3D segmentation, registration & resampling
Segment structures and register volumes to reference masks.
20 project-authored proposals · taxonomy v1.1 · 0 new benchmark runs. Verifiers and partial-credit rules below are designs, not executed results.
- Typical applications
- 3D Slicer · ITK-SNAP · MITK
- Class GUI-need label
G2· GUI-essential
Task proposals · 20
Expand to inspect the complete original wording and proposed checks. Original proposal text remains English in both interfaces.
R09.01.01.02-T01 Segment the liver with Grow from Seeds in 3D Slicer
Original instruction
In 3D Slicer load input/ct_abd_017.nii.gz (portal venous CT, synthetic-PHI public case). In Segment Editor create segments 'liver' and 'background', paint seeds on a few slices, run Grow from Seeds, apply, then smooth (median 3 mm) and fill holes. Export output/liver.nii.gz as a labelmap on the CT grid.
- Platforms
mac· macOSwin· Windows- Applications
- 3D Slicer
- Interfaces
desk· Desktop GUI- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G2· GUI-essential- Interaction primitives
path· Continuous pathvis· Visual judgmentnav· Pan/zoom/scroll- Work product
WP10· Annotations- Conditions
- None declared
- Proposed difficulty
- 3 / 5
Proposed verifier
Labelmap geometry identical to the CT; Dice>=0.90 and HD95<=8 mm vs gold liver mask. Slicer Python console and Jupyter kernel are disabled; scene history (.seg.nrrd effect metadata) must show Grow from Seeds use.
Proposed partial credit
10% export geometry; 15% seeds-based effect used; 50% Dice scaled to 0.90; 25% HD95
R09.01.01.02-T02 Repair leakage in a brain tumour edema label with ITK-SNAP paintbrush
Original instruction
In ITK-SNAP load input/flair_031.nii.gz with input/edema_auto.nii.gz. The automatic edema label leaks into 4 regions of normal tissue (ventricle cap, cortex). Use the paintbrush and polygon tools to remove the leakage and fill 2 under-segmented areas. Save output/edema_fixed.nii.gz.
- Platforms
mac· macOSwin· Windows- Applications
- ITK-SNAP
- Interfaces
desk· Desktop GUI- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G2· GUI-essential- Interaction primitives
path· Continuous pathvis· Visual judgmentnav· Pan/zoom/scroll- Work product
WP10· Annotations- Conditions
- None declared
- Proposed difficulty
- 3 / 5
Proposed verifier
Dice vs gold >=0.85 and improvement >=0.08 over the input; voxels outside the 6 edit regions changed <1%; label value preserved. ITK-SNAP has no scripting console and the VM has no shell, so itksnap-wt, c3d and Python NIfTI tools are unavailable; output writes by other processes are caught by file-access auditing.
Proposed partial credit
50% leakage removed (per region); 20% under-segmentation filled; 30% final Dice
R09.01.01.02-T03 Register MR to CT with six paired fiducials in 3D Slicer
Original instruction
In 3D Slicer load input/head_ct.nrrd and input/head_t1.nrrd. Place 6 corresponding anatomical fiducials on each (e.g. orbits, internal auditory canals, vertex, nasion), run Fiducial Registration Wizard (rigid), harden the transform and save output/mr_to_ct.h5 plus the resampled MR on the CT grid as output/t1_in_ct.nrrd.
- Platforms
mac· macOSwin· Windows- Applications
- 3D Slicer
- Interfaces
desk· Desktop GUI- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G2· GUI-essential- Interaction primitives
geom· Precise geometrynav· Pan/zoom/scrollvis· Visual judgment- Work product
WP04· 3D/spatial model- Conditions
- None declared
- Proposed difficulty
- 3 / 5
Proposed verifier
Checker applies the transform to 10 hidden landmark pairs: mean TRE<2 mm; resampled volume geometry equals CT grid; fiducial lists saved in scene with 6 points each. Slicer's Python Interactor is disabled and the launcher rejects --python-script/--python-code; the VM has no shell or SimpleITK, so registration must be done in Slicer's GUI modules.
