Evidence · ANTs

Atropos

Atropos labels the voxels of a mask with a few classes, alternating between the classes’ intensity models and the labelling, with a Markov random field (MRF). LARMORX’s port is a replica of ANTs 2.6.5’s program on ITK 5.4.5. In reproduce it gives ANTs’ images bit for bit in every compared case of its two records once ANTs uses correctly rounded maths (58 of 58 and 17 of 17). With ANTs as built, 27 of the 58 and 3 of the 17 match bit for bit, and the classified image matches in all of them; the others’ posteriors differ only where glibc’s expf, logf, exp, log or pow is not correctly rounded. sMRIPrep’s own call on five fMRIPrep runs is among the 58: identical with correctly rounded maths in all five, and as built in one. Its default, fix, corrects ITK-50 and eighteen other defects of ANTs and ITK (ITK-51, ANTS-90 to ANTS-106), among them partial-volume classes that the MRF shut out (ANTS-106).

Ports
  • replica larmorx Apache-2.0: translated from ANTs 2.6.5's and ITK 5.4.5's source (Apache-2.0)
Status
replica validated, with its data panel (marked before the statuses' definition)
Records
standard comparison, full comparison, data panel, all against ANTs 2.6.5
Reference
Atropos · API page ants-atropos · findings ants-atropos

What is compared

Atropos labels the voxels of a mask with a few classes (tissues), alternating between the classes’ intensity models and the labelling: K-means, Otsu, random or prior-based initializations; Gaussian, Parzen-window and tensor likelihoods; an MRF prior updated by iterated conditional modes. sMRIPrep calls it once per subject, in its brain extraction, to split the brain into CSF, grey and white matter.

The replica runs beside ANTs 2.6.5’s Atropos on the same files with the same arguments. ANTs is the repository’s build (GCC 14.3), run as built and with correctly rounded maths, at one thread in the records and at more for its own variability; three times with the clock’s seed, where its runs differ. fMRIPrep’s own ANTs 2.6.2 runs too, in fMRIPrep’s image.

bash
larmorx ants Atropos -d 3 -a t1w_n4.nii.gz -x brain_mask.nii.gz -i 'KMeans[3]' -k Gaussian -m '[0.1,1x1x1]' -c '[3,0]' -r 0 -o '[labels.nii.gz,POSTERIOR_%02d.nii.gz]'
python
import larmorx as lx

seg = lx.ants.atropos("t1w_n4.nii.gz", "brain_mask.nii.gz", classes=3, mrf_smoothing=0.1,
                      iterations=3, convergence_threshold=0, use_random_seed=False)
csf, gm, wm = seg.posteriors

The API page maps every option. How to read what follows: how to read the evidence.

Its port

ANTs and ITK come under the Apache-2.0 licence, as LARMORX does, so the replica is in larmorx itself. It is translated from ANTs’ and ITK’s source, step by step, to give their results bit for bit. A clean room of ANTs’ programs is planned after FreeSurfer (roadmap).

The replica computes the C library’s maths correctly rounded, the same on every platform. Where glibc’s expf and logf are not correctly rounded for a value a run meets, ANTs as built gives the same classified image and posteriors that differ in their last bits.

The verdicts at a glance

Agreement, replica against the original: ANTs Atropos 2.6.5 (standard comparison and data panel, 115 cases of the two records), by category and verdict
initializations6
posteriors16
likelihood models5
the MRF5
outliers5
partial-volume classes6
dimensions2
inputs3
messages31
verbose output5
random seeds2
real images: sMRIPrep's call5
real images: tissue priors2
real images: partial-volume classes1
real images: other calls3
data panel: fMRIPrep's calls14
data panel: other images4

Select a segment to list its cases below.

In reproduce, every case that writes images gives ANTs’ images bit for bit at one thread with correctly rounded maths, but one: Grubbs-Rosner from its second iteration, where ANTs weighs part of its samples with memory it never wrote (ANTS-107), so that its own result changes with the length of its output paths. Against ANTs as built the classified images are identical in every case, and the posteriors differ in their last bits where glibc’s expf and logf are not correctly rounded. Where ANTs aborts or crashes, the replica prints ANTs’ message and stops; where ANTs seeds from the clock, its runs differ from each other, and the replica uses a fixed seed.

In sMRIPrep: a real T1w segmented

fMRIPrep’s brain extraction segments the N4-corrected T1w into three classes. Its inputs and arguments go to ANTs at 1, 2, 4 and 8 threads, three times with the clock’s seed, to fMRIPrep’s own ANTs, and to the replica. Atropos does not depend on ANTs’ thread count, and fMRIPrep’s output is the repository’s ANTs result; the replica differs from it only in the last bits of a few posteriors (glibc’s maths). Its default, fix, corrects ITK-50, where ITK’s k-means leaves out the k-d tree’s split points: a change smaller than ANTs’ own spread when it seeds from the clock, shown on the slice below.

Worked example

sMRIPrep's segmentation of a real T1w

What this shows: In reproduce, on 1 and 6 threads, the replica gives ANTs’ one-thread result with correctly rounded maths bit for bit. ANTs as built, and fMRIPrep’s output, give the same classified image; a few posteriors differ in their last bits (by at most 1.2e-07), where glibc’s expf and logf are not correctly rounded. Atropos does not depend on ANTs’ thread count, and fMRIPrep’s output is ANTs 2.6.5’s at 8 threads bit for bit. With the clock’s seed, ANTs’ own runs differ from each other in up to 307 labels; the replica with its fixed seed differs from ANTs’ first run in 304. The default, fix, corrects ITK-50 in this call: 1 of 5,898,240 labels changes, and the posteriors by up to 0.038.

fMRIPrep 25.2.6’s brain extraction on OpenNeuro ds000005 sub-01 segments the N4-corrected T1w into three classes with Atropos; its inputs and arguments given to the repository’s ANTs 2.6.5 at 1 to 8 threads, with correctly rounded maths, with the clock’s seed, to fMRIPrep’s own ANTs at one thread, and to the replica in both modes.

Data
OpenNeuro ds000005 (Tom et al., 2007), snapshot 00001, sub-01’s T1w (PDDL-1.0); shown inside fMRIPrep’s brain mask, dilated by 2 voxels
Node
fMRIPrep 25.2.6, anat_fit_wf/brain_extraction_wf/atropos_wf/01_atropos, run with 8 threads
Command
Atropos --image-dimensionality 3 --initialization KMeans[3] --intensity-image /work/fmriprep_25_2_wf/sub_01_wf/anat_fit_wf/brain_extraction_wf/inu_n4/mapflow/_inu_n40/sub-01_T1w_noise_corrected_maths_corrected.nii.gz --likelihood-model Gaussian --mask-image /work/fmriprep_25_2_wf/sub_01_wf/anat_fit_wf/brain_extraction_wf/atropos_wf/get_brainmask/tpl-OASIS30ANTs_res-01_label-brain_probseg_trans_resampled_maths_maths.nii.gz --mrf [0.1,1x1x1] --convergence [3,0] --output [sub-01_T1w_noise_corrected_maths_corrected_labeled.nii.gz,POSTERIOR_%02d.nii.gz] --use-random-seed 0
Programs
ANTs v2.6.5 (fdce4d2f84) on ITK v5.4.5 (f51594ad88), x86_64-conda-linux-gnu-g++ (conda-forge gcc 14.3.0-20) 14.3.0; ANTs 2.6.2.dev1-gff61a3d (conda-forge) in fMRIPrep’s image; CORE-MATH preloaded where stated; the replica at commit d065bb4d4727
ITK-50
The replica’s default, fix, changes 1 of 5,898,240 labels and the posteriors by up to 0.0383; ANTs’ own runs with the clock’s seed differ in up to 307 labels
Grid
160 × 192 × 192 voxels of 1 × 1.33 × 1.33 mm; shown: 151 × 125 × 111 around the brain

Data shown

  • Mixed-gambles task (OpenNeuro ds000005, version 00001), Open Data Commons Public Domain Dedication and Licence 1.0.
  • OASIS30ANTs template (TemplateFlow, version tpl-OASIS30ANTs git e37b3f4e0e50b6c2f79e9e01e8f1aeffcb69fc7f), Creative Commons Attribution 4.0 (CC BY 4.0). Attribution: Avants BB, Tustison NJ, Song G, Cook PA, Klein A, Gee JC. OASIS30 - Generated with ANTs for MICCAI 2016. https://doi.org/10.1016/j.neuroimage.2010.09.025; https://doi.org/10.1162/jocn.2007.19.9.1498; https://figshare.com/articles/ANTs_ANTsR_Brain_Templates/915436. Distributed by TemplateFlow: Ciric R, et al. TemplateFlow: FAIR-sharing of multi-scale, multi-species brain models. Nat Methods 19, 1568-1571 (2022). doi:10.1038/s41592-022-01681-2 The images shown are adapted from it: reoriented to RAS, cropped to the region shown, masked outside it, and shown as the outputs of the programs compared.
Inputds000005 sub-01 T1w, N4-correctedAtropos' input in fMRIPrep's brain extraction (float32)
OriginalANTs Atropos 2.6.51 thread, as built · the repository's build (GCC 14.3), glibc's maths
fMRIPrep's outputfMRIPrep's outputfMRIPrep 25.2.6, as it ran: ANTs in its image, 8 threads
Replicareplicalarmorx, Apache-2.0 · 1 thread, reproduce · against ANTs Atropos 2.6.5
the original against itself: the clock's seed and a fixed seedANTs Atropos 2.6.5: the clock's seed − sMRIPrep's fixed seedscale ±0.139
the replica against the originalreplica − ANTs Atropos 2.6.5, 1 thread, as builtscale ±0.139
the replica against itself: fix and reproduce (ITK-50)replica, fix − reproduce (ITK-50)scale ±0.139

Axial slice 67 of 111; the crosshair where ANTs with the clock's seed differs most from its fixed-seed run in it; the differences on one symmetric log scale over six decades below the largest. The interactive viewer needs JavaScript; the metrics below are complete without it.

What separates fMRIPrep's output from the replica's

    • fMRIPrep 25.2.6, as it ranANTs in its image, 8 threads
    • ANTs Atropos 2.6.5, 8 threadsas built (fMRIPrep ran 8)
    • ANTs Atropos 2.6.5, 1 thread, as built1, 2, 4 and 8 threads: the same bytes
    • ANTs Atropos 2.6.5, correctly rounded maths, 1 threadthe same binary, CORE-MATH preloaded
    • replica, reproduce1 and 6 threads

Each box holds runs that give the same bytes. Atropos’ result does not depend on ANTs’ thread count, and fMRIPrep’s own run is the repository’s ANTs result; the replica differs from it only where glibc’s maths is not correctly rounded, in the last bits of a few posteriors. Its default, fix, then corrects ITK-50 (the metrics below), a change smaller than ANTs’ own spread when it seeds from the clock.

Every comparison, computed in float64 over every voxel of the outputs

Comparison (A − B), verdict and reasonOutputmax |Δ|Values that differRMS of Δmax |Δ| ÷ range1 − NCC
The original against itself: the yardstick
ANTs Atropos 2.6.5, 8 threads − 1 threadreferenceThe same bytes: Atropos’ result does not depend on ANTs’ thread count.every outputidentical: 0 of 5,898,240 values differ
ANTs Atropos 2.6.5: the clock's seed − sMRIPrep's fixed seedreferenceAtropos seeds its generator from the clock unless --use-random-seed 0 (sMRIPrep passes 0): the ICM visiting order and the random tie-breaks change from run to run.Classes (1 CSF, 2 GM, 3 WM)1250 of 5,898,240 (0.00424 %)0.006510.3333.48e−5
Posterior: CSF0.117736,832 of 5,898,240 (12.5 %)0.0005850.1176.17e−6
Posterior: grey matter0.139793,618 of 5,898,240 (13.5 %)0.0007050.145.52e−6
Posterior: white matter0.139673,316 of 5,898,240 (11.4 %)0.0003940.1391.78e−6
ANTs Atropos in fMRIPrep's image − 2.6.5, 1 thread each, as builtreferenceANTs’ version and build play no part at one thread.every outputidentical: 0 of 5,898,240 values differ
ANTs Atropos 2.6.5, 2 threads − 1 threadreferenceThe same bytes: Atropos’ result does not depend on ANTs’ thread count.every outputidentical: 0 of 5,898,240 values differ
ANTs Atropos 2.6.5, 4 threads − 1 threadreferenceThe same bytes: Atropos’ result does not depend on ANTs’ thread count.every outputidentical: 0 of 5,898,240 values differ
ANTs Atropos 2.6.5, the clock's seed: run 2 − run 1referenceAtropos seeds its generator from the clock unless --use-random-seed 0 (sMRIPrep passes 0): the ICM visiting order and the random tie-breaks change from run to run.Classes (1 CSF, 2 GM, 3 WM)1253 of 5,898,240 (0.00429 %)0.006550.3333.52e−5
Posterior: CSF0.174752,010 of 5,898,240 (12.7 %)0.0005950.1746.39e−6
Posterior: grey matter0.174795,323 of 5,898,240 (13.5 %)0.0006930.1755.34e−6
Posterior: white matter0.0963758,018 of 5,898,240 (12.9 %)0.0003550.09631.46e−6
ANTs Atropos 2.6.5, the clock's seed: run 3 − run 1referenceAtropos seeds its generator from the clock unless --use-random-seed 0 (sMRIPrep passes 0): the ICM visiting order and the random tie-breaks change from run to run.Classes (1 CSF, 2 GM, 3 WM)1307 of 5,898,240 (0.0052 %)0.007210.3334.27e−5
Posterior: CSF0.182755,341 of 5,898,240 (12.8 %)0.0006680.1828.03e−6
Posterior: grey matter0.182795,559 of 5,898,240 (13.5 %)0.0007660.1836.52e−6
Posterior: white matter0.0844750,363 of 5,898,240 (12.7 %)0.0003750.08441.62e−6
fMRIPrep's output − ANTs Atropos 2.6.5, 8 threadsreferencefMRIPrep’s own run gives the repository’s ANTs result bit for bit.every outputidentical: 0 of 5,898,240 values differ
The replica against the original and fMRIPrep's output
replica − ANTs Atropos 2.6.5, 1 thread, as builtIdentical with correctly rounded mathsThe classified images are identical; a few posteriors differ in their last bits where glibc’s expf and logf are not correctly rounded for a value the run meets: with CORE-MATH preloaded, the same ANTs binary gives the replica’s output bit for bit. platform mathsClasses (1 CSF, 2 GM, 3 WM)00 of 5,898,240000
Posterior: CSF5.96e−889 of 5,898,240 (0.00151 %)5.59e−115.96e−85.62e−20
Posterior: grey matter1.19e−753 of 5,898,240 (0.000899 %)7.73e−111.2e−76.63e−20
Posterior: white matter5.96e−823 of 5,898,240 (0.00039 %)4.07e−115.96e−81.91e−20
replica − ANTs Atropos 2.6.5, 1 thread, correctly rounded mathsBit-identicalWith correctly rounded maths, ANTs computes exactly what the replica computes.every outputidentical: 0 of 5,898,240 values differ
replica, seed 121212 − ANTs Atropos, the clock's seed (run 1)Within the original's own variabilityWith --use-random-seed 1 ANTs draws its seed from the clock and the replica uses 121212: another draw. The replica’s labels differ from ANTs’ run 1 in 304 voxels, ANTs’ own runs from each other in up to 307; the posteriors by up to 0.119, ANTs’ own by up to 0.182.Classes (1 CSF, 2 GM, 3 WM)1304 of 5,898,240 (0.00515 %)0.007180.3334.23e−5
Posterior: CSF0.119750,097 of 5,898,240 (12.7 %)0.0006090.1196.68e−6
Posterior: grey matter0.119795,053 of 5,898,240 (13.5 %)0.0007310.1195.94e−6
Posterior: white matter0.0793726,373 of 5,898,240 (12.3 %)0.0004050.07931.89e−6
replica − fMRIPrep's outputIdentical with correctly rounded mathsThe classified images are identical; a few posteriors differ in their last bits where glibc’s expf and logf are not correctly rounded for a value the run meets: with CORE-MATH preloaded, the same ANTs binary gives the replica’s output bit for bit. fMRIPrep’s own output is ANTs 2.6.5’s at 8 threads, which is its result at one thread. platform mathsClasses (1 CSF, 2 GM, 3 WM)00 of 5,898,240000
Posterior: CSF5.96e−889 of 5,898,240 (0.00151 %)5.59e−115.96e−85.62e−20
Posterior: grey matter1.19e−753 of 5,898,240 (0.000899 %)7.73e−111.2e−76.63e−20
Posterior: white matter5.96e−823 of 5,898,240 (0.00039 %)4.07e−115.96e−81.91e−20
The replica against itself: thread counts, fix and reproduce
replica, 6 threads − 1 threadBit-identicalThe replica’s result does not depend on its thread count.every outputidentical: 0 of 5,898,240 values differ
replica, fix − reproduce (ITK-50)Deliberate correctionITK-50, corrected by default with a warning: ITK’s k-means leaves out the point each divided k-d tree node is split at, so the starting centres miss those points; the replica counts every point. In reproduce, the replica gives ANTs’ result. ITK-50Classes (1 CSF, 2 GM, 3 WM)11 of 5,898,240 (1.7e−5 %)0.0004120.3331.39e−7
Posterior: CSF0.0383738,291 of 5,898,240 (12.5 %)2.4e−50.03831.04e−8
Posterior: grey matter0.0383795,884 of 5,898,240 (13.5 %)2.41e−50.03846.43e−9
Posterior: white matter0.00113674,550 of 5,898,240 (11.4 %)1.28e−60.001131.73e−11
replica, fix, 6 threads − 1 threadBit-identicalThe same bytes on any number of threads.every outputidentical: 0 of 5,898,240 values differ

