[CSimon-A]

Quick description

Fixes #1371. In BundleSeg.recognize(), _reduce_search_space uses the model centroids, while prune_far_from_model uses the full model streamlines. _register_model_to_neighb transformed the streamlines but not the centroids, so the second _reduce_search_space (14 mm) selected candidates around the model before the local SLR, while the pruning compared them to the model after the SLR.

Type of change

• bundleseg.py: also apply the SLR transform to model_centroids, so the 14 mm search and the pruning use the same (registered) model.
• test_bundleseg.py (was an empty stub): synthetic bundles, with the model shifted by 15 mm (between the 14 and 16 mm thresholds). Fails before the fix (0/50 streamlines recognized), passes after. Plus a basic recognition test with a 5 mm shift.

Check the relevant options.

• Bug fix (non-breaking change which fixes an issue)
• New feature (non-breaking change which adds functionality)
• Breaking change (fix or feature that would cause existing functionality to not work as expected)
• This change requires a documentation update

Validation

• test_tractogram_segment_with_bundleseg.py passes. On the fibercup test data, the outputs are identical before and after the fix.
• HCP test-retest: 3 subjects x 2 sessions, Zenodo BundleSeg atlas, seed 0, same code before/after except this fix.
  • Runs are reproducible, and the local SLR gives the same result with and without the fix, so all differences come from the 14 mm search.
  • Overall (51 bundles × 6 sessions): 34,829 streamlines gained and 4,897 lost, a net increase of ~1.0%. Streamline Dice between the outputs before and after the fix: 0.993.
  • Most bundles barely change. The largest changes are in small bundles (fewer than ~1,000 streamlines) that the SLR moved by more than ~5 mm.
  • In one session, the 14 mm search found no candidate for AC before the fix (133 after).
  • Test-retest voxel Dice unchanged (median over bundles 0.764 -> 0.765).
  • Similar run times before and after, even though 2.1% more candidates reach the pruning step.

Selected bundles (the most changed, two that barely change, the largest loss and the largest gain):

Bundle	Mean streamlines, before -> after	Mean Dice	Gained	Lost	Mean SLR (mm)
FX_R	96 -> 180	0.697	522	16	9.0
FX_L	104 -> 124	0.735	133	10	10.7
CG_R_curve	271 -> 304	0.846	378	177	6.8
CG_L_curve	607 -> 673	0.870	570	174	6.0
IFOF_R	6,333 -> 6,335	1.000	15	4	4.1
AF_L	9,788 -> 9,802	0.999	85	5	7.6
PYT_R	24,530 -> 24,280	0.993	79	1,574	3.7
CC_Fr_2	26,801 -> 29,302	0.963	15,011	1	6.3
All 51 bundles	2,979,511 -> 3,009,443	0.993	34,829	4,897

• Mean streamlines, Mean Dice, Mean SLR: averaged over the 6 sessions.
• Gained / Lost: summed over the 6 sessions.
• Dice: overlap of the recognized streamlines before and after the fix (1 = identical).
• Mean SLR: average distance the local SLR moved the model points (measured with temporary logging, not part of this PR).
• All 51 bundles: streamline counts summed over all bundles and sessions; Dice computed on all streamlines together.

As there is no ground truth, these numbers show how much the output changes, not which one is better. Gains and losses both come from the 14 mm search keeping or dropping whole tractogram clusters based on their centroid. The candidate search and the pruning now use the same (registered) model.

Full table (51 bundles)

Columns (each row combines the 6 sessions, bundles sorted by size):

• Mean before / Mean after, Mean Dice: averaged over the sessions. Min Dice: lowest session.
• Gained / Lost: summed over the sessions.
• Sessions changed: sessions where the output differs.
• Mean SLR: average distance the local SLR moved the model points, averaged over the sessions. Max SLR: largest distance moved by a single model point.
• 14 mm empty: sessions where the 14 mm search found no candidate (before -> after).