Proposed partial credit
20% fiducials placed; 50% TRE<2 mm (scaled from 5 mm); 30% resampled output
R09.01.01.02-T04 Resample a label map to 1 mm isotropic with nearest neighbour
Original instruction
In 3D Slicer resample input/roi_labels_0.7x0.7x3.nrrd (5 labels) to 1 mm isotropic spacing on the same physical extent using nearest-neighbour interpolation so no new label values appear, and save output/roi_labels_1mm.nrrd.
- Platforms
mac· macOSwin· Windows- Applications
- 3D Slicer
- Interfaces
desk· Desktop GUI- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
CL· CLI-sufficient- Interaction primitives
click· Point & clicktype· Text & forms- Work product
WP10· Annotations- Conditions
- None declared
- Proposed difficulty
- 2 / 5
Proposed verifier
Header: spacing 1,1,1, origin and direction preserved, extent covering the original; label set equals {0..5}; per-label volume within 3% of original; voxelwise agreement with reference resampling >=99%. Slicer's Python Interactor is disabled and the launcher rejects --python-script/--python-code; the VM has no shell or SimpleITK, so resampling must be done in Slicer's GUI modules.
Proposed partial credit
30% spacing/origin/direction; 20% no new labels; 50% voxel agreement
R09.01.01.02-T05 Segment the lateral ventricles with ITK-SNAP active contours
Original instruction
In ITK-SNAP load input/t1_brain_09.nii.gz. Use Active Contour (snake) mode with thresholding pre-segmentation, place bubbles in both lateral ventricles, evolve until filled without leaking into the third ventricle, then save output/ventricles.nii.gz (label 1).
- Platforms
mac· macOSwin· Windows- Applications
- ITK-SNAP
- Interfaces
desk· Desktop GUI- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G2· GUI-essential- Interaction primitives
click· Point & clickvis· Visual judgmentdrag· Drag & drop- Work product
WP10· Annotations- Conditions
- None declared
- Proposed difficulty
- 3 / 5
Proposed verifier
Dice>=0.85 vs gold lateral ventricle mask; overlap with third/fourth ventricle gold masks <2% of segmented volume; ITK-SNAP workspace file must record snake mode parameters. ITK-SNAP has no scripting console and the VM has no shell, so itksnap-wt, c3d and Python NIfTI tools are unavailable; output writes by other processes are caught by file-access auditing.
Proposed partial credit
15% snake workspace saved; 60% Dice scaled to 0.85; 25% no leakage
R09.01.01.02-T06 Deformably register inhale to exhale lung CT in MITK
Original instruction
In MITK load the inhale and exhale CT phases of input/4dct_case03/. Using the MITK registration (MatchPoint) plugin, run a rigid then B-spline deformable registration of inhale to exhale with lung masks, check the result with the checkerboard view, and save the deformation field output/def.nrrd and warped image output/inhale_warped.nrrd.
- Platforms
win· Windowslin· Linux- Applications
- MITK
- Interfaces
desk· Desktop GUI- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G2· GUI-essential- Interaction primitives
click· Point & clickvis· Visual judgmentnav· Pan/zoom/scroll- Work product
WP04· 3D/spatial model- Conditions
- None declared
- Proposed difficulty
- 5 / 5
Proposed verifier
Checker applies the field to 300 hidden landmark pairs: mean TRE<3 mm (initial ~10 mm) and 95th percentile <7 mm; Jacobian determinant positive everywhere inside the lungs. MITK Workbench runs without its Python console plugin and the VM has no shell for MatchPoint/MITK command-line apps; output writes by other processes are caught by file-access auditing.
Proposed partial credit
20% rigid stage; 50% TRE scaled from 10 to 3 mm; 15% 95th percentile; 15% no folding
R09.01.01.02-T07 Segment bladder, rectum and prostate on pelvic MRI in 3D Slicer
Original instruction
In 3D Slicer load input/pelvis_t2_12.nrrd and segment three structures in Segment Editor: bladder (threshold + islands), rectum wall and lumen (paint/draw), prostate (draw on axial slices then fill between slices). Export output/pelvis_seg.nii.gz with labels 1/2/3.