Δ is A − B. Every number is computed in float64 over all 5,898,240 voxels of the 160 × 192 × 192 outputs, not only the brain shown; the range is B’s maximum minus its minimum. The rows the original against itself are ANTs’ own variability on this input: its thread counts, its builds and its runs with the clock’s seed.

Provenance of this example
  • Data: OpenNeuro ds000005 (Tom et al., 2007), snapshot 00001, sub-01, PDDL-1.0
  • Inputs: N4-corrected T1w (sha256 3f9a610cd436863c…); brain mask (sha256 88951bc71271901f…); fMRIPrep’s POSTERIOR_01.nii.gz (sha256 f3c1068ede36eaca…); fMRIPrep’s POSTERIOR_02.nii.gz (sha256 aa3556ba097eb997…); fMRIPrep’s POSTERIOR_03.nii.gz (sha256 6ad6730b100f5887…); fMRIPrep’s sub-01_T1w_noise_corrected_maths_corrected_labeled.nii.gz (sha256 8adb20583f8430fe…)
  • Runs: figures cache entry ants-atropos/fmriprep-ds000005-atropos, computed 2026-10-22T17:36:55+00:00 with larmorx d065bb4d4727; ANTs v2.6.5 (fdce4d2f84) on ITK v5.4.5 (f51594ad88), x86_64-conda-linux-gnu-g++ (conda-forge gcc 14.3.0-20) 14.3.0; ANTs Version: 2.6.2.dev1-gff61a3d Compiled: Jul 31 2025 06:55:26; every command, its environment, threads and exit status are in its RUN_INFO.json

Every number above is in ants-atropos.json, exported by python -m larmorx_validation evidence ants-atropos.

A segmentation with tissue priors

The same T1w segmented as antsCorticalThickness segments one: OASIS30ANTs’ CSF, cortical grey matter and white matter priors, brought into the T1w’s space with the transform fMRIPrep computed, weigh 0.25 against the likelihoods.

Worked example

A segmentation of the same T1w with tissue priors

What this shows: In reproduce, on 1 and 6 threads, the replica gives ANTs’ result with correctly rounded maths bit for bit. ANTs as built gives the same classified image; the grey matter’s posterior differs in its last bits (by at most 1.2e-07), where glibc’s expf and logf are not correctly rounded. No correction by default changes this call.

The same T1w segmented as antsCorticalThickness segments one: OASIS30ANTs’ CSF, cortical grey matter and white matter priors, brought into the T1w’s space with the transform fMRIPrep’s brain extraction computed, weigh 0.25 against the likelihoods; MRF 0.1, five iterations. Shown: the classified image and the grey matter’s posterior, beside its prior.

Data
OpenNeuro ds000005 (Tom et al., 2007), snapshot 00001, sub-01’s T1w (PDDL-1.0); OASIS30ANTs (Avants et al., MICCAI 2016), from TemplateFlow (CC-BY 4.0)
Command
Atropos -d 3 -a <fMRIPrep run>/ds000005-sub-01-noreconall/work/fmriprep_25_2_wf/sub_01_wf/anat_fit_wf/brain_extraction_wf/inu_n4/mapflow/_inu_n40/sub-01_T1w_noise_corrected_maths_corrected.nii.gz -x <fMRIPrep run>/ds000005-sub-01-noreconall/work/fmriprep_25_2_wf/sub_01_wf/anat_fit_wf/brain_extraction_wf/atropos_wf/get_brainmask/tpl-OASIS30ANTs_res-01_label-brain_probseg_trans_resampled_maths_maths.nii.gz -i PriorProbabilityImages[3,<the OASIS30ANTs priors in the T1w's space>/prior%02d.nii.gz,0.25] -k Gaussian -m [0.1,1x1x1] -c [5,0] -g 1 -p Socrates[1] -r 0 -e 0 -o [out.nii.gz,out-post%02d.nii.gz]
Programs
ANTs v2.6.5 (fdce4d2f84) on ITK v5.4.5 (f51594ad88), x86_64-conda-linux-gnu-g++ (conda-forge gcc 14.3.0-20) 14.3.0; CORE-MATH preloaded where stated; the replica at commit dc5e4b828079
Grid
160 × 192 × 192 voxels of 1 × 1.33 × 1.33 mm; shown: 151 × 125 × 111 around the brain

Data shown

  • Mixed-gambles task (OpenNeuro ds000005, version 00001), Open Data Commons Public Domain Dedication and Licence 1.0.
  • OASIS30ANTs template (TemplateFlow, version tpl-OASIS30ANTs git e37b3f4e0e50b6c2f79e9e01e8f1aeffcb69fc7f), Creative Commons Attribution 4.0 (CC BY 4.0). Attribution: Avants BB, Tustison NJ, Song G, Cook PA, Klein A, Gee JC. OASIS30 - Generated with ANTs for MICCAI 2016. https://doi.org/10.1016/j.neuroimage.2010.09.025; https://doi.org/10.1162/jocn.2007.19.9.1498; https://figshare.com/articles/ANTs_ANTsR_Brain_Templates/915436. Distributed by TemplateFlow: Ciric R, et al. TemplateFlow: FAIR-sharing of multi-scale, multi-species brain models. Nat Methods 19, 1568-1571 (2022). doi:10.1038/s41592-022-01681-2 The images shown are adapted from it: reoriented to RAS, cropped to the region shown, masked outside it, and shown as the outputs of the programs compared.
Inputds000005 sub-01 T1w, N4-correctedAtropos' input
InputOASIS30ANTs prior: cortical grey matterin the T1w's space (fMRIPrep's transform, Gaussian interpolation)
OriginalANTs Atropos 2.6.51 thread, as built · the repository's build (GCC 14.3), glibc's maths
Replicareplicalarmorx, Apache-2.0 · 1 thread, reproduce · against ANTs Atropos 2.6.5
the replica against the originalreplica − ANTs Atropos 2.6.5, 1 thread, as builtscale ±1.19e−7

Axial slice 67 of 111; the crosshair where the replica and ANTs as built differ most in it. The interactive viewer needs JavaScript; the metrics below are complete without it.

Every comparison, computed in float64 over every voxel of the outputs

Comparison (A − B), verdict and reasonOutputmax |Δ|Values that differRMS of Δmax |Δ| ÷ range1 − NCC
The replica against the original
replica − ANTs Atropos 2.6.5, 1 thread, as builtIdentical with correctly rounded mathsThe classified images are identical; some posteriors differ in their last bits where glibc’s expf and logf are not correctly rounded for a value the run meets; with CORE-MATH preloaded the outputs are bit-identical. platform mathsClasses (1 CSF, 2 GM, 3 WM)00 of 5,898,240000
Posterior: grey matter1.19e−71,079 of 5,898,240 (0.0183 %)3.75e−101.19e−72.02e−18
replica − ANTs Atropos 2.6.5, 1 thread, correctly rounded mathsBit-identicalWith correctly rounded maths, ANTs computes exactly what the replica computes.both outputsidentical: 0 of 5,898,240 values differ
The replica against itself: thread counts, fix and reproduce
replica, 6 threads − 1 threadBit-identicalThe replica’s result does not depend on its thread count.both outputsidentical: 0 of 5,898,240 values differ
replica, fix − reproduceBit-identicalNo defect the replica corrects is met: the default, fix, gives ANTs’ result.both outputsidentical: 0 of 5,898,240 values differ

Δ is A − B. Every number is computed in float64 over all 5,898,240 voxels of the outputs, not only the brain shown.

Provenance of this example
  • Data: OpenNeuro ds000005 (Tom et al., 2007), snapshot 00001, sub-01, PDDL-1.0; OASIS30ANTs (Avants et al., MICCAI 2016), from TemplateFlow, CC-BY 4.0
  • Inputs: OASIS30ANTs CSF prior (sha256 ebbe851b6d5898c5…); OASIS30ANTs CGM prior (sha256 c4be7df569fa12d0…); OASIS30ANTs WM prior (sha256 3579a79d197cb3e8…); N4-corrected T1w (sha256 3f9a610cd436863c…); brain mask (sha256 88951bc71271901f…)
  • Runs: figures cache entry ants-atropos/ds000005-oasis-priors, computed 2026-10-22T17:08:39+00:00 with larmorx dc5e4b828079; ANTs v2.6.5 (fdce4d2f84) on ITK v5.4.5 (f51594ad88), x86_64-conda-linux-gnu-g++ (conda-forge gcc 14.3.0-20) 14.3.0; every command, its environment, threads and exit status are in its RUN_INFO.json

Every number above is in ants-atropos.json, exported by python -m larmorx_validation evidence ants-atropos.

Partial-volume classes with the MRF

With --use-partial-volume-likelihoods 1, Atropos adds a class for each pair of tissues given with -s (here 1x2 and 2x3), for the voxels that mix them. In ANTs’ MRF a partial-volume label is +1 against every neighbour, its own two tissues included, while a tissue next to its own kind is −2. So with any MRF the partial-volume classes hardly win a voxel, and Atropos stops when one wins none (“2 non-zero label sets but 5 is requested”; sMRIPrep’s MRF of 0.1 on a real T1w stops this way). The replica corrects it by default (ANTS-106): a partial-volume label counts as a member of its own set, −1 with its tissues, −2 with itself, +1 with any other label; reproduce keeps ANTs’ energies. A phantom whose voxels’ tissue fractions are known measures what the correction does.

Worked example

Partial-volume classes with the MRF, on a phantom of known fractions

What this shows: In reproduce the replica gives ANTs’ labels voxel for voxel, as built and with correctly rounded maths. Its default, fix, corrects ANTS-106 (with ITK-50 and ANTS-96): against the known fractions it labels 97.7 % of the voxels inside the phantom’s mask as the truth, ANTs 94.0 %. The two partial-volume classes get 1,418 and 454 voxels; the truth has 1,368 and 408, and ANTs gives them 237 and 106.

The record’s partial-volume phantom segmented with classes 1x2 and 2x3, their likelihoods and an MRF of 0.1: ANTs gives the partial-volume classes few voxels (ANTS-106); the replica’s default, fix, with a partial-volume label a member of its own set, labels the boundaries between tissues as their mixtures. The truth comes from the fractions the phantom was built with.

Data
synthetic (the comparison suite’s atropos-pv): nested spheres of radii 8.3, 13.7 and 17.9 voxels (200, 130, 60 over 0), 40³ voxels of 1 mm
Command
Atropos -d 3 -a pv.nii.gz -x pv-mask.nii.gz -i KMeans[3] -s 1x2 -s 2x3 --use-partial-volume-likelihoods 1 -m [0.1,1x1x1] -c [5,0] -r 0 -o [labels.nii.gz,post%02d.nii.gz]
Programs
ANTs v2.6.5 (fdce4d2f84) on ITK v5.4.5 (f51594ad88), x86_64-conda-linux-gnu-g++ (conda-forge gcc 14.3.0-20) 14.3.0; the replica at commit 15fd54702655
Classes
0 background, 1-3 tissues (dark to bright), 4 = 1x2, 5 = 2x3
Inputthe partial-volume phantomthree tissues box-averaged from a grid five times finer, noise sd 6
Inputthe truthfrom the known fractions: a tissue where one share is at least 0.8, else the partial-volume class of the two largest
OriginalANTs Atropos 2.6.51 thread · as built
Replicareplicalarmorx, Apache-2.0 · fix (the default), 6 threads · against ANTs Atropos 2.6.5
the original against the known answerANTs Atropos − the truthscale ±3
the replica against the originalreplica, fix − ANTs Atropos as builtscale ±3
the replica against the known answerreplica, fix − the truthscale ±3

Axial slice 20 of 40; the phantom, the truth and each program's labels. The interactive viewer needs JavaScript; the metrics below are complete without it.

Every comparison, computed in float64 over every voxel of the outputs

Comparison (A − B), verdict and reasonmax |Δ|Values that differRMS of Δmax |Δ| ÷ range1 − NCC
The original against the known answer
ANTs Atropos − the truthreferenceANTs labels 94.0% of the mask’s 23,984 voxels as the truth; it gives the partial-volume classes 237 and 106 voxels, the truth has 1368 and 408.31,437 of 64,000 (2.25 %)0.3760.60.066
The original against itself: the yardstick
ANTs Atropos with CORE-MATH − as builtreferenceThe same labels.identical: 0 of 64,000 values differ
The replica against the original and the known answer
replica, reproduce − ANTs Atropos as builtBit-identicalThe same labels, voxel for voxel.identical: 0 of 64,000 values differ
replica, reproduce − ANTs Atropos 2.6.5, correctly rounded mathsBit-identicalThe same labels, voxel for voxel.identical: 0 of 64,000 values differ
replica, fix − ANTs Atropos as builtDeliberate correctionANTS-106 corrected by default (with ITK-50 and ANTS-96): a partial-volume label is a member of its own set in the MRF, so the boundary voxels between two tissues can take it. Against the known fractions: ANTs 94.0%, the replica’s default, fix, 97.7% of the mask’s 23,984 voxels (the table below has the Dice per class). ANTS-10631,529 of 64,000 (2.39 %)0.3930.60.0706
replica, fix − the truthreferenceThe replica’s default, fix, labels 97.7% of the mask’s 23,984 voxels as the truth; 1418 and 454 voxels in the partial-volume classes.3552 of 64,000 (0.863 %)0.2470.60.0286
The replica against itself: thread counts
replica, fix: 1 thread − 6 threadsBit-identicalThe same labels at any thread count.identical: 0 of 64,000 values differ

Δ is A − B between the label images, over every voxel of the 40³ grid.

