Computer-Use Agent Taxonomy中文
R09.02.01.01 · TASK CLASS

Radiotherapy planning & dose evaluation

Contour, plan and evaluate dose against constraints.

20 project-authored proposals · taxonomy v1.1 · 0 new benchmark runs. Verifiers and partial-credit rules below are designs, not executed results.

Typical applications
matRad · SlicerRT · Eclipse · RayStation
Class GUI-need label
HY · Hybrid

Task proposals · 20

Expand to inspect the complete original wording and proposed checks. Original proposal text remains English in both interfaces.

R09.02.01.01-T01 Compute DVH metrics for a prostate plan in SlicerRT

Original instruction

In 3D Slicer with SlicerRT, load the DICOM-RT study input/prostate_plan/ (CT, RTSTRUCT, RTDOSE; synthetic). Compute DVHs for PTV, bladder and rectum and write output/dvh.json with PTV D95%, bladder V65Gy and rectum V70Gy (percent volume), plus the DVH table export output/dvh.csv.

Platforms
mac · macOS win · Windows
Applications
3D Slicer · SlicerRT
Interfaces
desk · Desktop GUI
GUI need · task setting
HY · Hybrid
GUI need · ordinary access
CL · CLI-sufficient
Interaction primitives
click · Point & click nav · Pan/zoom/scroll
Work product
WP07 · Simulation result
Conditions
None declared
Proposed difficulty
2 / 5

Proposed verifier

Checker recomputes with dicompyler-core on the same files: D95 within 0.5 Gy, V65/V70 within 1 percentage point; dvh.csv contains 3 structures with cumulative curves.

Proposed partial credit

30% PTV D95; 30% bladder V65; 30% rectum V70; 10% DVH export

R09.02.01.01-T02 Create an anisotropic PTV margin in 3D Slicer Segment Editor

Original instruction

In 3D Slicer load input/hn_case/ (CT + RTSTRUCT with CTV_70). Create PTV_70 by expanding CTV_70 10 mm superior/inferior and 5 mm in all other directions using the Margin/Hollow tools or Segment Editor with anisotropic margins, crop 3 mm inside the body contour, and export a new RTSTRUCT output/rtstruct_ptv.dcm.

Platforms
mac · macOS win · Windows
Applications
3D Slicer · SlicerRT
Interfaces
desk · Desktop GUI
GUI need · task setting
HY · Hybrid
GUI need · ordinary access
CL · CLI-sufficient
Interaction primitives
click · Point & click type · Text & forms vis · Visual judgment
Work product
WP10 · Annotations
Conditions
None declared
Proposed difficulty
2 / 5

Proposed verifier

PTV_70 in the exported RTSTRUCT: Dice>=0.95 vs gold expansion; minimum distance to body surface >=3 mm; original structures unchanged.

Proposed partial credit

60% PTV geometry; 25% body crop; 15% export valid

R09.02.01.01-T03 Build a 7-field prostate IMRT plan in matRad and repair the structure set

Original instruction

In matRad (MATLAB/Octave) load input/prostate_case.mat. Fix the 3 planted structure problems listed in input/qa_notes.txt (wrong rectum label, missing PTV priority, couch included in body), then plan 7 equispaced 6 MV IMRT fields for 78 Gy in 39 fractions meeting input/constraints.csv, and export DICOM RTPLAN/RTDOSE to output/.

Platforms
mac · macOS win · Windows lin · Linux
Applications
matRad · MATLAB
Interfaces
desk · Desktop GUI term · Terminal
GUI need · task setting
HY · Hybrid
GUI need · ordinary access
CL · CLI-sufficient
Interaction primitives
code · Code & commands click · Point & click vis · Visual judgment
Work product
WP07 · Simulation result
Conditions
long · Long-horizon
Proposed difficulty
4 / 5

Proposed verifier

Checker recomputes DVH from exported dose: PTV D95>=74.1 Gy and Dmax<=83.5 Gy, rectum V70<=20%, bladder V75<=15%; 3 structure fixes verified; 7 beams at equispaced gantry angles in RTPLAN.

Proposed partial credit

20% structure fixes; 15% beam geometry; 35% PTV coverage; 30% OAR constraints

R09.02.01.01-T04 Compare photon IMRT and proton IMPT plans for a pediatric case in matRad

Original instruction

For synthetic pediatric medulloblastoma boost case input/peds_pf.mat, create in matRad (a) a 9-field photon IMRT plan and (b) a 3-field IMPT plan, both normalized to PTV D50 = 54 Gy(RBE) with RBE 1.1. Write output/compare.json with integral dose (J), cochlea mean doses and brainstem D1cc for both.