- Platforms
mac· macOSwin· Windows- Applications
- 3D Slicer
- Interfaces
desk· Desktop GUI- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G2· GUI-essential- Interaction primitives
path· Continuous pathnav· Pan/zoom/scrollvis· Visual judgment- Work product
WP10· Annotations- Conditions
long· Long-horizon- Proposed difficulty
- 4 / 5
Proposed verifier
Per structure Dice vs gold: bladder>=0.90, rectum>=0.75, prostate>=0.80; labels exact; geometry equals input. Console disabled; segment history must show interactive effects.
Proposed partial credit
30% bladder; 30% rectum; 30% prostate; 10% label values and geometry
R09.01.01.02-T08 Crop a CT to a region of interest and reorient to RAS in 3D Slicer
Original instruction
In 3D Slicer, crop input/ct_thorax.nrrd to the ROI described in input/roi.txt (centre and size in RAS mm) with Crop Volume, keeping original spacing, and save it as output/thorax_crop.nii.gz with an identity direction matrix (RAS) without moving anatomy in physical space.
- Platforms
mac· macOSwin· Windows- Applications
- 3D Slicer
- Interfaces
desk· Desktop GUI- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
CL· CLI-sufficient- Interaction primitives
geom· Precise geometryclick· Point & click- Work product
WP04· 3D/spatial model- Conditions
- None declared
- Proposed difficulty
- 2 / 5
Proposed verifier
Header: spacing equal to input, physical bounds equal to the ROI within one voxel, direction identity in RAS; intensity at 20 random physical points equals the original (nearest voxel). Slicer's Python Interactor is disabled and the launcher rejects --python-script/--python-code; the VM has no shell or SimpleITK, so cropping must be done in Slicer's GUI modules.
Proposed partial credit
40% ROI bounds; 20% spacing; 20% orientation; 20% physical consistency
R09.01.01.02-T09 Segment the spleen by sparse slices and interpolation in ITK-SNAP
Original instruction
In ITK-SNAP load input/ct_spleen_22.nii.gz. Manually outline the spleen with the polygon tool on roughly every 5th axial slice, then use Interpolate Labels to fill the slices in between and touch up errors. Save output/spleen.nii.gz.
- Platforms
mac· macOSwin· Windows- Applications
- ITK-SNAP
- Interfaces
desk· Desktop GUI- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G2· GUI-essential- Interaction primitives
path· Continuous pathnav· Pan/zoom/scrollvis· Visual judgment- Work product
WP10· Annotations- Conditions
- None declared
- Proposed difficulty
- 3 / 5
Proposed verifier
Dice>=0.88 and HD95<=6 mm vs gold; label 1 only; geometry matches input. ITK-SNAP has no scripting console and the VM has no shell, so itksnap-wt, c3d and Python NIfTI tools are unavailable; output writes by other processes are caught by file-access auditing.
Proposed partial credit
60% Dice scaled to 0.88; 30% HD95; 10% geometry and label
R09.01.01.02-T10 Export a smoothed decimated femur surface model from 3D Slicer
Original instruction
In 3D Slicer segment the right femur in input/ct_hip_05.nrrd (threshold 200 HU + islands + fill holes), then export a surface model with smoothing factor 0.5 and 50% decimation to output/femur.stl in RAS coordinates (mm).
- Platforms
mac· macOSwin· Windows- Applications
- 3D Slicer
- Interfaces
desk· Desktop GUI- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
CL· CLI-sufficient- Interaction primitives
click· Point & clickvis· Visual judgmentpath· Continuous path- Work product
WP04· 3D/spatial model- Conditions
- None declared
- Proposed difficulty
- 3 / 5
Proposed verifier
STL is watertight and manifold; triangle count within 40-60% of the undecimated export; symmetric Hausdorff (95%) to gold femur surface <1.5 mm; only the right femur present (no pelvis or left femur). Slicer's Python Interactor is disabled and the launcher rejects --python-script/--python-code; the VM has no shell, VTK or SimpleITK, so the model must be made in Slicer's GUI modules.