Provenance of this example
  • Data: synthetic phantom, Apache-2.0
  • Inputs: the partial-volume phantom (sha256 db2e464ca8ebec17…); its mask (sha256 1fa2300b4f927cd6…); the truth from the known fractions (sha256 e8a727207c3a7e66…)
  • Runs: figures cache entry ants-atropos/pv-phantom, computed 2026-10-23T10:06:35+00:00 with larmorx 15fd54702655; ANTs v2.6.5 (fdce4d2f84) on ITK v5.4.5 (f51594ad88), x86_64-conda-linux-gnu-g++ (conda-forge gcc 14.3.0-20) 14.3.0; every command, its environment, threads and exit status are in its RUN_INFO.json

Every number above is in ants-atropos.json, exported by python -m larmorx_validation evidence ants-atropos.

Partial-volume classes with the MRF on the phantom of known fractions: Dice per class and the share of the mask’s voxels labelled as the truth (the record’s cases)

RunDice 1Dice 2Dice 3Dice 4 (1x2)Dice 5 (2x3)voxels of the mask as the truthVerdict
MRF 0: ANTs 2.6.50.9750.9440.9750.7280.5780.937the original
MRF 0: the replica, fix0.9710.9480.9710.7110.5940.934Deliberate correctionITK-50 and ANTS-96 corrected
MRF 0.1: ANTs 2.6.50.980.9510.9640.2950.4050.94the original
MRF 0.1: the replica, fix0.9910.9830.9850.8620.8050.977Deliberate correctionANTS-106 corrected (and ITK-50, ANTS-96)
MRF 0.3: ANTs 2.6.5stops (exit 1)–––––the original
MRF 0.3: the replica, fix0.9910.9830.9860.8350.8010.978Deliberate correctionANTS-106 corrected (and ITK-50, ANTS-96)

Classes 1-3 are the tissues, 4 and 5 the partial-volume classes 1x2 and 2x3. ANTS-106: in ANTs’ MRF a partial-volume label is +1 against every neighbour, its own tissues included; corrected, it is a member of its own set (−1 with its tissues, −2 with itself, +1 with any other label). The replica’s default, fix, also corrects ITK-50 and ANTS-96 (the k-means split points, the partial-volume means), which alone change the MRF-0 row. ANTs stops when a class wins no voxel.

On the data panel

The data panel is the fixed set of fourteen OpenNeuro subjects on which LARMORX’s programs are validated: 1.5, 3 and 7 T; Siemens, GE and Philips; single-band, multiband and multi-echo BOLD; a 3.5-year-old to an 82-year-old; a tumour, a stroke with its lesion mask, strong motion. Its record replays every call fMRIPrep 25.2.6 made of the program in its fourteen runs on these subjects, the node’s own command.txt with its inputs from the run’s working directory (calls repeated word for word are compared once), and runs the program on panel images fMRIPrep never gives it. ANTs runs at one thread, as built and with correctly rounded maths, and at 2 and 3 threads for its own variability where it has any; the replica in reproduce, and in fix for the corrections. The panel’s cases are in the case explorer below (the categories data panel: fMRIPrep’s calls and data panel: other images).

17 of 17 bit-identical to ANTs with correctly rounded maths, the classes identical in all 17: sMRIPrep’s segmentation in the brain extraction of all fourteen runs, and three calls fMRIPrep never makes (the stroke outside its manual lesion mask, a child’s T1w and T2w as two channels, a FLAIR of 3 mm slices). Against fMRIPrep’s own files (ANTs 2.6.2 with glibc’s maths) the segmentation is the same in every run, and the posteriors differ in their last bits in 0 to 4,175 voxels. The panel also found a check the replica lacked: an input on another grid (a mask written from the image’s affine, its spacing 2.9e-6 mm off) makes ITK throw, and now the replica too (one more case, rejected by both).

fMRIPrep’s Atropos nodes in the fourteen fMRIPrep runs, and what the panel record compares of them

Node (normalised path)NodesRunsComparedSame call as a compared oneStatusesNot compared: why
anat_fit_wf/brain_extraction_wf/atropos_wf/01_atropos141414014 pass–

From the panel record (docs/validation/ants-atropos-panel.md): every node of the finished fMRIPrep 25.2.6 runs that ran the program, grouped by its path with the subject, run, space and item removed; a call repeated word for word is compared once.

Images on a small grid (a BOLD reference, a template at 2 mm) are shown as whole volumes; the panel’s large anatomical images (up to 17.5 million voxels) as three full-resolution slices through the voxel where the replica and ANTs as built differ most (the centre of the field of view where they agree), the whole field of view shown, defaced heads included (the data panel’s defacing check).

Worked example

Data panel: fMRIPrep's Atropos on a brain with a tumour

What this shows: In reproduce, on 1 and 8 threads, the replica gives ANTs’ bytes with correctly rounded maths; ANTs as built differs only where glibc’s maths is not correctly rounded (largest difference 1.2e-07). Its default, fix, corrects ITK-50 here.

fMRIPrep 25.2.6’s own call (sMRIPrep’s brain extraction, 01_atropos: K-means initialisation, 3 classes, MRF 0.1, 3 iterations) on ds005003 sub-RS001’s N4-corrected T1w with a 70 mL right frontal oligodendroglioma, in the brain extraction’s mask: its command.txt and inputs from the fMRIPrep run.

Data
OpenNeuro ds005003, sub-RS001 (CC0-1.0); the whole field of view shown, only margins where every array is 0 left out
Node
anat_fit/brain_extraction/atropos/01_atropos of the fMRIPrep run ds005003-sub-RS001-noreconall
Command
Atropos --image-dimensionality 3 --initialization KMeans[3] --intensity-image sub-RS001_ses-1_T1w_noise_corrected_maths_corrected.nii.gz --likelihood-model Gaussian --mask-image tpl-OASIS30ANTs_res-01_label-brain_probseg_trans_resampled_maths_maths.nii.gz --mrf [0.1,1x1x1] --convergence [3,0] --output [sub-RS001_ses-1_T1w_noise_corrected_maths_corrected_labeled.nii.gz,POSTERIOR_%02d.nii.gz] --use-random-seed 0
Programs
ANTs v2.6.5 (fdce4d2f84) on ITK v5.4.5 (f51594ad88), x86_64-conda-linux-gnu-g++ (conda-forge gcc 14.3.0-20) 14.3.0; the replica at commit 1fa019a08fbb
Grid
144 × 256 × 256 voxels of 1.2 × 0.977 × 0.977 mm; shown 144 × 256 × 256, three full-resolution slices

Data shown

  • Retrospective Task/Rest fMRI Data from Tumor Patients (OpenNeuro ds005003, version 2.0.0, doi:10.18112/openneuro.ds005003.v2.0.0), CC0 1.0 public-domain dedication.
  • OASIS30ANTs template (TemplateFlow, version tpl-OASIS30ANTs git e37b3f4e0e50b6c2f79e9e01e8f1aeffcb69fc7f), Creative Commons Attribution 4.0 (CC BY 4.0). Attribution: Avants BB, Tustison NJ, Song G, Cook PA, Klein A, Gee JC. OASIS30 - Generated with ANTs for MICCAI 2016. https://doi.org/10.1016/j.neuroimage.2010.09.025; https://doi.org/10.1162/jocn.2007.19.9.1498; https://figshare.com/articles/ANTs_ANTsR_Brain_Templates/915436. Distributed by TemplateFlow: Ciric R, et al. TemplateFlow: FAIR-sharing of multi-scale, multi-species brain models. Nat Methods 19, 1568-1571 (2022). doi:10.1038/s41592-022-01681-2 The images shown are adapted from it: reoriented to RAS, cropped to the region shown, masked outside it, and shown as the outputs of the programs compared.
Inputds005003 sub-RS001, N4-corrected T1w (fMRIPrep's)sub-RS001_ses-1_T1w_noise_corrected_maths_corrected.nii.gz
OriginalANTs Atropos 2.6.51 thread · as built (GCC 14.3), glibc's maths
Replicareplicalarmorx, Apache-2.0 · reproduce, 8 threads · against ANTs Atropos 2.6.5
Replicareplicalarmorx, Apache-2.0 · fix (the default), 8 threads

Bit-identical9,437,184 of 9,437,184 voxels equal

the replica against the originalreplica − ANTs Atropos 2.6.5, 1 thread, as built

Bit-identical9,437,184 of 9,437,184 voxels equal

the replica against the original: correctly rounded mathsreplica − ANTs Atropos 2.6.5, correctly rounded maths
the replica against itself: fix and reproducereplica, fix − reproduce

Axial slice 128 of 256; the crosshair where the replica differs most from ANTs as built; the three slices through it are stored at full resolution. The interactive viewer needs JavaScript; the metrics below are complete without it.

Stored for this example: three full-resolution slices, one per plane, through display voxel (72, 128, 128): where the main comparison differs most, or the field's centre where nothing differs; every metric is computed over every voxel of the files.

Every comparison, computed in float64 over every voxel of the outputs

Comparison (A − B), verdict and reasonOutputmax |Δ|Values that differRMS of Δmax |Δ| ÷ range1 − NCC
The original against itself: the yardstick
ANTs Atropos 2.6.5, 8 threads − 1 threadreferenceThe same bytes: ANTs’ thread count changes nothing in this call.every outputidentical: 0 of 9,437,184 values differ
The replica against the original
replica − ANTs Atropos 2.6.5, 1 thread, as builtIdentical with correctly rounded mathsOnly where glibc’s maths is not correctly rounded (post1, post2, post3): with CORE-MATH preloaded, ANTs gives the replica’s bytes.classes00 of 9,437,184000
posterior 15.96e−8216 of 9,437,184 (0.00229 %)4.11e−115.96e−83.31e−20
posterior 21.19e−779 of 9,437,184 (0.000837 %)6.9e−111.2e−73.91e−20
posterior 32.98e−855 of 9,437,184 (0.000583 %)1.04e−112.98e−88.6e−22
replica − ANTs Atropos 2.6.5, correctly rounded mathsBit-identicalThe replica gives ANTs’ bytes with correctly rounded maths.every outputidentical: 0 of 9,437,184 values differ
The replica against itself: thread counts, fix and reproduce
replica, fix − reproduceDeliberate correctionThe replica’s default, fix, corrects ITK-50 (register); reproduce gives ANTs’ result. Changed: seg, post1, post2, post3.classes12 of 9,437,184 (2.12e−5 %)0.000460.3331.27e−7
posterior 10.008891,460,242 of 9,437,184 (15.5 %)4.48e−60.008893.93e−10
posterior 20.0681,618,764 of 9,437,184 (17.2 %)5.26e−50.06842.27e−8
posterior 30.0681,449,776 of 9,437,184 (15.4 %)5.23e−50.0682.17e−8
replica, 1 thread − 8 threads (reproduce)Bit-identicalThe replica’s result does not depend on its thread count.every outputidentical: 0 of 9,437,184 values differ

Δ is A − B over all 9,437,184 values of each output.

Provenance of this example
  • Data: OpenNeuro ds005003, sub-RS001, CC0-1.0
  • Inputs: input 1 (sha256 0fb91c1cc56dc134…)
  • Runs: figures cache entry ants-atropos/panel-fmriprep-tumour, computed 2026-10-26T23:26:17+00:00 with larmorx 1fa019a08fbb; ANTs v2.6.5 (fdce4d2f84) on ITK v5.4.5 (f51594ad88), x86_64-conda-linux-gnu-g++ (conda-forge gcc 14.3.0-20) 14.3.0; every command, its environment, threads and exit status are in its RUN_INFO.json

Every number above is in ants-atropos.json, exported by python -m larmorx_validation evidence ants-atropos.

Worked example

Data panel: fMRIPrep's Atropos on a brain with a stroke

What this shows: In reproduce, on 1 and 8 threads, the replica gives ANTs’ bytes with correctly rounded maths; ANTs as built differs only where glibc’s maths is not correctly rounded (largest difference 6e-08). Its default, fix, corrects ITK-50 here.

fMRIPrep 25.2.6’s own call on ds004884 sub-M2079, a chronic left-hemisphere stroke (176 mL): the lesion’s fluid-filled tissue enters the K-means classes.

Data
OpenNeuro ds004884, sub-M2079 (CC0-1.0); the whole field of view shown, only margins where every array is 0 left out
Node
anat_fit/brain_extraction/atropos/01_atropos of the fMRIPrep run ds004884-sub-M2079-noreconall
Command
Atropos --image-dimensionality 3 --initialization KMeans[3] --intensity-image sub-M2079_ses-217_acq-tfl3p2_run-3_T1w_noise_corrected_maths_corrected.nii.gz --likelihood-model Gaussian --mask-image tpl-OASIS30ANTs_res-01_label-brain_probseg_trans_resampled_maths_maths.nii.gz --mrf [0.1,1x1x1] --convergence [3,0] --output [sub-M2079_ses-217_acq-tfl3p2_run-3_T1w_noise_corrected_maths_corrected_labeled.nii.gz,POSTERIOR_%02d.nii.gz] --use-random-seed 0
Programs
ANTs v2.6.5 (fdce4d2f84) on ITK v5.4.5 (f51594ad88), x86_64-conda-linux-gnu-g++ (conda-forge gcc 14.3.0-20) 14.3.0; the replica at commit 1fa019a08fbb
Grid
192 × 256 × 256 voxels of 1 × 1 × 1 mm; shown 192 × 256 × 256, three full-resolution slices

Data shown

  • Aphasia Recovery Cohort (OpenNeuro ds004884, version 1.0.2, doi:10.18112/openneuro.ds004884.v1.0.2), CC0 1.0 public-domain dedication.
  • OASIS30ANTs template (TemplateFlow, version tpl-OASIS30ANTs git e37b3f4e0e50b6c2f79e9e01e8f1aeffcb69fc7f), Creative Commons Attribution 4.0 (CC BY 4.0). Attribution: Avants BB, Tustison NJ, Song G, Cook PA, Klein A, Gee JC. OASIS30 - Generated with ANTs for MICCAI 2016. https://doi.org/10.1016/j.neuroimage.2010.09.025; https://doi.org/10.1162/jocn.2007.19.9.1498; https://figshare.com/articles/ANTs_ANTsR_Brain_Templates/915436. Distributed by TemplateFlow: Ciric R, et al. TemplateFlow: FAIR-sharing of multi-scale, multi-species brain models. Nat Methods 19, 1568-1571 (2022). doi:10.1038/s41592-022-01681-2 The images shown are adapted from it: reoriented to RAS, cropped to the region shown, masked outside it, and shown as the outputs of the programs compared.
Inputds004884 sub-M2079, N4-corrected T1w (fMRIPrep's)sub-M2079_ses-217_acq-tfl3p2_run-3_T1w_noise_corrected_maths_corrected.nii.gz
OriginalANTs Atropos 2.6.51 thread · as built (GCC 14.3), glibc's maths
Replicareplicalarmorx, Apache-2.0 · reproduce, 8 threads · against ANTs Atropos 2.6.5
Replicareplicalarmorx, Apache-2.0 · fix (the default), 8 threads

Bit-identical12,582,912 of 12,582,912 voxels equal

the replica against the originalreplica − ANTs Atropos 2.6.5, 1 thread, as built

Bit-identical12,582,912 of 12,582,912 voxels equal

the replica against the original: correctly rounded mathsreplica − ANTs Atropos 2.6.5, correctly rounded maths
the replica against itself: fix and reproducereplica, fix − reproduce

Axial slice 128 of 256; the crosshair where the replica differs most from ANTs as built; the three slices through it are stored at full resolution. The interactive viewer needs JavaScript; the metrics below are complete without it.

