%	—
`metric_voxels`	3536.00 ±0.00	3536.00 ±0.00	+0.0%	—
`metric_labels`	7.00 ±0.00	7.00 ±0.00	+0.0%	—
`metric_foreground_pct`	50.17 ±0.00	50.17 ±0.00	+0.0%	—
`poi_save`	2.12 ±0.00	2.09 ±0.00	-1.1% (noise)	—
`nii_get_array`	2.95 ±0.00	2.93 ±0.00	-0.8% (noise)	—

Gate: a measurement fails when the baseline is ≥ 1 MiB, the median grows by ≥ 50%, and Welch's t-test gives p < 0.05. metric_* rows are context only.

robert-graf and others added 30 commits September 17, 2026 11:51

more auto accepts

94bb5d6

add versioning

afe8329

add Verida lordosie version

9467429

add S1 Endplate

06a5f13

add warning to not use is_ct in "DAExt"

840e353

claude_remapping

431a320

fix angls.

7d5b1a5

logging

cfa270a

Merge branch 'tanja2' of github.com:Hendrik-code/TPTBox into cleanup/…

a134232

…cobb-endplate-v3

Merge remote-tracking branch 'origin/poi-label-names' into cleanup/co…

3a08b81

…bb-endplate-v3

fix example test case compile

1a723a8

Merge branch 'main' into cleanup/cobb-endplate-v3

662afff

skipt T13 for apex

cbf9afa

last_vert must be in poi

d58bfd0

Merge branch 'cleanup/cobb-endplate-v3' into dicom-extract-cleanup

7eb56ab

fix min_value, smaller overlap nii to save RAM

b0df6f0

add c_val option

ad09d9e

only warn if memory is not trival

be514be

Refactor stiching to us TPTBox function which are more efficent.

7436d9a

Merge remote-tracking branch 'origin/robert-pc' into robert-pc-pr-sti…

6008d96

…tching-refactor # Conflicts: # TPTBox/stitching/stitching.py

more 4D chatches in DICOM

664183b

skip missing plane/source POIs; widen sacrum-endplate crop; bump run_…

cf0d29d

…all defaults Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

Merge branch 'robert-develop' of github.com:Hendrik-code/TPTBox into …

9f4ee3a

…robert-develop

bug fix _rib_assingn

449ebcd

style fixes by ruff

41e4a15

deepali point registration and clean up

2716437

robert-graf added 9 commits October 7, 2026 11:00

rib gen fix

4d1e0d9

Merge branch 'robert-develop' of github.com:Hendrik-code/TPTBox into …

ae8ad57

…robert-develop

Merge branch 'tanja' into robert-develop

1f1f54c

better cms handling

00c8bbe

ruff

08aff66

Merge remote-tracking branch 'origin/robert-develop' into dicom-extra…

b585f7a

…ct-cleanup

add 2D support for dicom

8f88511

Merge branch 'main' into dicom-extract-cleanup

200ea16

ruff and pin ubuntu version for 3.9

017a054

change in behavior; update test

aae34bc

robert-graf requested a review from Hendrik-code October 28, 2026 19:01

Navigation Menu

Repository navigation

DICOM extract cleanup, spinestats v3 (endplate Cobb), point registration, stitching refactor - #160

DICOM extract cleanup, spinestats v3 (endplate Cobb), point registration, stitching refactor#160
robert-graf wants to merge 40 commits into
mainfrom
dicom-extract-cleanup

robert-graf commented Oct 28, 2026

Uh oh!

github-actions Bot commented Oct 28, 2026 •

edited

Loading

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Conversation

robert-graf commented Oct 28, 2026

Summary

Test plan

Uh oh!

github-actions Bot commented Oct 28, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Benchmark comparison

Speed (wall time per call)

ct_3d — shape (73, 47, 73)

ct_2d — shape (73, 47, 1)

mri_3d — shape (68, 52, 67)

mri_2d — shape (68, 52, 1)

Memory (peak RSS growth per call)

ct_3d — shape (73, 47, 73)

ct_2d — shape (73, 47, 1)

mri_3d — shape (68, 52, 67)

mri_2d — shape (68, 52, 1)

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

github-actions Bot commented Oct 28, 2026 •

edited

Loading

`ct_3d` — shape (73, 47, 73)

`ct_2d` — shape (73, 47, 1)

`mri_3d` — shape (68, 52, 67)

`mri_2d` — shape (68, 52, 1)

`ct_3d` — shape (73, 47, 73)

`ct_2d` — shape (73, 47, 1)

`mri_3d` — shape (68, 52, 67)

`mri_2d` — shape (68, 52, 1)