Platforms
mac · macOS win · Windows lin · Linux
Applications
matRad · MATLAB
Interfaces
desk · Desktop GUI term · Terminal
GUI need · task setting
HY · Hybrid
GUI need · ordinary access
CL · CLI-sufficient
Interaction primitives
code · Code & commands vis · Visual judgment
Work product
WP07 · Simulation result
Conditions
None declared
Proposed difficulty
4 / 5

Proposed verifier

Checker recomputes metrics from saved dose cubes: normalization within 0.2 Gy; reported metrics within 2% of recomputation; IMPT integral dose lower than IMRT and PTV D95>=95% in both.

Proposed partial credit

20% both plans optimized; 20% normalization; 40% metrics consistent; 20% coverage

R09.02.01.01-T05 Contour head-and-neck organs at risk in Eclipse training mode

Original instruction

In the Eclipse (non-clinical training database) Contouring workspace, open synthetic patient HN-TRAIN-07 and contour spinal cord, brainstem, both parotids and mandible on the planning CT using the brush and interpolation tools, naming them per input/naming_tg263.txt. Save the structure set 'OAR_v1'.

Platforms
win · Windows
Applications
Eclipse
Interfaces
desk · Desktop GUI
GUI need · task setting
G2 · GUI-essential
GUI need · ordinary access
G2 · GUI-essential
Interaction primitives
path · Continuous path nav · Pan/zoom/scroll vis · Visual judgment
Work product
WP10 · Annotations
Conditions
licensed · Licensed pro app long · Long-horizon
Proposed difficulty
4 / 5

Proposed verifier

Checker exports the structure set via DICOM from the training DB: per-structure Dice vs gold (cord>=0.8, brainstem>=0.8, parotids>=0.75, mandible>=0.85); names exactly TG-263; ESAPI scripting disabled for the agent account.

Proposed partial credit

15% names; 85% Dice per structure (equal share each)

R09.02.01.01-T06 Optimize a VMAT plan to clinical goals in RayStation research mode

Original instruction

In RayStation (research license, synthetic lung case LUNG-R-02), create a 2-arc 6 MV VMAT plan for 60 Gy in 30 fractions, set up optimization objectives, and iterate until as many goals in input/clinical_goals.xlsx as possible are met. Approve is not allowed; save the plan and export RTPLAN/RTDOSE.

Platforms
win · Windows
Applications
RayStation
Interfaces
desk · Desktop GUI
GUI need · task setting
G2 · GUI-essential
GUI need · ordinary access
CL · CLI-sufficient
Interaction primitives
click · Point & click type · Text & forms vis · Visual judgment
Work product
WP07 · Simulation result
Conditions
licensed · Licensed pro app long · Long-horizon
Proposed difficulty
5 / 5

Proposed verifier

Checker evaluates the 14 clinical goals on exported dose with its own DVH engine: score = fraction met (hard goals weighted 2x); plan state must be unapproved. RayStation scripting (Python console and Script Management) is disabled for the research user, the VM has no shell, and any script run in the RayStation audit log fails.

Proposed partial credit

15% plan created with 2 arcs; 60% clinical goals met (weighted); 15% PTV homogeneity; 10% unapproved and exported

R09.02.01.01-T07 Accumulate dose from two courses with deformable registration in SlicerRT

Original instruction

In 3D Slicer with SlicerRT, load course 1 and course 2 CTs with their RTDOSEs (input/reirr_case/). Deformably register course 1 CT to course 2 CT, warp the course 1 dose, add it to course 2 dose, and report maximum spinal cord dose (D0.03cc) of the sum in output/sum.json; save the summed dose as output/sum_dose.nrrd.

Platforms
mac · macOS win · Windows
Applications
3D Slicer · SlicerRT
Interfaces
desk · Desktop GUI
GUI need · task setting
HY · Hybrid
GUI need · ordinary access
CL · CLI-sufficient
Interaction primitives
click · Point & click vis · Visual judgment nav · Pan/zoom/scroll
Work product
WP07 · Simulation result
Conditions
None declared
Proposed difficulty
3 / 5

Proposed verifier

Summed dose on course 2 grid; cord D0.03cc within 3% of gold accumulation; registration TRE on 8 hidden landmarks <3 mm.