Stored for this example: three full-resolution slices, one per plane, through display voxel (96, 128, 128): where the main comparison differs most, or the field's centre where nothing differs; every metric is computed over every voxel of the files.

Every comparison, computed in float64 over every voxel of the outputs

Comparison (A − B), verdict and reasonOutputmax |Δ|Values that differRMS of Δmax |Δ| ÷ range1 − NCC
The original against itself: the yardstick
ANTs Atropos 2.6.5, 8 threads − 1 threadreferenceThe same bytes: ANTs’ thread count changes nothing in this call.every outputidentical: 0 of 12,582,912 values differ
The replica against the original
replica − ANTs Atropos 2.6.5, 1 thread, as builtIdentical with correctly rounded mathsOnly where glibc’s maths is not correctly rounded (post1, post2, post3): with CORE-MATH preloaded, ANTs gives the replica’s bytes.classes00 of 12,582,912000
posterior 15.96e−86 of 12,582,912 (4.77e−5 %)1.68e−115.96e−84.29e−21
posterior 23.73e−93 of 12,582,912 (2.38e−5 %)1.49e−123.74e−92.9e−23
posterior 39.09e−132 of 12,582,912 (1.59e−5 %)2.56e−169.1e−138.25e−31
replica − ANTs Atropos 2.6.5, correctly rounded mathsBit-identicalThe replica gives ANTs’ bytes with correctly rounded maths.every outputidentical: 0 of 12,582,912 values differ
The replica against itself: thread counts, fix and reproduce
replica, fix − reproduceDeliberate correctionThe replica’s default, fix, corrects ITK-50 (register); reproduce gives ANTs’ result. Changed: seg, post1, post2, post3.classes17 of 12,582,912 (5.56e−5 %)0.0007460.3334.95e−7
posterior 10.051,380,771 of 12,582,912 (11 %)5.11e−50.053.95e−8
posterior 20.051,589,360 of 12,582,912 (12.6 %)6.61e−50.05015.74e−8
posterior 30.04631,397,380 of 12,582,912 (11.1 %)4.19e−50.04632.2e−8
replica, 1 thread − 8 threads (reproduce)Bit-identicalThe replica’s result does not depend on its thread count.every outputidentical: 0 of 12,582,912 values differ

Δ is A − B over all 12,582,912 values of each output.

Provenance of this example
  • Data: OpenNeuro ds004884, sub-M2079, CC0-1.0
  • Inputs: input 1 (sha256 744689d057da9c5e…)
  • Runs: figures cache entry ants-atropos/panel-fmriprep-stroke, computed 2026-10-26T23:29:29+00:00 with larmorx 1fa019a08fbb; ANTs v2.6.5 (fdce4d2f84) on ITK v5.4.5 (f51594ad88), x86_64-conda-linux-gnu-g++ (conda-forge gcc 14.3.0-20) 14.3.0; every command, its environment, threads and exit status are in its RUN_INFO.json

Every number above is in ants-atropos.json, exported by python -m larmorx_validation evidence ants-atropos.

Every case

Each case of the two records, the standard comparison and the data panel, with what it tests, the command line, its verdict and the reason. Search, filter by category or verdict (or select a segment of the chart above), sort by a column; #case=<name> links to one case.

115 cases

CaseCategoryWhat it testsVerdictChecks
kmeans-3initializationsK-means, 3 classes, MRF 0.1 (sMRIPrep's options on the phantom)
command lineAtropos -i KMeans[3] -d 3 -a '<generated: atropos-t1>' -x '<generated: atropos-mask>' -c [3,0] -r 0 -m [0.1,1x1x1] -o [seg.nii.gz,post%02d.nii.gz]
Identical with correctly rounded maths

post02: as built, 1 of 63360 differ; max |diff| 1.82e-12, 1 - NCC 0; bit-identical with correctly rounded maths

29/29
kmeans-4initializationsK-means, 4 classes, MRF 0.3
command lineAtropos -i KMeans[4] -d 3 -a '<generated: atropos-t1>' -x '<generated: atropos-mask>' -c [3,0] -r 0 -m [0.3,1x1x1] -o [seg.nii.gz,post%02d.nii.gz]
Identical with correctly rounded maths

post01: as built, 42 of 63360 differ; max |diff| 5.82e-11, 1 - NCC 0; post02: as built, 36 of 63360 differ; max |diff| 2.33e-10, 1 - NCC 1.1e-16; post03: as built, 242 of 63360 differ; max |diff| 5.96e-08, 1 - NCC 1.2e-15; post04: as built, 16 of 63360 differ; max |diff| 3.73e-09, 1 - NCC 2.4e-15; bit-identical with correctly rounded maths

33/33
kmeans-centresinitializationsK-means from given centres (50, 100, 150)
command lineAtropos -i KMeans[3,50x100x150] -d 3 -a '<generated: atropos-t1>' -x '<generated: atropos-mask>' -c [3,0] -r 0 -o [seg.nii.gz,post%02d.nii.gz]
Bit-identical24/24
otsuinitializationsOtsu thresholds, 3 classes
command lineAtropos -i Otsu[3] -d 3 -a '<generated: atropos-t1>' -x '<generated: atropos-mask>' -c [3,0] -r 0 -o [seg.nii.gz,post%02d.nii.gz]
Bit-identical24/24
randominitializationsrandom initial labels (fixed seed)
command lineAtropos -i Random[3] -d 3 -a '<generated: atropos-t1>' -x '<generated: atropos-mask>' -c [3,0] -r 0 -o [seg.nii.gz,post%02d.nii.gz]
Identical with correctly rounded maths

post01: as built, 24 of 63360 differ; max |diff| 5.96e-08, 1 - NCC 0; post02: as built, 32 of 63360 differ; max |diff| 5.96e-08, 1 - NCC 0; post03: as built, 25 of 63360 differ; max |diff| 5.96e-08, 1 - NCC 0; bit-identical with correctly rounded maths

27/27
kmeans-multivariateinitializationsK-means on two images (ANTS-90: only the first decides the initial labels)
command lineAtropos -i KMeans[3] -d 3 -a '<generated: atropos-t1>' -x '<generated: atropos-mask>' -c [3,0] -r 0 -a '<generated: atropos-t2>' -o [seg.nii.gz,post%02d.nii.gz]
Bit-identical24/24
priors-socratesposteriorsprior probability images (weight 0.5), Socrates
command lineAtropos -i 'PriorProbabilityImages[3,<generated: atropos-prior%02d>,0.5]' -d 3 -a '<generated: atropos-t1>' -x '<generated: atropos-mask>' -c [3,0] -r 0 -o [seg.nii.gz,post%02d.nii.gz]
Identical with correctly rounded maths

post01: as built, 38 of 63360 differ; max |diff| 3.73e-09, 1 - NCC 0; post02: as built, 72 of 63360 differ; max |diff| 5.96e-08, 1 - NCC 2.2e-16; post03: as built, 28 of 63360 differ; max |diff| 1.16e-10, 1 - NCC 1.1e-16; bit-identical with correctly rounded maths

27/27
priors-platoposteriorsprior probability images, Plato
command lineAtropos -i 'PriorProbabilityImages[3,<generated: atropos-prior%02d>,0.5]' -d 3 -a '<generated: atropos-t1>' -x '<generated: atropos-mask>' -c [3,0] -r 0 -p Plato -o [seg.nii.gz,post%02d.nii.gz]
Identical with correctly rounded maths

post01: as built, 41 of 63360 differ; max |diff| 3.73e-09, 1 - NCC 2.2e-16; post02: as built, 94 of 63360 differ; max |diff| 5.96e-08, 1 - NCC 0; post03: as built, 28 of 63360 differ; max |diff| 1.16e-10, 1 - NCC 1.1e-16; bit-identical with correctly rounded maths

27/27
priors-aristotleposteriorsprior probability images, Aristotle
command lineAtropos -i 'PriorProbabilityImages[3,<generated: atropos-prior%02d>,0.5]' -d 3 -a '<generated: atropos-t1>' -x '<generated: atropos-mask>' -c [3,0] -r 0 -p Aristotle -o [seg.nii.gz,post%02d.nii.gz]
Identical with correctly rounded maths

post01: as built, 55 of 63360 differ; max |diff| 2.27e-13, 1 - NCC 1.1e-16; post02: as built, 48 of 63360 differ; max |diff| 3.55e-15, 1 - NCC 0; post03: as built, 44 of 63360 differ; max |diff| 8.88e-16, 1 - NCC 0; bit-identical with correctly rounded maths

27/27
priors-sigmoidposteriorsprior probability images, Sigmoid
command lineAtropos -i 'PriorProbabilityImages[3,<generated: atropos-prior%02d>,0.5]' -d 3 -a '<generated: atropos-t1>' -x '<generated: atropos-mask>' -c [3,0] -r 0 -p Sigmoid -o [seg.nii.gz,post%02d.nii.gz]
Identical with correctly rounded maths

post01: as built, 8 of 63360 differ; max |diff| 1.16e-10, 1 - NCC 0; post02: as built, 17 of 63360 differ; max |diff| 7.28e-12, 1 - NCC 0; post03: as built, 25 of 63360 differ; max |diff| 1.14e-13, 1 - NCC 0; bit-identical with correctly rounded maths

27/27
priors-euclideanposteriorsprior probability images (weight 0.25), Euclidean distance priors
command lineAtropos -i 'PriorProbabilityImages[3,<generated: atropos-prior%02d>,0.25]' -d 3 -a '<generated: atropos-t1>' -x '<generated: atropos-mask>' -c [3,0] -r 0 -e 1 -o [seg.nii.gz,post%02d.nii.gz]
Identical with correctly rounded maths

post01: as built, 33 of 63360 differ; max |diff| 1.86e-09, 1 - NCC 1.7e-15; post02: as built, 33 of 63360 differ; max |diff| 5.82e-11, 1 - NCC 0; post03: as built, 28 of 63360 differ; max |diff| 2.33e-10, 1 - NCC 1.1e-16; bit-identical with correctly rounded maths

27/27
priors-thresholdposteriorsprior probability images with a probability threshold of 0.1
command lineAtropos -i 'PriorProbabilityImages[3,<generated: atropos-prior%02d>,0.5,0.1]' -d 3 -a '<generated: atropos-t1>' -x '<generated: atropos-mask>' -c [3,0] -r 0 -o [seg.nii.gz,post%02d.nii.gz]
Identical with correctly rounded maths

post01: as built, 30 of 63360 differ; max |diff| 1.49e-08, 1 - NCC 0; post02: as built, 40 of 63360 differ; max |diff| 5.82e-11, 1 - NCC 0; post03: as built, 39 of 63360 differ; max |diff| 1.16e-10, 1 - NCC 1.1e-15; bit-identical with correctly rounded maths

27/27
priors-vectorposteriorsthe priors as one vector image (ANTS-100: every class reads the last component's normalized copy)
command lineAtropos -i 'PriorProbabilityImages[3,<generated: atropos-priors-vector>,0.5]' -d 3 -a '<generated: atropos-t1>' -x '<generated: atropos-mask>' -c [3,0] -r 0 -o [seg.nii.gz,post%02d.nii.gz]
Identical with correctly rounded maths

post01: as built, 23 of 63360 differ; max |diff| 1.86e-09, 1 - NCC 3.3e-16; post02: as built, 35 of 63360 differ; max |diff| 5.96e-08, 1 - NCC 3.3e-16; post03: as built, 38 of 63360 differ; max |diff| 9.31e-10, 1 - NCC 0; bit-identical with correctly rounded maths

27/27
priors-minimize-memoryposteriorsprior probability images with --minimize-memory-usage and a mask (ANTS-105: ANTs crashes)
command lineAtropos -i 'PriorProbabilityImages[3,<generated: atropos-prior%02d>,0.5]' -d 3 -a '<generated: atropos-t1>' -x '<generated: atropos-mask>' -c [3,0] -r 0 -u 1 -o [seg.nii.gz,post%02d.nii.gz]
Expected divergence

ANTs aborts; the replica exits 1 with its text

4/4
priors-all-outputsposteriorsprior probability images: posteriors, likelihoods and distance priors written
command lineAtropos -i 'PriorProbabilityImages[3,<generated: atropos-prior%02d>,0.5]' -d 3 -a '<generated: atropos-t1>' -x '<generated: atropos-mask>' -c [3,0] -r 0 -o [seg.nii.gz,post%02d.nii.gz,lik%02d.nii.gz,dist%02d.nii.gz]
Identical with correctly rounded maths

post01: as built, 38 of 63360 differ; max |diff| 3.73e-09, 1 - NCC 0; post02: as built, 72 of 63360 differ; max |diff| 5.96e-08, 1 - NCC 2.2e-16; post03: as built, 28 of 63360 differ; max |diff| 1.16e-10, 1 - NCC 1.1e-16; lik01: as built, 5 of 63360 differ; max |diff| 4.66e-10, 1 - NCC 0; lik02: as built, 140 of 63360 differ; max |diff| 4.66e-10, 1 - NCC 2.4e-15; lik03: as built, 4 of 63360 diffe…

60/60
priors-adaptive-smoothingposteriorsprior probability images with adaptive smoothing (0.5) of the intensities: B-spline fits per class, 3 levels
command lineAtropos -i 'PriorProbabilityImages[3,<generated: atropos-prior%02d>,0.5]' -d 3 -a '[<generated: atropos-t1>,0.5]' -x '<generated: atropos-mask>' -c [2,0] -r 0 -b [3,1,3] -o [seg.nii.gz,post%02d.nii.gz,lik%02d.nii.gz,dist%02d.nii.gz,bsp%02d.nii.gz]
Identical with correctly rounded maths

post01: as built, 28 of 63360 differ; max |diff| 5.96e-08, 1 - NCC 0; post02: as built, 49 of 63360 differ; max |diff| 5.96e-08, 1 - NCC 0; post03: as built, 44 of 63360 differ; max |diff| 3.73e-09, 1 - NCC 0; lik01: as built, 2 of 63360 differ; max |diff| 4.66e-10, 1 - NCC 3.4e-15; lik02: as built, 46 of 63360 differ; max |diff| 1.86e-09, 1 - NCC 0; lik03: as built, 16 of 63360 differ; max |diff…