Proposed partial credit
30% correct structure isolated; 30% surface distance; 20% watertight; 20% decimation level
R09.01.01.02-T11 Segment lung nodules with region growing in MITK and report volumes
Original instruction
In MITK load input/ldct_44.nrrd. Segment the 3 nodules listed in input/nodules.csv (approximate RAS centres) using the region growing tool and manual correction, save them as one multilabel output/nodules.nrrd (labels 1-3), and write volumes in mm^3 to output/volumes.csv.
- Platforms
win· Windowslin· Linux- Applications
- MITK
- Interfaces
desk· Desktop GUI- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G2· GUI-essential- Interaction primitives
click· Point & clickpath· Continuous pathvis· Visual judgment- Work product
WP10· Annotations- Conditions
- None declared
- Proposed difficulty
- 3 / 5
Proposed verifier
Per nodule Dice>=0.75 vs gold; volumes within 10% of gold and equal to the saved mask volumes within 1%. MITK Workbench runs without its Python console plugin and the VM has no shell for MITK command-line apps; output writes by other processes are caught by file-access auditing.
Proposed partial credit
60% per-nodule Dice (equal share each); 30% volumes; 10% consistent with masks
R09.01.01.02-T12 Propagate atlas labels to a subject by landmark registration in 3D Slicer
Original instruction
In 3D Slicer register atlas input/atlas_t1.nrrd to subject input/subj_t1.nrrd using 8 landmark pairs (Landmark Registration module, affine plus thin-plate spline), then apply the transform to input/atlas_labels.nrrd with nearest-neighbour resampling onto the subject grid. Save output/subj_labels.nrrd.
- Platforms
mac· macOSwin· Windows- Applications
- 3D Slicer
- Interfaces
desk· Desktop GUI- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G2· GUI-essential- Interaction primitives
geom· Precise geometrydrag· Drag & dropvis· Visual judgment- Work product
WP10· Annotations- Conditions
- None declared
- Proposed difficulty
- 4 / 5
Proposed verifier
Mean Dice across 6 structures vs subject gold labels >=0.75; output on the subject grid; label set unchanged; landmark TRE on hidden pairs <3 mm. Slicer's Python Interactor is disabled and the launcher rejects --python-script/--python-code; the VM has no shell or SimpleITK, so registration must be done in Slicer's GUI modules.
Proposed partial credit
25% landmark set; 20% TRE; 45% mean Dice; 10% grid and labels
R09.01.01.02-T13 Affinely register a T1 scan to MNI space with BRAINSFit in 3D Slicer
Original instruction
In 3D Slicer register input/t1_sub07.nii.gz to the MNI152 1 mm template (input/mni152.nii.gz) using General Registration (BRAINS) with rigid+affine stages, mutual information, and a brain mask initialization by centre of mass. Harden and resample to template space; save output/t1_mni.nii.gz and output/affine.tfm.
- Platforms
mac· macOSwin· Windowslin· Linux- Applications
- 3D Slicer
- Interfaces
desk· Desktop GUI- GUI need · task setting
HY· Hybrid- GUI need · ordinary access
CL· CLI-sufficient- Interaction primitives
click· Point & clicktype· Text & formsvis· Visual judgment- Work product
WP04· 3D/spatial model- Conditions
- None declared
- Proposed difficulty
- 3 / 5
Proposed verifier
Output on MNI grid; NCC with template within the brain mask >=0.85; checker applies affine.tfm to 12 hidden landmarks: mean error <3 mm.
Proposed partial credit
20% output grid; 40% NCC; 40% landmark error
R09.01.01.02-T14 Split a merged hippocampus label into left and right in ITK-SNAP
Original instruction
input/hippo_merged.nii.gz labels both hippocampi as 1 on input/t1_hippo.nii.gz. In ITK-SNAP relabel the left hippocampus to 2 and the right to 3 (radiological orientation shown on screen), keeping every voxel. Save output/hippo_lr.nii.gz.