Bundle	Mean before	Mean after	Mean Dice	Min Dice	Gained	Lost	Sessions changed	Mean SLR (mm)	Max SLR (mm)	14 mm empty
PC	1	1	0.944	0.667	0	1	1/6	8.6	25.7	0 -> 0
FX_R	96	180	0.697	0.220	522	16	6/6	9.0	22.6	0 -> 0
FX_L	104	124	0.735	0.216	133	10	6/6	10.7	23.6	0 -> 0
AC	226	236	0.965	0.836	66	8	2/6	7.7	18.2	1 -> 0
CG_R_curve	271	304	0.846	0.579	378	177	6/6	6.8	23.2	0 -> 0
CG_L_curve	607	673	0.870	0.629	570	174	5/6	6.0	19.0	0 -> 0
CG_L_Po	648	669	0.946	0.733	148	24	4/6	6.8	16.2	0 -> 0
SCP_L	875	875	0.973	0.933	107	108	6/6	6.6	21.4	0 -> 0
CC_Te	1042	1043	0.998	0.994	11	1	4/6	5.3	22.7	0 -> 0
CG_R_Po	1114	1187	0.970	0.832	461	24	5/6	5.9	14.1	0 -> 0
UF_R	1130	1128	0.996	0.986	6	15	4/6	9.8	23.4	0 -> 0
SCP_R	1230	1250	0.978	0.943	156	39	6/6	6.6	22.0	0 -> 0
UF_L	1275	1282	0.997	0.986	41	1	3/6	8.2	17.4	0 -> 0
ICP_R	1476	1480	0.963	0.781	24	4	2/6	5.9	14.1	0 -> 0
ICP_L	2098	2098	0.997	0.991	30	33	5/6	6.6	15.6	0 -> 0
MdLF_R	2699	2697	0.987	0.951	70	85	6/6	8.2	28.8	0 -> 0
MdLF_L	3502	3609	0.958	0.777	694	50	6/6	8.6	29.9	0 -> 0
CG_R_An	3572	3617	0.987	0.961	297	23	5/6	8.8	28.0	0 -> 0
CG_L_An	4872	4886	0.997	0.992	105	24	4/6	8.7	25.4	0 -> 0
OR_ML_R	5487	5527	0.996	0.989	248	5	5/6	10.1	21.1	0 -> 0
CG_L	5782	5787	0.995	0.982	134	103	6/6	7.9	19.9	0 -> 0
CG_R	5840	5809	0.994	0.986	95	284	6/6	7.5	17.0	0 -> 0
CC_Oc	6084	6162	0.994	0.987	472	0	6/6	5.5	12.0	0 -> 0
IFOF_R	6333	6335	1.000	0.999	15	4	5/6	4.1	13.7	0 -> 0
SLF_L	6751	6773	0.996	0.984	173	44	6/6	8.6	23.4	0 -> 0
OR_ML_L	6775	6804	0.998	0.992	180	3	5/6	9.3	23.9	0 -> 0
POPT_R_Brainstem	6945	7207	0.975	0.886	1637	66	5/6	4.2	15.0	0 -> 0
IFOF_L	7433	7420	0.998	0.994	33	110	3/6	3.6	12.3	0 -> 0
AF_L	9788	9802	0.999	0.998	85	5	6/6	7.6	18.8	0 -> 0
POPT_L_Brainstem	10699	10872	0.993	0.969	1099	64	6/6	6.5	17.9	0 -> 0
POPT_R	11489	11479	0.999	0.995	73	134	5/6	5.7	15.5	0 -> 0
FPT_L_Brainstem	12847	13133	0.975	0.860	1793	76	6/6	5.5	16.0	0 -> 0
CC_Pa	13314	13322	0.999	0.998	101	51	5/6	5.8	18.4	0 -> 0
FAT_R	13692	13836	0.991	0.976	1147	284	6/6	8.2	16.3	0 -> 0
POPT_L	14229	14318	0.996	0.990	593	64	6/6	6.1	15.8	0 -> 0
FPT_R_Brainstem	14420	14438	0.996	0.986	376	272	5/6	5.5	15.4	0 -> 0
PYT_L_Brainstem	14487	14531	0.997	0.993	465	200	6/6	4.4	16.1	0 -> 0
PYT_R_Brainstem	16051	16050	1.000	1.000	12	16	4/6	3.2	12.7	0 -> 0
FAT_L	16815	17180	0.990	0.968	2243	54	6/6	9.9	22.4	0 -> 0
FPT_L	18054	18104	0.997	0.986	339	42	5/6	4.2	15.4	0 -> 0
CC_Fr_1	18420	18746	0.991	0.967	1955	0	6/6	7.4	15.7	0 -> 0
CC_Pr_Po	19566	19683	0.997	0.987	716	12	5/6	3.4	10.7	0 -> 0
ILF_L	20604	20652	0.999	0.998	290	0	6/6	8.6	17.8	0 -> 0
MCP	21273	21299	0.999	0.996	157	1	4/6	4.2	10.0	0 -> 0
ILF_R	21667	21764	0.998	0.996	587	1	5/6	8.1	17.8	0 -> 0
PYT_L	21992	21979	1.000	0.999	0	73	2/6	3.5	11.3	0 -> 0
AF_R	22372	22401	0.999	0.996	200	28	6/6	6.2	15.7	0 -> 0
FPT_R	22778	22781	1.000	1.000	20	1	3/6	4.9	16.3	0 -> 0
PYT_R	24530	24280	0.993	0.967	79	1574	5/6	3.7	11.9	0 -> 0
SLF_R	26430	26459	0.996	0.990	682	508	6/6	8.2	22.9	0 -> 0
CC_Fr_2	26801	29302	0.963	0.922	15011	1	6/6	6.3	14.7	0 -> 0

Checklist

• My code follows the style guidelines of this project (run autopep8)
• I added relevant citations to scripts, modules and functions docstrings and descriptions
• I have performed a self-review of my code
• I have commented my code, particularly in hard-to-understand areas
• I have made corresponding changes to the documentation
• My changes generate no new warnings
• I moved all functions from the script file (except the argparser and main) to scilpy modules
• I have added tests that prove my fix is effective or that my feature works
• New and existing unit tests pass locally with my changes

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 72.85%. Comparing base (c787973) to head (884772b).

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #1373      +/-   ##
==========================================
+ Coverage   72.57%   72.85%   +0.27%     
==========================================
  Files         301      301              
  Lines       26279    26281       +2     
  Branches     3702     3702              
==========================================
+ Hits        19073    19148      +75     
+ Misses       5655     5572      -83     
- Partials     1551     1561      +10     

Flag	Coverage Δ
smoketests	69.72% <0.00%> (-0.02%)	⬇️
unittests	14.69% <100.00%> (+0.47%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

Components	Coverage Δ
Scripts	75.26% <ø> (-0.02%)	⬇️
Library	69.96% <100.00%> (+0.63%)	⬆️

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.