75/75
label-fast-marchingposteriorsa prior label image (holes filled by fast marching)
command lineAtropos -i 'PriorLabelImage[3,<generated: atropos-prior-label>,0.5]' -d 3 -a '<generated: atropos-t1>' -x '<generated: atropos-mask>' -c [3,0] -r 0 -o [seg.nii.gz,post%02d.nii.gz]
Identical with correctly rounded maths

post01: as built, 8 of 63360 differ; max |diff| 4.66e-10, 1 - NCC 0; post02: as built, 28 of 63360 differ; max |diff| 3.73e-09, 1 - NCC 1.1e-15; post03: as built, 34 of 63360 differ; max |diff| 1.16e-10, 1 - NCC 7.8e-16; bit-identical with correctly rounded maths

27/27
label-euclideanposteriorsa prior label image, Euclidean distances
command lineAtropos -i 'PriorLabelImage[3,<generated: atropos-prior-label>,0.5]' -d 3 -a '<generated: atropos-t1>' -x '<generated: atropos-mask>' -c [3,0] -r 0 -e 1 -o [seg.nii.gz,post%02d.nii.gz]
Identical with correctly rounded maths

post01: as built, 8 of 63360 differ; max |diff| 4.66e-10, 1 - NCC 0; post02: as built, 28 of 63360 differ; max |diff| 3.73e-09, 1 - NCC 1.1e-15; post03: as built, 34 of 63360 differ; max |diff| 1.16e-10, 1 - NCC 7.8e-16; bit-identical with correctly rounded maths

27/27
label-propagationposteriorsa prior label image with label propagation for every class ([0.5,0.8])
command lineAtropos -i 'PriorLabelImage[3,<generated: atropos-prior-label>,0.5]' -d 3 -a '<generated: atropos-t1>' -x '<generated: atropos-mask>' -c [3,0] -r 0 -l [0.5,0.8] -o [seg.nii.gz,post%02d.nii.gz]
Identical with correctly rounded maths

post01: as built, 32 of 63360 differ; max |diff| 3.73e-09, 1 - NCC 2.2e-16; post02: as built, 62 of 63360 differ; max |diff| 2.98e-08, 1 - NCC 2.3e-15; post03: as built, 57 of 63360 differ; max |diff| 1.49e-08, 1 - NCC 0; bit-identical with correctly rounded maths

27/27
label-propagation-per-labelposteriorslabel propagation for labels 1 and 3
command lineAtropos -i 'PriorLabelImage[3,<generated: atropos-prior-label>,0.5]' -d 3 -a '<generated: atropos-t1>' -x '<generated: atropos-mask>' -c [3,0] -r 0 -l 1[0.3,0.9] -l 3[1,0.5] -o [seg.nii.gz,post%02d.nii.gz]
Identical with correctly rounded maths

post01: as built, 31 of 63360 differ; max |diff| 5.96e-08, 1 - NCC 1e-15; post02: as built, 36 of 63360 differ; max |diff| 5.96e-08, 1 - NCC 1.2e-15; post03: as built, 77 of 63360 differ; max |diff| 7.45e-09, 1 - NCC 4.4e-16; bit-identical with correctly rounded maths

27/27
histogram-parzenlikelihood modelshistogram Parzen windows (sigma 1, 32 bins)
command lineAtropos -i KMeans[3] -d 3 -a '<generated: atropos-t1>' -x '<generated: atropos-mask>' -c [3,0] -r 0 -k HistogramParzenWindows[1,32] -o [seg.nii.gz,post%02d.nii.gz]
Bit-identical24/24
manifold-parzenlikelihood modelsmanifold Parzen windows (k = 10; ANTS-95: the weights ignored)
command lineAtropos -i KMeans[3] -d 3 -a '<generated: atropos-t1>' -x '<generated: atropos-mask>' -c [3,0] -r 0 -k ManifoldParzenWindows[1,10] -c [2,0] -o [seg.nii.gz,post%02d.nii.gz]
Bit-identical24/24
manifold-parzen-covariancelikelihood modelsmanifold Parzen windows with local covariances (5 neighbours; ANTS-104)
command lineAtropos -i KMeans[3] -d 3 -a '<generated: atropos-t1>' -x '<generated: atropos-mask>' -c [3,0] -r 0 -k ManifoldParzenWindows[1,10,5,2] -c [1,0] -o [seg.nii.gz,post%02d.nii.gz]
Identical with correctly rounded maths

post01: as built, 1 of 63360 differ; max |diff| 1.86e-09, 1 - NCC 0; post02: as built, 1 of 63360 differ; max |diff| 1.12e-08, 1 - NCC 1.3e-15; bit-identical with correctly rounded maths

26/26
log-euclideanlikelihood modelslog-Euclidean Gaussians of six tensor components
command lineAtropos -i KMeans[3] -d 3 -a '<generated: atropos-tensor0>' -a '<generated: atropos-tensor1>' -a '<generated: atropos-tensor2>' -a '<generated: atropos-tensor3>' -a '<generated: atropos-tensor4>' -a '<generated: atropos-tensor5>' -x '<generated: atropos-mask>' -c [2,0] -r 0 -k LogEuclideanGaussian -o [seg.nii.gz,post%02d.nii.gz]
Identical with correctly rounded maths

post01: as built, 187 of 63360 differ; max |diff| 3.54e-08, 1 - NCC 0; post02: as built, 145 of 63360 differ; max |diff| 1.19e-07, 1 - NCC 0; post03: as built, 232 of 63360 differ; max |diff| 5e-16, 1 - NCC 0; bit-identical with correctly rounded maths

27/27
joint-shape-orientationlikelihood modelsjoint shape and orientation histograms of six tensor components (ANTS-103: an unset coordinate)
command lineAtropos -i KMeans[3] -d 3 -a '<generated: atropos-tensor0>' -a '<generated: atropos-tensor1>' -a '<generated: atropos-tensor2>' -a '<generated: atropos-tensor3>' -a '<generated: atropos-tensor4>' -a '<generated: atropos-tensor5>' -x '<generated: atropos-mask>' -c [2,0] -r 0 -k JointShapeAndOrientationProbability -o [seg.nii.gz,post%02d.nii.gz]
Bit-identical24/24
mrf-synchronousthe MRFsynchronous updating (no ICM)
command lineAtropos -i KMeans[3] -d 3 -a '<generated: atropos-t1>' -x '<generated: atropos-mask>' -c [3,0] -r 0 -g 0 -o [seg.nii.gz,post%02d.nii.gz]
Bit-identical24/24
mrf-radiusthe MRFan anisotropic MRF radius (2x1x2), smoothing 0.2
command lineAtropos -i KMeans[3] -d 3 -a '<generated: atropos-t1>' -x '<generated: atropos-mask>' -c [3,0] -r 0 -m [0.2,2x1x2] -o [seg.nii.gz,post%02d.nii.gz]
Identical with correctly rounded maths

post01: as built, 5 of 63360 differ; max |diff| 2.91e-11, 1 - NCC 3.3e-16; post02: as built, 23 of 63360 differ; max |diff| 3.64e-12, 1 - NCC 0; post03: as built, 7 of 63360 differ; max |diff| 1.78e-15, 1 - NCC 0; bit-identical with correctly rounded maths

27/27
mrf-coefficient-imagethe MRFan MRF coefficient image
command lineAtropos -i KMeans[3] -d 3 -a '<generated: atropos-t1>' -x '<generated: atropos-mask>' -c [3,0] -r 0 -m '[<generated: atropos-mrf-coefficients>,1x1x1]' -o [seg.nii.gz,post%02d.nii.gz]
Identical with correctly rounded maths

post01: as built, 38 of 63360 differ; max |diff| 5.96e-08, 1 - NCC 8.9e-16; post02: as built, 83 of 63360 differ; max |diff| 2.33e-10, 1 - NCC 0; post03: as built, 95 of 63360 differ; max |diff| 3.73e-09, 1 - NCC 3.3e-16; bit-identical with correctly rounded maths

27/27
mrf-nonethe MRFno MRF (smoothing 0)
command lineAtropos -i KMeans[3] -d 3 -a '<generated: atropos-t1>' -x '<generated: atropos-mask>' -c [3,0] -r 0 -m [0,1x1x1] -o [seg.nii.gz,post%02d.nii.gz]
Bit-identical24/24
mrf-icm-codesthe MRFtwo ICM iterations and the ICM code image
command lineAtropos -i KMeans[3] -d 3 -a '<generated: atropos-t1>' -x '<generated: atropos-mask>' -c [3,0] -r 0 -g 1[1,2,icm.nii.gz] -o [seg.nii.gz,post%02d.nii.gz]
Bit-identical29/29
no-proportionsposteriorsno mixture proportions (Socrates[0])
command lineAtropos -i KMeans[3] -d 3 -a '<generated: atropos-t1>' -x '<generated: atropos-mask>' -c [3,0] -r 0 -p Socrates[0] -o [seg.nii.gz,post%02d.nii.gz]
Bit-identical24/24
annealingposteriorsannealing from 2 at a rate of 0.5 down to 0.1
command lineAtropos -i KMeans[3] -d 3 -a '<generated: atropos-t1>' -x '<generated: atropos-mask>' -c [3,0] -r 0 -p Socrates[1,2,0.5,0.1] -o [seg.nii.gz,post%02d.nii.gz]
Identical with correctly rounded maths

post01: as built, 23 of 63360 differ; max |diff| 4.55e-13, 1 - NCC 0; post02: as built, 13 of 63360 differ; max |diff| 3.16e-30, 1 - NCC 0; post03: as built, 12 of 63360 differ; max |diff| 3.94e-31, 1 - NCC 0; bit-identical with correctly rounded maths

27/27
box-plotoutliersbox-plot winsorizing (ANTS-94: the filter re-runs on its own output)
command lineAtropos -i KMeans[3] -d 3 -a '<generated: atropos-t1>' -x '<generated: atropos-mask>' -c [3,0] -r 0 -w BoxPlot[0.25,0.75,1.5] -o [seg.nii.gz,post%02d.nii.gz]
Bit-identical24/24
box-plot-wideoutliersbox-plot winsorizing with a whisker of 100 (nothing beyond)
command lineAtropos -i KMeans[3] -d 3 -a '<generated: atropos-t1>' -x '<generated: atropos-mask>' -c [1,0] -r 0 -w BoxPlot[0.25,0.75,100] -o [seg.nii.gz,post%02d.nii.gz]
Bit-identical24/24
grubbs-rosneroutliersGrubbs-Rosner winsorizing, three iterations (ANTS-94; ANTS-107: from the second iteration on, ANTs weighs the grown sample with uninitialized memory)
command lineAtropos -i KMeans[3] -d 3 -a '<generated: atropos-t1>' -x '<generated: atropos-mask>' -c [3,-1] -r 0 -w GrubbsRosner[0.05,0.1] -o [seg.nii.gz,post%02d.nii.gz]
Expected divergence

ANTs weighs part of its samples with memory it never wrote (ANTS-107): its values depend on the heap and change with the output paths’ length; the replica weighs them 0. Everything else (exit status, headers, types, shapes) is compared; the record passes the case on it, the heap’s values not required, and this page counts it with the expected divergences.

20/20
grubbs-rosner-outliersoutliersGrubbs-Rosner on an image with six far outliers (ITK-51: ITK's Student quantiles are NaN, no outlier is found)
command lineAtropos -i KMeans[3] -d 3 -a '<generated: atropos-t1-outliers>' -x '<generated: atropos-mask>' -c [2,0] -r 0 -w GrubbsRosner[0.05,0.1] -o [seg.nii.gz,post%02d.nii.gz]
Bit-identical20/20
grubbs-rosner-multivariateoutliersGrubbs-Rosner on two images: the class covariance becomes singular in both
command lineAtropos -i KMeans[3] -d 3 -a '<generated: atropos-t1>' -a '<generated: atropos-t2>' -x '<generated: atropos-mask>' -c [1,0] -r 0 -w GrubbsRosner[0.2,0.3] -o [seg.nii.gz,post%02d.nii.gz]
Rejected by both

ANTs exit 1: ‘\nSLATEC/DGAMMA: X SO BIG GAMMA OVERFLOWS\nSLATEC/DGAMMA: X SO BIG GAMMA OVERFLOWS\nSLATEC/DGAMMA: X SO BIG GAMMA OVERFLOWS\nSLATEC/DGAMMA: X SO BIG GAMMA OVERFLOWS’; the replica exit 1: "x: upstream defect ANTS-94 (Atropos) reproduced: the outlier …

4/4
partial-volume-mrf-onlypartial-volume classesa partial-volume label set used by the MRF only
command lineAtropos -i KMeans[3] -d 3 -a '<generated: atropos-t1>' -x '<generated: atropos-mask>' -c [3,0] -r 0 -s 1x2 -o [seg.nii.gz,post%02d.nii.gz]
Bit-identical24/24
partial-volume-likelihoodspartial-volume classespartial-volume classes 1x2 and 2x3 with their likelihoods, no MRF
command lineAtropos -i KMeans[3] -d 3 -a '<generated: atropos-t1>' -x '<generated: atropos-mask>' -c [3,0] -r 0 -s 1x2 -s 2x3 --use-partial-volume-likelihoods 1 -p Socrates[0] -m [0,1x1x1] -o [seg.nii.gz,post%02d.nii.gz]
Bit-identical24/24
partial-volume-with-mrfpartial-volume classespartial-volume classes with the MRF (ANTS-106: they can never win a voxel; both stop)
command lineAtropos -i KMeans[3] -d 3 -a '<generated: atropos-t1>' -x '<generated: atropos-mask>' -c [3,0] -r 0 -s 1x2 -s 2x3 --use-partial-volume-likelihoods 1 -o [seg.nii.gz,post%02d.nii.gz]
Rejected by both

ANTs exit 1: ‘’; the replica exit 1: ‘rrects it]\nlarmorx: upstream defect ANTS-96 (Atropos) reproduced: the partial-volume classes keep their initial means [LARMORX_UPSTREAM_DEFECTS=fix corrects it]’

4/4
partial-volume-phantom-mrf-0partial-volume classesthe partial-volume phantom, classes 1x2 and 2x3 with their likelihoods, MRF 0 (ANTS-106), scored against the known fractions
command lineAtropos -d 3 -a '<generated: atropos-pv>' -x '<generated: atropos-pv-mask>' -i KMeans[3] -s 1x2 -s 2x3 --use-partial-volume-likelihoods 1 -m [0,1x1x1] -c [5,0] -r 0 -o [seg.nii.gz,post%02d.nii.gz]
Bit-identical26/26
partial-volume-phantom-mrf-0.1partial-volume classesthe partial-volume phantom, classes 1x2 and 2x3 with their likelihoods, MRF 0.1 (ANTS-106), scored against the known fractions
command lineAtropos -d 3 -a '<generated: atropos-pv>' -x '<generated: atropos-pv-mask>' -i KMeans[3] -s 1x2 -s 2x3 --use-partial-volume-likelihoods 1 -m [0.1,1x1x1] -c [5,0] -r 0 -o [seg.nii.gz,post%02d.nii.gz]
Bit-identical26/26
partial-volume-phantom-mrf-0.3partial-volume classesthe partial-volume phantom, classes 1x2 and 2x3 with their likelihoods, MRF 0.3 (ANTS-106), scored against the known fractions
command lineAtropos -d 3 -a '<generated: atropos-pv>' -x '<generated: atropos-pv-mask>' -i KMeans[3] -s 1x2 -s 2x3 --use-partial-volume-likelihoods 1 -m [0.3,1x1x1] -c [5,0] -r 0 -o [seg.nii.gz,post%02d.nii.gz]
Rejected by both