- Platforms
mac· macOSwin· Windows- Applications
- ITK-SNAP
- Interfaces
desk· Desktop GUI- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
CL· CLI-sufficient- Interaction primitives
click· Point & clickpath· Continuous pathvis· Visual judgment- Work product
WP10· Annotations- Conditions
- None declared
- Proposed difficulty
- 2 / 5
Proposed verifier
Union of labels 2 and 3 equals the input label 1 voxel set exactly; left/right assignment correct in patient coordinates (checker uses the header direction); no voxels remain with label 1. ITK-SNAP has no scripting console and the VM has no shell, so itksnap-wt, c3d and Python NIfTI tools are unavailable; output writes by other processes are caught by file-access auditing.
Proposed partial credit
50% correct sides; 30% voxels preserved; 20% no label 1 left
R09.01.01.02-T15 Extract a vessel centreline and measure its length in 3D Slicer
Original instruction
In 3D Slicer with SlicerVMTK, segment the left internal carotid artery from input/cta_neck_02.nrrd (vesselness-guided threshold plus manual cleanup), extract its centreline from the bifurcation to the carotid siphon (endpoints in input/endpoints.fcsv), and save output/ica_seg.nrrd and output/centreline.json with length in mm.
- Platforms
mac· macOSwin· Windows- Applications
- 3D Slicer · SlicerVMTK
- Interfaces
desk· Desktop GUI- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G2· GUI-essential- Interaction primitives
path· Continuous pathvis· Visual judgmentnav· Pan/zoom/scroll- Work product
WP04· 3D/spatial model- Conditions
tutorial· Follow docs/tutorial- Proposed difficulty
- 5 / 5
Proposed verifier
Vessel segment Dice>=0.80 vs gold; centreline endpoints within 3 mm of the given fiducials and inside the gold lumen along 95% of its length; length within 5% of gold centreline. Slicer's Python Interactor is disabled and the launcher rejects --python-script/--python-code; the VM has no shell or VMTK scripts, so the work must be done in Slicer's GUI modules.
Proposed partial credit
40% vessel segmentation; 20% endpoints; 20% centreline inside lumen; 20% length
R09.01.01.02-T16 Detect and fix a left-right flipped CT orientation in 3D Slicer
Original instruction
input/ct_chest_flip.nrrd has an incorrect direction matrix that mirrors left and right (the heart appears on the patient's right). Confirm from anatomy in 3D Slicer, correct the orientation with a transform so anatomy is in the correct physical positions, harden it and save output/ct_chest_fixed.nrrd. Do not alter voxel values.
- Platforms
mac· macOSwin· Windows- Applications
- 3D Slicer
- Interfaces
desk· Desktop GUI- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G2· GUI-essential- Interaction primitives
vis· Visual judgmentgeom· Precise geometryclick· Point & click- Work product
WP04· 3D/spatial model- Conditions
hidden· Hidden/stale state- Proposed difficulty
- 4 / 5
Proposed verifier
Checker compares the corrected image to the hidden correctly oriented original: physical-space voxel agreement 100% (nearest neighbour) and heart centroid on the patient's left; voxel value histogram identical. Slicer's Python Interactor is disabled and the launcher rejects --python-script/--python-code; the VM has no shell or SimpleITK, so the fix must be made in Slicer's GUI modules.
Proposed partial credit
30% flip correctly diagnosed (direction sign); 50% physical agreement; 20% voxel values untouched
R09.01.01.02-T17 Build a tumour time-intensity curve from 4D DCE-MRI in MITK
Original instruction
In MITK load the 4D DCE-MRI series input/dce_breast_06/ (40 timepoints). Segment the enhancing lesion on the peak-enhancement frame, then compute mean intensity inside the mask over time and export output/tic.csv (time_s, mean) plus output/lesion.nrrd.