ANTs exit 1: ‘’; the replica exit 1: ‘rrects it]\nlarmorx: upstream defect ANTS-96 (Atropos) reproduced: the partial-volume classes keep their initial means [LARMORX_UPSTREAM_DEFECTS=fix corrects it]’

6/6
2ddimensionsa 2D slice
command lineAtropos -i KMeans[3] -d 2 -a '<generated: atropos-t1-2d>' -x '<generated: atropos-mask-2d>' -c [3,0] -r 0 -o [seg.nii.gz,post%02d.nii.gz]
Bit-identical24/24
4ddimensionsa 4D series (two volumes)
command lineAtropos -i KMeans[3] -d 4 -a '<generated: atropos-t1-4d>' -x '<generated: atropos-mask-4d>' -c [2,0] -r 0 -o [seg.nii.gz,post%02d.nii.gz]
Bit-identical24/24
mask-255inputsa mask of 255s (one LabelMap label, ITK's error)
command lineAtropos -i KMeans[3] -d 3 -a '<generated: atropos-t1>' -x '<generated: atropos-mask255>' -c [3,0] -r 0 -o [seg.nii.gz,post%02d.nii.gz]
Rejected by both

ANTs exit 1: ‘nListSampleFunction.hxx, line 105\nGaussianListSampleFunction (0x5c9d89232f10): The input list sample has <= 1 element. Function evaluations will be equal to 0.’; the replica exit 1: "x: upstream defect ANTS-98 (Atropos) reproduced: the first image’…

4/4
mask-floatinputsa float32 mask with 0.6 at its rim (read as 0)
command lineAtropos -i KMeans[3] -d 3 -a '<generated: atropos-t1>' -x '<generated: atropos-mask-float>' -c [2,0] -r 0 -o [seg.nii.gz,post%02d.nii.gz]
Bit-identical24/24
no-dimensioninputsno -d: Atropos assumes 3D (ANTS-99)
command lineAtropos -i KMeans[3] -a '<generated: atropos-t1>' -x '<generated: atropos-mask>' -c [2,0] -r 0 -o [seg.nii.gz,post%02d.nii.gz]
Bit-identical24/24
no-argumentsmessagesno arguments: the short help
command lineAtropos
Rejected by both

ANTs exit 1: ’ ). \n\n -v, --verbose (0)/1\n Verbose output. \n\n -h \n Print the help menu (short version). \n\n --help \n Print the help menu.'; the replica exit 1: ’ ). \n\n -v, --verbose (0)/1\n Verbose output. \n\n -h \n Print the help menu (short version). …

4/4
helpmessages--help
command lineAtropos --help
Bit-identical3/3
short-helpmessages-h
command lineAtropos -h
Bit-identical3/3
no-initializationmessagesno -i
command lineAtropos -d 3 -a '<generated: atropos-t1>' -x '<generated: atropos-mask>'
Rejected by both

ANTs exit 1: ‘No initialization option specified.’; the replica exit 1: ‘No initialization option specified.’

4/4
initialization-without-classesmessages-i KMeans (no classes)
command lineAtropos -d 3 -i KMeans -a '<generated: atropos-t1>' -x '<generated: atropos-mask>' -v 1
Rejected by both

ANTs exit 1: ‘usage option (see -m). With that option one can specify a prior probability threshold in which only those pixels exceeding that threshold are stored in memory.’; the replica exit 1: 'usage option (see -m). With that option one can specify a prior pr…

4/4
unknown-initializationmessages-i Foo[3]
command lineAtropos -d 3 -i Foo[3] -a '<generated: atropos-t1>' -x '<generated: atropos-mask>' -v 1
Rejected by both

ANTs exit 1: ‘Unrecognized initialization strategy request.’; the replica exit 1: ‘Unrecognized initialization strategy request.’

4/4
bad-numbermessages-i KMeans[3x] (the parser cannot convert)
command lineAtropos -d 3 -i KMeans[3x] -a '<generated: atropos-t1>' -x '<generated: atropos-mask>'
Expected divergence

ANTs aborts; the replica exits 1 with its text

4/4
wrong-centresmessagestwo centres for three classes
command lineAtropos -d 3 -i KMeans[3,10x20] -a '<generated: atropos-t1>' -x '<generated: atropos-mask>' -v 1
Rejected by both

ANTs exit 1: ‘The cluster center vector size does not equal the specified number of classes.’; the replica exit 1: ‘The cluster center vector size does not equal the specified number of classes.’

4/4
priors-without-weightmessagesPriorProbabilityImages without a weight
command lineAtropos -d 3 -i 'PriorProbabilityImages[3,<generated: atropos-prior%02d>]' -a '<generated: atropos-t1>' -x '<generated: atropos-mask>' -v 1
Rejected by both

ANTs exit 1: ‘usage option (see -m). With that option one can specify a prior probability threshold in which only those pixels exceeding that threshold are stored in memory.’; the replica exit 1: 'usage option (see -m). With that option one can specify a prior pr…

4/4
priors-missingmessagesprior images that do not exist
command lineAtropos -d 3 -i PriorProbabilityImages[3,nope%02d.nii.gz,0.5] -a '<generated: atropos-t1>' -x '<generated: atropos-mask>'
Rejected by both

ANTs exit 1: ‘file /nope01.nii.gz does not exist .’; the replica exit 1: ‘file /nope01.nii.gz does not exist .’

4/4
prior-label-missingmessagesa prior label image that does not exist
command lineAtropos -d 3 -i PriorLabelImage[3,nope.nii.gz,0.5] -a '<generated: atropos-t1>' -x '<generated: atropos-mask>'
Rejected by both

ANTs exit 1: ‘file /nope.nii.gz does not exist .’; the replica exit 1: ‘file /nope.nii.gz does not exist .’

4/4
no-maskmessagesno -x
command lineAtropos -d 3 -i KMeans[3] -a '<generated: atropos-t1>' -v 1
Rejected by both

ANTs exit 1: ‘An image mask is required. Specify a mask image with the -x option.’; the replica exit 1: ‘An image mask is required. Specify a mask image with the -x option.’

4/4
mask-missingmessagesa mask that does not exist
command lineAtropos -d 3 -i KMeans[3] -a '<generated: atropos-t1>' -x nope.nii.gz
Rejected by both

ANTs exit 1: ‘file /nope.nii.gz does not exist .’; the replica exit 1: ‘file /nope.nii.gz does not exist .’

4/4
no-imagemessagesno -a
command lineAtropos -d 3 -i KMeans[3] -x '<generated: atropos-mask>' -v 1
Rejected by both

ANTs exit 1: ‘No input images were specified. Specify an input image with the -a option.’; the replica exit 1: ‘No input images were specified. Specify an input image with the -a option.’

4/4
image-missingmessagesan image that does not exist
command lineAtropos -d 3 -i KMeans[3] -a nope.nii.gz -x '<generated: atropos-mask>'
Rejected by both

ANTs exit 1: ‘file /nope.nii.gz does not exist .’; the replica exit 1: ‘file /nope.nii.gz does not exist .’

4/4
dimension-5messages-d 5
command lineAtropos -d 5 -i KMeans[3] -a '<generated: atropos-t1>' -x '<generated: atropos-mask>'
Rejected by both

ANTs exit 1: ‘Unsupported dimension’; the replica exit 1: ‘Unsupported dimension’

4/4
negative-temperaturemessagesa negative annealing temperature
command lineAtropos -d 3 -i KMeans[3] -a '<generated: atropos-t1>' -x '<generated: atropos-mask>' -p Socrates[1,-1] -v 1
Rejected by both

ANTs exit 1: ‘Annealing temperature must be positive.’; the replica exit 1: ‘Annealing temperature must be positive.’

4/4
rate-above-onemessagesan annealing rate of 2
command lineAtropos -d 3 -i KMeans[3] -a '<generated: atropos-t1>' -x '<generated: atropos-mask>' -p Socrates[1,1,2] -v 1
Rejected by both

ANTs exit 1: ‘Annealing rate must be in the range [0, 1].’; the replica exit 1: ‘Annealing rate must be in the range [0, 1].’

4/4
wrong-levelsmessages-b with two levels in 3D
command lineAtropos -d 3 -i KMeans[3] -a '<generated: atropos-t1>' -x '<generated: atropos-mask>' -b [2x2] -v 1
Rejected by both

ANTs exit 1: ‘Incorrect number of levels’; the replica exit 1: ‘Incorrect number of levels’

4/4
wrong-meshmessages-b with a 2D mesh in 3D
command lineAtropos -d 3 -i KMeans[3] -a '<generated: atropos-t1>' -x '<generated: atropos-mask>' -b [2,1x1] -v 1
Rejected by both

ANTs exit 1: ‘Incorrect mesh resolution’; the replica exit 1: ‘Incorrect mesh resolution’

4/4
wrong-radiusmessages-m with a 2D radius in 3D
command lineAtropos -d 3 -i KMeans[3] -a '<generated: atropos-t1>' -x '<generated: atropos-mask>' -m [0.1,1x1] -v 1
Rejected by both

ANTs exit 1: ‘MRF radius size needs to be equal to the image dimension.’; the replica exit 1: ‘MRF radius size needs to be equal to the image dimension.’

4/4
unknown-likelihoodmessages-k Foo
command lineAtropos -d 3 -i KMeans[3] -a '<generated: atropos-t1>' -x '<generated: atropos-mask>' -k Foo -v 1
Rejected by both

ANTs exit 1: ‘Unrecognized likelihood model request.’; the replica exit 1: ‘Unrecognized likelihood model request.’

4/4
tensor-image-countmessagesa tensor model with one image
command lineAtropos -d 3 -i KMeans[3] -a '<generated: atropos-t1>' -x '<generated: atropos-mask>' -k LogEuclideanGaussian -v 1
Rejected by both

ANTs exit 1: ‘Expect images in upper triangular order \n xx xy xz yy yz zz \nIncorrect number of intensity images specified.’; the replica exit 1: ‘Expect images in upper triangular order \n xx xy xz yy yz zz \nIncorrect number of intensity images specified.’

4/4
partial-volume-labelsmessages-s 1x2x3 (three labels)
command lineAtropos -d 3 -i KMeans[3] -a '<generated: atropos-t1>' -x '<generated: atropos-mask>' -s 1x2x3 -v 1
Rejected by both

ANTs exit 1: ‘Error: Currently Atropos only supports partial volume label sets of size equal to 2.’; the replica exit 1: ‘Error: Currently Atropos only supports partial volume label sets of size equal to 2.’

4/4
partial-volume-label-rangemessages-s 1x4 with three classes
command lineAtropos -d 3 -i KMeans[3] -a '<generated: atropos-t1>' -x '<generated: atropos-mask>' -c [1,0] -r 0 -s 1x4 -o seg.nii.gz
Bit-identical9/9
unknown-outliersmessages-w Foo
command lineAtropos -d 3 -i KMeans[3] -a '<generated: atropos-t1>' -x '<generated: atropos-mask>' -w Foo -v 1
Rejected by both

ANTs exit 1: ‘Unrecognized outlier handling strategy request.’; the replica exit 1: ‘Unrecognized outlier handling strategy request.’

4/4
icm-bracketsmessages--icm [1,2], the documented form (ANTS-102: ANTs aborts)
command lineAtropos -d 3 -i KMeans[3] -a '<generated: atropos-t1>' -x '<generated: atropos-mask>' -g [1,2]
Expected divergence

ANTs aborts; the replica exits 1 with its text

4/4
label-propagation-without-parametersmessages-l 2 (the parser cannot convert the missing parameters: ANTs aborts)
command lineAtropos -d 3 -i KMeans[3] -a '<generated: atropos-t1>' -x '<generated: atropos-mask>' -l 2
Expected divergence

ANTs aborts; the replica exits 1 with its text

4/4
unknown-formulationmessages-p Zeno (Socrates kept, ANTS-101)
command lineAtropos -d 3 -i KMeans[3] -a '<generated: atropos-t1>' -x '<generated: atropos-mask>' -c [1,0] -r 0 -p Zeno -o seg.nii.gz
Bit-identical9/9
output-extensionmessagesa classified image with an unknown extension
command lineAtropos -d 3 -i KMeans[3] -a '<generated: atropos-t1>' -x '<generated: atropos-mask>' -c [1,0] -r 0 -o seg.foo
Expected divergence

ANTs aborts; the replica exits 1 with its text

3/4
posterior-without-percentmessagesa posterior name without %d (every class writes the same file)
command lineAtropos -d 3 -i KMeans[3] -a '<generated: atropos-t1>' -x '<generated: atropos-mask>' -c [1,0] -r 0 -o [seg.nii.gz,p.nii.gz]
Bit-identical14/14
verboseverbose output-v 1: progress, the filter's settings and its models
command lineAtropos -d 3 -i KMeans[3] -a '<generated: atropos-t1>' -x '<generated: atropos-mask>' -c [2,0] -r 0 -o [seg.nii.gz,post%02d.nii.gz] -v 1
Bit-identical26/26
verbose-grubbsverbose output-v 1 with Grubbs-Rosner (its outlier listing; one iteration, before ANTS-107)
command lineAtropos -d 3 -i KMeans[3] -a '<generated: atropos-t1>' -x '<generated: atropos-mask>' -c [1,0] -r 0 -w GrubbsRosner[0.05,0.1] -o [seg.nii.gz,post%02d.nii.gz] -v 1
Bit-identical26/26
verbose-box-plotverbose output-v 1 with box-plot winsorizing
command lineAtropos -d 3 -i KMeans[3] -a '<generated: atropos-t1>' -x '<generated: atropos-mask>' -c [2,0] -r 0 -w BoxPlot[0.25,0.75,1.5] -o [seg.nii.gz,post%02d.nii.gz] -v 1
Bit-identical26/26
verbose-priorsverbose output-v 1 with prior probability images and label propagation
command lineAtropos -d 3 -i 'PriorProbabilityImages[3,<generated: atropos-prior%02d>,0.5]' -a '<generated: atropos-t1>' -x '<generated: atropos-mask>' -c [2,0] -r 0 -l [0.5,0.8] -o [seg.nii.gz,post%02d.nii.gz] -v 1
Identical with correctly rounded maths

post01: as built, 34 of 63360 differ; max |diff| 2.98e-08, 1 - NCC 2.2e-15; post02: as built, 111 of 63360 differ; max |diff| 5.96e-08, 1 - NCC 1.8e-15; post03: as built, 60 of 63360 differ; max |diff| 1.16e-10, 1 - NCC 7.8e-16; bit-identical with correctly rounded maths

29/29
verbose-parzenverbose output-v 1 with histogram Parzen windows
command lineAtropos -d 3 -i KMeans[3] -a '<generated: atropos-t1>' -x '<generated: atropos-mask>' -c [1,0] -r 0 -k HistogramParzenWindows[1,32] -o [seg.nii.gz,post%02d.nii.gz] -v 1
Identical with correctly rounded maths

post01: as built, 1 of 63360 differ; max |diff| 1.73e-18, 1 - NCC 0; bit-identical with correctly rounded maths

27/27
clock-seedrandom seeds--use-random-seed 1: ANTs seeds from the clock (two runs); the replica uses the seed 121212
command lineAtropos -i KMeans[3] -d 3 -a '<generated: atropos-t1>' -x '<generated: atropos-mask>' -c [3,0] -r 1 -m [0.1,1x1x1] -o [seg.nii.gz,post%02d.nii.gz]
Expected divergence

ANTs draws its seed from the clock; the replica uses 121212

7/7
clock-seed-random-labelsrandom seedsRandom[3] with the clock's seed
command lineAtropos -i Random[3] -d 3 -a '<generated: atropos-t1>' -x '<generated: atropos-mask>' -c [3,0] -o [seg.nii.gz,post%02d.nii.gz]
Expected divergence

ANTs draws its seed from the clock; the replica uses 121212

7/7
real/ds000005-sub-01/smriprep worked examplereal images: sMRIPrep's callsMRIPrep's Atropos call in ds000005-sub-01 (fMRIPrep 25.2.6's command line, its N4-corrected T1w and mask)
command lineAtropos <fMRIPrep's command line: ds000005-sub-01-noreconall/anat_fit_wf/brain_extraction_wf/atropos_wf/01_atropos/command.txt>
Identical with correctly rounded maths

post01: as built, 89 of 5898240 differ; max |diff| 5.96e-08, 1 - NCC 0; post02: as built, 53 of 5898240 differ; max |diff| 1.19e-07, 1 - NCC 0; post03: as built, 23 of 5898240 differ; max |diff| 5.96e-08, 1 - NCC 0; bit-identical with correctly rounded maths

35/35
real/ds001600-sub-1/smriprepreal images: sMRIPrep's callsMRIPrep's Atropos call in ds001600-sub-1 (fMRIPrep 25.2.6's command line, its N4-corrected T1w and mask)
command lineAtropos <fMRIPrep's command line: ds001600-sub-1-noreconall/anat_fit_wf/brain_extraction_wf/atropos_wf/01_atropos/command.txt>
Bit-identical28/28
real/ds005040-sub-002/smriprepreal images: sMRIPrep's callsMRIPrep's Atropos call in ds005040-sub-002 (fMRIPrep 25.2.6's command line, its N4-corrected T1w and mask)
command lineAtropos <fMRIPrep's command line: ds005040-sub-002-noreconall/anat_fit_wf/brain_extraction_wf/atropos_wf/01_atropos/command.txt>
Identical with correctly rounded maths

post01: as built, 80 of 15842304 differ; max |diff| 5.96e-08, 1 - NCC 4.8e-15; post02: as built, 60 of 15842304 differ; max |diff| 8.94e-08, 1 - NCC 0; post03: as built, 67 of 15842304 differ; max |diff| 5.96e-08, 1 - NCC 0; bit-identical with correctly rounded maths

31/31
real/ds005454-sub-16/smriprepreal images: sMRIPrep's callsMRIPrep's Atropos call in ds005454-sub-16 (fMRIPrep 25.2.6's command line, its N4-corrected T1w and mask)
command lineAtropos <fMRIPrep's command line: ds005454-sub-16-noreconall/anat_fit_wf/brain_extraction_wf/atropos_wf/01_atropos/command.txt>
Identical with correctly rounded maths

post01: as built, 4175 of 17547264 differ; max |diff| 1.19e-07, 1 - NCC 3.6e-14; post02: as built, 3655 of 17547264 differ; max |diff| 1.19e-07, 1 - NCC 0; post03: as built, 3805 of 17547264 differ; max |diff| 1.19e-07, 1 - NCC 3.3e-16; bit-identical with correctly rounded maths

31/31
real/ds005752-sub-ON01016/smriprepreal images: sMRIPrep's callsMRIPrep's Atropos call in ds005752-sub-ON01016 (fMRIPrep 25.2.6's command line, its N4-corrected T1w and mask)
command lineAtropos <fMRIPrep's command line: ds005752-sub-ON01016-noreconall/anat_fit_wf/brain_extraction_wf/atropos_wf/01_atropos/command.txt>
Identical with correctly rounded maths

post01: as built, 11 of 12845056 differ; max |diff| 5.96e-08, 1 - NCC 2e-15; post02: as built, 23 of 12845056 differ; max |diff| 5.96e-08, 1 - NCC 7.9e-15; post03: as built, 21 of 12845056 differ; max |diff| 2.98e-08, 1 - NCC 8.9e-15; bit-identical with correctly rounded maths

31/31
real/ds000005-sub-01/priors-3 worked examplereal images: tissue priorsantsCorticalThickness-style segmentation in ds000005-sub-01: the OASIS30ANTs priors (CSF, CGM, WM) in T1w space, weight 0.25, MRF 0.1, five iterations, Socrates
command lineAtropos -d 3 -a '<ds000005-sub-01-noreconall: N4-corrected T1w>' -x '<ds000005-sub-01-noreconall: brain mask>' -i 'PriorProbabilityImages[3,<OASIS30ANTs priors%02d>,0.25]' -k Gaussian -m [0.1,1x1x1] -c [5,0] -g 1 -p Socrates[1] -r 0 -e 0 -o [seg.nii.gz,post%02d.nii.gz]
Identical with correctly rounded maths

post01: as built, 1101 of 5898240 differ; max |diff| 1.19e-07, 1 - NCC 0; post02: as built, 1079 of 5898240 differ; max |diff| 1.19e-07, 1 - NCC 0; post03: as built, 934 of 5898240 differ; max |diff| 1.19e-07, 1 - NCC 0; bit-identical with correctly rounded maths

27/27
real/ds005040-sub-002/priors-3real images: tissue priorsantsCorticalThickness-style segmentation in ds005040-sub-002: the OASIS30ANTs priors (CSF, CGM, WM) in T1w space, weight 0.25, MRF 0.1, five iterations, Socrates
command lineAtropos -d 3 -a '<ds005040-sub-002-noreconall: N4-corrected T1w>' -x '<ds005040-sub-002-noreconall: brain mask>' -i 'PriorProbabilityImages[3,<OASIS30ANTs priors%02d>,0.25]' -k Gaussian -m [0.1,1x1x1] -c [5,0] -g 1 -p Socrates[1] -r 0 -e 0 -o [seg.nii.gz,post%02d.nii.gz]
Identical with correctly rounded maths

post01: as built, 3731 of 15842304 differ; max |diff| 1.19e-07, 1 - NCC 0; post02: as built, 3056 of 15842304 differ; max |diff| 1.19e-07, 1 - NCC 7.5e-15; post03: as built, 2971 of 15842304 differ; max |diff| 1.19e-07, 1 - NCC 0; bit-identical with correctly rounded maths

27/27
real/ds000005-sub-01/smriprep-partial-volumereal images: partial-volume classessMRIPrep's options on ds000005-sub-01's N4-corrected T1w and brain mask with partial-volume classes 1x2 and 2x3 (ANTS-106: ANTs stops)
command lineAtropos -d 3 -a '<ds000005-sub-01-noreconall: N4-corrected T1w>' -x '<ds000005-sub-01-noreconall: brain mask>' -i KMeans[3] -k Gaussian -s 1x2 -s 2x3 --use-partial-volume-likelihoods 1 -m [0.1,1x1x1] -c [3,0] -r 0 -o [seg.nii.gz,post%02d.nii.gz]
Rejected by both

ANTs exit 1: ‘’; the replica exit 1: ‘rrects it]\nlarmorx: upstream defect ANTS-96 (Atropos) reproduced: the partial-volume classes keep their initial means [LARMORX_UPSTREAM_DEFECTS=fix corrects it]’

5/5
real/ds000005/otsureal images: other callsOtsu[3] on ds000005's N4-corrected T1w
command lineAtropos -d 3 -a '<ds000005-sub-01-noreconall: N4-corrected T1w>' -x '<ds000005-sub-01-noreconall: brain mask>' -i Otsu[3] -m [0.1,1x1x1] -c [3,0] -r 0 -o [seg.nii.gz,post%02d.nii.gz]
Identical with correctly rounded maths

post01: as built, 23 of 5898240 differ; max |diff| 5.96e-08, 1 - NCC 8.9e-16; post02: as built, 76 of 5898240 differ; max |diff| 1.19e-07, 1 - NCC 0; post03: as built, 37 of 5898240 differ; max |diff| 1.49e-08, 1 - NCC 1.1e-16; bit-identical with correctly rounded maths

27/27
real/ds000005/randomreal images: other callsRandom[3] on ds000005's T1w (fixed seed)
command lineAtropos -d 3 -a '<ds000005-sub-01-noreconall: N4-corrected T1w>' -x '<ds000005-sub-01-noreconall: brain mask>' -i Random[3] -m [0.1,1x1x1] -c [5,0] -r 0 -o [seg.nii.gz,post%02d.nii.gz]
Identical with correctly rounded maths

post01: as built, 39 of 5898240 differ; max |diff| 5.96e-08, 1 - NCC 2.1e-14; post02: as built, 30 of 5898240 differ; max |diff| 5.96e-08, 1 - NCC 0; post03: as built, 23 of 5898240 differ; max |diff| 5.96e-08, 1 - NCC 0; bit-identical with correctly rounded maths

27/27
real/ds000005/smriprep-clock-seedreal images: other callssMRIPrep's call with --use-random-seed 1: ANTs seeds from the clock (two ANTs runs, ANTs' seed band)
command lineAtropos -d 3 -a '<ds000005-sub-01-noreconall: N4-corrected T1w>' -x '<ds000005-sub-01-noreconall: brain mask>' -i KMeans[3] -k Gaussian -m [0.1,1x1x1] -c [3,0] -r 1 -o [seg.nii.gz,post%02d.nii.gz]
Expected divergence

ANTs draws its seed from the clock; the replica uses 121212

7/7
panel/fmriprep/ds000005-sub-01/anat_fit/brain_extraction/atropos/01_atroposdata panel: fMRIPrep's callsfMRIPrep 25.2.6's Atropos node `anat_fit/brain_extraction/atropos/01_atropos` of the fMRIPrep run ds000005-sub-01 (anat_fit_wf/brain_extraction_wf/atropos_wf/01_atropos): its command.txt replayed, its inputs from the run's working directory
command lineAtropos <fMRIPrep's command line: ds000005-sub-01-noreconall/anat_fit_wf/brain_extraction_wf/atropos_wf/01_atropos/command.txt>
Identical with correctly rounded maths

post01: as built, 89 of 5898240 differ; max |diff| 5.96e-08, 1 - NCC 2.2e-16; post02: as built, 53 of 5898240 differ; max |diff| 1.19e-07, 1 - NCC 0; post03: as built, 23 of 5898240 differ; max |diff| 5.96e-08, 1 - NCC 0; bit-identical with correctly rounded maths

35/35
panel/fmriprep/ds000228-sub-pixar008/anat_fit/brain_extraction/atropos/01_atroposdata panel: fMRIPrep's callsfMRIPrep 25.2.6's Atropos node `anat_fit/brain_extraction/atropos/01_atropos` of the fMRIPrep run ds000228-sub-pixar008 (anat_fit_wf/brain_extraction_wf/atropos_wf/01_atropos): its command.txt replayed, its inputs from the run's working directory
command lineAtropos <fMRIPrep's command line: ds000228-sub-pixar008-noreconall/anat_fit_wf/brain_extraction_wf/atropos_wf/01_atropos/command.txt>
Bit-identical28/28
panel/fmriprep/ds000256-sub-CTS203/anat_fit/brain_extraction/atropos/01_atroposdata panel: fMRIPrep's callsfMRIPrep 25.2.6's Atropos node `anat_fit/brain_extraction/atropos/01_atropos` of the fMRIPrep run ds000256-sub-CTS203 (anat_fit_wf/brain_extraction_wf/atropos_wf/01_atropos): its command.txt replayed, its inputs from the run's working directory
command lineAtropos <fMRIPrep's command line: ds000256-sub-CTS203-noreconall/anat_fit_wf/brain_extraction_wf/atropos_wf/01_atropos/command.txt>
Identical with correctly rounded maths

post01: as built, 354 of 14680064 differ; max |diff| 5.96e-08, 1 - NCC 0; post02: as built, 916 of 14680064 differ; max |diff| 1.19e-07, 1 - NCC 0; post03: as built, 460 of 14680064 differ; max |diff| 1.19e-07, 1 - NCC 0; bit-identical with correctly rounded maths

31/31
panel/fmriprep/ds001600-sub-1/anat_fit/brain_extraction/atropos/01_atroposdata panel: fMRIPrep's callsfMRIPrep 25.2.6's Atropos node `anat_fit/brain_extraction/atropos/01_atropos` of the fMRIPrep run ds001600-sub-1 (anat_fit_wf/brain_extraction_wf/atropos_wf/01_atropos): its command.txt replayed, its inputs from the run's working directory
command lineAtropos <fMRIPrep's command line: ds001600-sub-1-noreconall/anat_fit_wf/brain_extraction_wf/atropos_wf/01_atropos/command.txt>
Bit-identical28/28
panel/fmriprep/ds002422-sub-02/anat_fit/brain_extraction/atropos/01_atroposdata panel: fMRIPrep's callsfMRIPrep 25.2.6's Atropos node `anat_fit/brain_extraction/atropos/01_atropos` of the fMRIPrep run ds002422-sub-02 (anat_fit_wf/brain_extraction_wf/atropos_wf/01_atropos): its command.txt replayed, its inputs from the run's working directory
command lineAtropos <fMRIPrep's command line: ds002422-sub-02-noreconall/anat_fit_wf/brain_extraction_wf/atropos_wf/01_atropos/command.txt>
Identical with correctly rounded maths

post01: as built, 601 of 11534336 differ; max |diff| 1.19e-07, 1 - NCC 0; post02: as built, 570 of 11534336 differ; max |diff| 1.19e-07, 1 - NCC 0; post03: as built, 190 of 11534336 differ; max |diff| 5.96e-08, 1 - NCC 2.2e-16; bit-identical with correctly rounded maths

30/30
panel/fmriprep/ds002603-sub-01/anat_fit/brain_extraction/atropos/01_atroposdata panel: fMRIPrep's callsfMRIPrep 25.2.6's Atropos node `anat_fit/brain_extraction/atropos/01_atropos` of the fMRIPrep run ds002603-sub-01 (anat_fit_wf/brain_extraction_wf/atropos_wf/01_atropos): its command.txt replayed, its inputs from the run's working directory
command lineAtropos <fMRIPrep's command line: ds002603-sub-01-noreconall/anat_fit_wf/brain_extraction_wf/atropos_wf/01_atropos/command.txt>
Identical with correctly rounded maths

post02: as built, 2 of 16384000 differ; max |diff| 3.73e-09, 1 - NCC 0; bit-identical with correctly rounded maths