- Platforms
win· Windowslin· Linux- Applications
- MITK
- Interfaces
desk· Desktop GUI- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G2· GUI-essential- Interaction primitives
path· Continuous pathnav· Pan/zoom/scrollvis· Visual judgment- Work product
WP10· Annotations- Conditions
- None declared
- Proposed difficulty
- 4 / 5
Proposed verifier
Lesion Dice>=0.75 vs gold; tic.csv has 40 rows with times from the DICOM headers; curve values within 5% of the curve computed from the gold mask; curve type (washout) identifiable. MITK Workbench runs without its Python console plugin and the VM has no shell for MITK command-line apps; output writes by other processes are caught by file-access auditing.
Proposed partial credit
40% lesion segmentation; 20% timepoints; 40% curve values
R09.01.01.02-T18 Control pair of T04: resample the label map with SimpleITK
Original instruction
Control pair of T04. Write output/resample.py using SimpleITK that resamples input/roi_labels_0.7x0.7x3.nrrd to 1 mm isotropic on the same physical extent with nearest-neighbour interpolation and writes output/roi_labels_1mm.nrrd.
- Platforms
lin· Linuxmac· macOSwin· Windows- Applications
- Python · SimpleITK
- Interfaces
term· Terminal- GUI need · task setting
CL· CLI-sufficient- GUI need · ordinary access
CL· CLI-sufficient- Interaction primitives
code· Code & commands- Work product
WP10· Annotations- Conditions
- None declared
- Proposed difficulty
- 3 / 5
Proposed verifier
Same checks as T04 (spacing, origin, direction, label set, per-label volume within 3%, voxel agreement >=99%) for a direct GUI-vs-script comparison.
Proposed partial credit
30% header geometry; 20% label set; 50% voxel agreement
R09.01.01.02-T19 Manually segment the pancreas on contrast CT in 3D Slicer
Original instruction
In 3D Slicer segment the pancreas (head, uncinate, body, tail) on input/ct_panc_33.nrrd using Draw/Paint/Erase with intensity masking and Fill between slices, excluding the splenic vein and duodenum. Export output/pancreas.nii.gz.
- Platforms
mac· macOSwin· Windows- Applications
- 3D Slicer
- Interfaces
desk· Desktop GUI- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G2· GUI-essential- Interaction primitives
path· Continuous pathnav· Pan/zoom/scrollvis· Visual judgment- Work product
WP10· Annotations- Conditions
long· Long-horizon- Proposed difficulty
- 5 / 5
Proposed verifier
Dice>=0.75 and HD95<=10 mm vs gold; overlap with gold splenic vein and duodenum masks <3% of segmentation volume. Console disabled; effect history checked.
Proposed partial credit
55% Dice scaled to 0.75; 25% HD95; 20% excluded structures
R09.01.01.02-T20 Manually align two MR volumes with ITK-SNAP's interactive registration
Original instruction
In ITK-SNAP load input/t2_axial.nii.gz as main and input/t2_coronal_recon.nii.gz as additional image (misaligned by about 8 mm and 6 degrees). Use the Registration panel's manual mode, dragging the rotation and translation widgets while checking the overlay, then run automatic rigid refinement and save output/manual.txt (ITK transform).
- Platforms
mac· macOSwin· Windows- Applications
- ITK-SNAP
- Interfaces
desk· Desktop GUI- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G2· GUI-essential- Interaction primitives
drag· Drag & dropgeom· Precise geometryvis· Visual judgment- Work product
WP04· 3D/spatial model- Conditions
- None declared
- Proposed difficulty
- 4 / 5
Proposed verifier
Transform error vs gold: translation <2 mm and rotation <2 degrees; ITK-SNAP workspace records a manual transform before the automatic step (initial-pose log). ITK-SNAP has no scripting console and the VM has no shell, so itksnap-wt, c3d and Python NIfTI tools are unavailable; output writes by other processes are caught by file-access auditing.
Proposed partial credit
20% manual alignment recorded; 40% translation error; 40% rotation error
Mapped research references
Existing benchmark-to-class links from the research catalog. They are not record-level annotations, executions or product endorsements.