29/29
panel/fmriprep/ds003592-sub-280/anat_fit/brain_extraction/atropos/01_atroposdata panel: fMRIPrep's callsfMRIPrep 25.2.6's Atropos node `anat_fit/brain_extraction/atropos/01_atropos` of the fMRIPrep run ds003592-sub-280 (anat_fit_wf/brain_extraction_wf/atropos_wf/01_atropos): its command.txt replayed, its inputs from the run's working directory
command lineAtropos <fMRIPrep's command line: ds003592-sub-280-noreconall/anat_fit_wf/brain_extraction_wf/atropos_wf/01_atropos/command.txt>
Identical with correctly rounded maths

post01: as built, 11 of 12582912 differ; max |diff| 1.19e-07, 1 - NCC 1.1e-16; post02: as built, 45 of 12582912 differ; max |diff| 5.96e-08, 1 - NCC 0; post03: as built, 5 of 12582912 differ; max |diff| 2.24e-08, 1 - NCC 2.2e-16; bit-identical with correctly rounded maths

31/31
panel/fmriprep/ds004173-sub-603309/anat_fit/brain_extraction/atropos/01_atroposdata panel: fMRIPrep's callsfMRIPrep 25.2.6's Atropos node `anat_fit/brain_extraction/atropos/01_atropos` of the fMRIPrep run ds004173-sub-603309 (anat_fit_wf/brain_extraction_wf/atropos_wf/01_atropos): its command.txt replayed, its inputs from the run's working directory
command lineAtropos <fMRIPrep's command line: ds004173-sub-603309-noreconall/anat_fit_wf/brain_extraction_wf/atropos_wf/01_atropos/command.txt>
Bit-identical28/28
panel/fmriprep/ds004884-sub-M2079/anat_fit/brain_extraction/atropos/01_atroposdata panel: fMRIPrep's callsfMRIPrep 25.2.6's Atropos node `anat_fit/brain_extraction/atropos/01_atropos` of the fMRIPrep run ds004884-sub-M2079 (anat_fit_wf/brain_extraction_wf/atropos_wf/01_atropos): its command.txt replayed, its inputs from the run's working directory
command lineAtropos <fMRIPrep's command line: ds004884-sub-M2079-noreconall/anat_fit_wf/brain_extraction_wf/atropos_wf/01_atropos/command.txt>
Identical with correctly rounded maths

post01: as built, 6 of 12582912 differ; max |diff| 5.96e-08, 1 - NCC 0; post02: as built, 3 of 12582912 differ; max |diff| 3.73e-09, 1 - NCC 0; post03: as built, 2 of 12582912 differ; max |diff| 9.09e-13, 1 - NCC 0; bit-identical with correctly rounded maths

31/31
panel/fmriprep/ds004958-sub-fsm64ps/anat_fit/brain_extraction/atropos/01_atroposdata panel: fMRIPrep's callsfMRIPrep 25.2.6's Atropos node `anat_fit/brain_extraction/atropos/01_atropos` of the fMRIPrep run ds004958-sub-fsm64ps (anat_fit_wf/brain_extraction_wf/atropos_wf/01_atropos): its command.txt replayed, its inputs from the run's working directory
command lineAtropos <fMRIPrep's command line: ds004958-sub-fsm64ps-noreconall/anat_fit_wf/brain_extraction_wf/atropos_wf/01_atropos/command.txt>
Identical with correctly rounded maths

post01: as built, 31 of 11534336 differ; max |diff| 4.66e-10, 1 - NCC 0; post02: as built, 9 of 11534336 differ; max |diff| 5.96e-08, 1 - NCC 0; post03: as built, 2 of 11534336 differ; max |diff| 4.66e-10, 1 - NCC 0; bit-identical with correctly rounded maths

31/31
panel/fmriprep/ds005003-sub-RS001/anat_fit/brain_extraction/atropos/01_atroposdata panel: fMRIPrep's callsfMRIPrep 25.2.6's Atropos node `anat_fit/brain_extraction/atropos/01_atropos` of the fMRIPrep run ds005003-sub-RS001 (anat_fit_wf/brain_extraction_wf/atropos_wf/01_atropos): its command.txt replayed, its inputs from the run's working directory
command lineAtropos <fMRIPrep's command line: ds005003-sub-RS001-noreconall/anat_fit_wf/brain_extraction_wf/atropos_wf/01_atropos/command.txt>
Identical with correctly rounded maths

post01: as built, 216 of 9437184 differ; max |diff| 5.96e-08, 1 - NCC 0; post02: as built, 79 of 9437184 differ; max |diff| 1.19e-07, 1 - NCC 0; post03: as built, 55 of 9437184 differ; max |diff| 2.98e-08, 1 - NCC 0; bit-identical with correctly rounded maths

31/31
panel/fmriprep/ds005040-sub-002/anat_fit/brain_extraction/atropos/01_atroposdata panel: fMRIPrep's callsfMRIPrep 25.2.6's Atropos node `anat_fit/brain_extraction/atropos/01_atropos` of the fMRIPrep run ds005040-sub-002 (anat_fit_wf/brain_extraction_wf/atropos_wf/01_atropos): its command.txt replayed, its inputs from the run's working directory
command lineAtropos <fMRIPrep's command line: ds005040-sub-002-noreconall/anat_fit_wf/brain_extraction_wf/atropos_wf/01_atropos/command.txt>
Identical with correctly rounded maths

post01: as built, 80 of 15842304 differ; max |diff| 5.96e-08, 1 - NCC 0; post02: as built, 60 of 15842304 differ; max |diff| 8.94e-08, 1 - NCC 0; post03: as built, 67 of 15842304 differ; max |diff| 5.96e-08, 1 - NCC 0; bit-identical with correctly rounded maths

31/31
panel/fmriprep/ds005454-sub-16/anat_fit/brain_extraction/atropos/01_atroposdata panel: fMRIPrep's callsfMRIPrep 25.2.6's Atropos node `anat_fit/brain_extraction/atropos/01_atropos` of the fMRIPrep run ds005454-sub-16 (anat_fit_wf/brain_extraction_wf/atropos_wf/01_atropos): its command.txt replayed, its inputs from the run's working directory
command lineAtropos <fMRIPrep's command line: ds005454-sub-16-noreconall/anat_fit_wf/brain_extraction_wf/atropos_wf/01_atropos/command.txt>
Identical with correctly rounded maths

post01: as built, 4175 of 17547264 differ; max |diff| 1.19e-07, 1 - NCC 0; post02: as built, 3655 of 17547264 differ; max |diff| 1.19e-07, 1 - NCC 0; post03: as built, 3805 of 17547264 differ; max |diff| 1.19e-07, 1 - NCC 0; bit-identical with correctly rounded maths

31/31
panel/fmriprep/ds005752-sub-ON01016/anat_fit/brain_extraction/atropos/01_atroposdata panel: fMRIPrep's callsfMRIPrep 25.2.6's Atropos node `anat_fit/brain_extraction/atropos/01_atropos` of the fMRIPrep run ds005752-sub-ON01016 (anat_fit_wf/brain_extraction_wf/atropos_wf/01_atropos): its command.txt replayed, its inputs from the run's working directory
command lineAtropos <fMRIPrep's command line: ds005752-sub-ON01016-noreconall/anat_fit_wf/brain_extraction_wf/atropos_wf/01_atropos/command.txt>
Identical with correctly rounded maths

post01: as built, 11 of 12845056 differ; max |diff| 5.96e-08, 1 - NCC 1.1e-16; post02: as built, 23 of 12845056 differ; max |diff| 5.96e-08, 1 - NCC 0; post03: as built, 21 of 12845056 differ; max |diff| 2.98e-08, 1 - NCC 0; bit-identical with correctly rounded maths

31/31
panel/images/stroke/outside-lesiondata panel: other imagessMRIPrep's options on the stroke's N4-corrected T1w (the fMRIPrep run's), its brain mask without the manual lesion mask (ds004884 sub-M2079: the stroke's T1w; ds004884 sub-M2079: the stroke's manual lesion mask, drawn on the T2w (uint8))
command lineAtropos -d 3 -a '<ds004884-sub-M2079-noreconall: N4-corrected T1w>' -x '<generated: atropos-panel-stroke-mask>' -i KMeans[3] -k Gaussian -m [0.1,1x1x1] -c [3,0] -r 0 -o [seg.nii.gz,post%02d.nii.gz]
Identical with correctly rounded maths

post01: as built, 15 of 12582912 differ; max |diff| 7.45e-09, 1 - NCC 0; post02: as built, 2 of 12582912 differ; max |diff| 3.73e-09, 1 - NCC 0; post03: as built, 4 of 12582912 differ; max |diff| 4.55e-13, 1 - NCC 0; bit-identical with correctly rounded maths

27/27
panel/images/child/t1w-t2wdata panel: other imagessMRIPrep's options with two channels: a child's N4-corrected T1w (the fMRIPrep run's) and its T2w on the same grid, in its brain mask (ds000256 sub-CTS203: a 7.5-year-old's low-quality 0.8 mm T1w; ds000256 sub-CTS203: a 7.5-year-old's 0.8 mm T2w)
command lineAtropos -d 3 -a '<ds000256-sub-CTS203-noreconall: N4-corrected T1w>' -a '<generated: atropos-panel-child-t2w>' -x '<ds000256-sub-CTS203-noreconall: brain mask>' -i KMeans[3] -k Gaussian -m [0.1,1x1x1] -c [3,0] -r 0 -o [seg.nii.gz,post%02d.nii.gz]
Identical with correctly rounded maths

post01: as built, 340 of 14680064 differ; max |diff| 1.19e-07, 1 - NCC 0; post02: as built, 798 of 14680064 differ; max |diff| 1.19e-07, 1 - NCC 0; post03: as built, 390 of 14680064 differ; max |diff| 1.19e-07, 1 - NCC 2.2e-16; bit-identical with correctly rounded maths

27/27
panel/images/flair/kmeans-3data panel: other imagessMRIPrep's options on a FLAIR of 0.42 x 0.42 x 3 mm voxels, in its head mask (ds003592 sub-280: an 82-year-old's FLAIR (0.42 x 0.42 x 3 mm, 44 slices))
command lineAtropos -d 3 -a {in:openneuro/ds003592/sub-280/ses-1/anat/sub-280_ses-1_FLAIR.nii.gz} -x '<generated: atropos-panel-flair-mask>' -i KMeans[3] -k Gaussian -m [0.1,1x1x1] -c [3,0] -r 0 -o [seg.nii.gz,post%02d.nii.gz]
Identical with correctly rounded maths

post01: as built, 16 of 8650752 differ; max |diff| 1.19e-07, 1 - NCC 0; post02: as built, 6 of 8650752 differ; max |diff| 1.19e-07, 1 - NCC 2.2e-16; post03: as built, 18 of 8650752 differ; max |diff| 5.96e-08, 1 - NCC 0; bit-identical with correctly rounded maths

27/27
panel/images/flair/mask-from-affinedata panel: other imagesthe same with the head mask written from the FLAIR's affine (its spacing recomputed, 3e-6 mm off): ITK's VerifyInputInformation throws, ANTs exits 1 (ds003592 sub-280: an 82-year-old's FLAIR (0.42 x 0.42 x 3 mm, 44 slices))
command lineAtropos -d 3 -a {in:openneuro/ds003592/sub-280/ses-1/anat/sub-280_ses-1_FLAIR.nii.gz} -x '<generated: atropos-panel-flair-mask-affine>' -i KMeans[3] -k Gaussian -m [0.1,1x1x1] -c [3,0] -r 0 -o [seg.nii.gz,post%02d.nii.gz] -v 1
Rejected by both

ANTs exit 1: ‘InputImage Spacing: [4.2187500e-01, 4.2187500e-01, 3.0000000e+00], InputImage_1 Spacing: [4.2187503e-01, 4.2187500e-01, 2.9999971e+00]\n\tTolerance: 4.2187500e-07’; the replica exit 1: 'InputImage Spacing: [4.2187500e-01, 4.2187500e-01, 3.0000000e+0…

4/4

Corrections by default

Where ANTs has a deterministic defect that changes a result, the replica corrects it by default, fix, and warns. upstream_defects="reproduce" (on the command line --upstream-defects reproduce from the next release, LARMORX_UPSTREAM_DEFECTS=reproduce in 0.1.0rc1) gives ANTs’ behaviour, and the records compare with it. Every Atropos defect of the register but ANTS-107 is corrected by default: ITK-50, ITK-51 and ANTS-90 to ANTS-106. The record’s list of what fix changed meets fifteen of them (ANTS-90, ANTS-92, ANTS-94 to ANTS-96, ANTS-98, ANTS-100 to ANTS-106, ITK-50 and ITK-51). ITK-50 changes sMRIPrep’s own call, as the worked example measures; ANTS-106 changes partial-volume classes with an MRF, as the phantom above measures. ANTS-107, where ANTs weighs part of its samples with memory it never wrote, is not met by default and is filled deterministically when reproduced.

Random testing

Random testing runs random valid inputs and option combinations through the original and the replica, in long runs of several hours: inputs from seeded plain arithmetic (up to 32³ voxels; orientations, data types, scale factors and value ranges of every kind), with options drawn in random combinations from those the program accepts (how random testing works). A difference is shrunk to the smallest input and fewest options that still show it, then fixed and kept as a case of the record, registered as a defect of the original, or reported. The latest long run:

Random testing, the replica against the original: ANTs Atropos 2.6.5, the long run of 2026-10-25 (seed 20261025, 6.2 h for 21 programs): random valid inputs and option combinations, each run through both

ProgramRandom casesExtremeOptions coveredAgreeglibcNoticesUndefinedDifferHeavyGuardUnclear
Atroposreplica1,3630201,211·126719···
  • fixed in LARMORX Atropos (replica), 19 case(s): the fix pass, stderr: two of LARMORX’s messages: the one-line explanation of an empty mask or a constant channel in a Parzen histogram (ANTS-45: ANTs explains its exception only with --verbose 1) held ITK’s two-line description, now one line; and the ‘<= 1 element’ warning of a class with one sample was printed as the Gaussian model’s for every model, where ANTs prints each model’s own (ManifoldParzenWindows at line 55, HistogramParzenWindows at line 57: another file, class and text), now each model’s. The remaining fix-pass notices name defects that changed no value

Notices: a correction named in a case where it changed nothing the input shows (defect, cases): Atropos: ANTS-107 (25), ANTS-90 (19), ANTS-91 (22), ANTS-92 (25), ANTS-93 (11), ANTS-94 (27), ANTS-95 (7), ITK-50 (30), ITK-51 (4).

Every case is drawn from its tool, the run's seed and its number, and replays anywhere: how random testing works; each difference's triage: validation/difftest/triage.toml.

Known limits

The comparison with ANTs is made on Linux x86-64, against ANTs at one thread. ANTs’ clock-seeded runs have no single result to compare with: they are compared as one more draw. There is no benchmark report yet. The cross-platform checks (how they work) hold the other five platforms to LARMORX’s own Linux output.

History

  • 2026-10-22: ANTS-106 corrected by default, the partial-volume classes with an MRF (the user’s decision), measured on the phantom of known fractions.
  • 2026-10-23: the records of the standard and full comparisons shown here.
  • 2026-10-26: the data panel’s record, sMRIPrep’s segmentation in all fourteen runs and three more calls.

Data credits

The data this page shows or names: each dataset with its version, DOI and licence, from LARMORX's test-data catalog. Public-domain dedications (CC0, PDDL) ask nothing; the datasets are cited all the same.