Proposed partial credit

35% registration quality; 25% dose warped and summed on correct grid; 40% cord metric

R09.02.01.01-T08 Re-optimize a prostate plan from 5 to 9 fields and report conformity

Original instruction

Starting from input/prostate_5f.mat (matRad), change the beam arrangement to 9 equispaced fields, re-optimize with the same objectives, and report the Paddick conformity index and gradient index for the 95% isodose in output/ci.json; save the new plan.

Platforms
mac · macOS win · Windows lin · Linux
Applications
matRad · MATLAB
Interfaces
term · Terminal desk · Desktop GUI
GUI need · task setting
HY · Hybrid
GUI need · ordinary access
CL · CLI-sufficient
Interaction primitives
code · Code & commands vis · Visual judgment
Work product
WP07 · Simulation result
Conditions
None declared
Proposed difficulty
3 / 5

Proposed verifier

Saved plan has 9 beams at 0,40,...,320; recomputed Paddick CI>=0.80; reported CI and GI within 2% of checker recomputation.

Proposed partial credit

25% beam setup; 40% CI threshold; 35% reported metrics correct

R09.02.01.01-T09 Run a 3%/3 mm gamma analysis between planned and measured dose

Original instruction

In 3D Slicer with SlicerRT, compare planned dose input/qa/plan_dose.dcm with measured dose input/qa/measured.dcm (phantom QA) using Dose Comparison with 3%/3 mm global gamma and 10% low-dose threshold. Report pass rate in output/gamma.json and save the gamma volume output/gamma.nrrd.

Platforms
mac · macOS win · Windows
Applications
3D Slicer · SlicerRT
Interfaces
desk · Desktop GUI
GUI need · task setting
HY · Hybrid
GUI need · ordinary access
CL · CLI-sufficient
Interaction primitives
click · Point & click type · Text & forms
Work product
WP07 · Simulation result
Conditions
None declared
Proposed difficulty
3 / 5

Proposed verifier

Pass rate within 0.5 percentage points of reference computed with pymedphys using the same criteria; gamma volume dimensions match the dose grid.

Proposed partial credit

20% correct criteria; 60% pass rate; 20% gamma volume saved

R09.02.01.01-T10 Evaluate a lung plan against protocol constraints in Eclipse

Original instruction

In Eclipse training mode open synthetic plan LUNG-EVAL-03. Review DVHs in Plan Evaluation and fill output/eval.csv per input/protocol_constraints.csv: for each of 12 constraints the achieved value, pass/minor deviation/major deviation, and an overall plan verdict.

Platforms
win · Windows
Applications
Eclipse
Interfaces
desk · Desktop GUI
GUI need · task setting
G2 · GUI-essential
GUI need · ordinary access
AP · API-sufficient
Interaction primitives
click · Point & click nav · Pan/zoom/scroll vis · Visual judgment
Work product
WP12 · Answer/decision
Conditions
licensed · Licensed pro app
Proposed difficulty
3 / 5

Proposed verifier

Achieved values within 0.5 Gy or 1% of checker's DVH from exported dose; classifications exact; verdict exact (2 planted failures: lung V20 major, esophagus mean minor). ESAPI script execution (Tools > Scripts) is disabled for the training-mode user and the VM has no shell, so values must be read from Eclipse's Plan Evaluation and DVH views.

Proposed partial credit

40% achieved values; 40% classifications; 20% verdict

R09.02.01.01-T11 Robustness-evaluate a proton plan under setup and range errors in matRad

Original instruction

For proton plan input/hn_impt.mat in matRad, compute dose for 9 scenarios (nominal, +-3 mm setup shifts on 3 axes, +-3.5% range) and report worst-case CTV D95 and the voxelwise worst-case cord Dmax in output/robust.json. Save scenario dose cubes.

Platforms
mac · macOS win · Windows lin · Linux
Applications
matRad · MATLAB
Interfaces
term · Terminal desk · Desktop GUI
GUI need · task setting
HY · Hybrid
GUI need · ordinary access
CL · CLI-sufficient
Interaction primitives
code · Code & commands
Work product
WP07 · Simulation result
Conditions
None declared
Proposed difficulty
5 / 5

Proposed verifier

Checker verifies 9 scenarios with correct shifts and range scalings; worst-case CTV D95 within 1% and cord Dmax within 2% of reference computation.

Proposed partial credit

40% scenario setup; 30% worst-case CTV D95; 30% worst-case cord

R09.02.01.01-T12 Display isodose lines and capture a chart screenshot in SlicerRT

Original instruction

In 3D Slicer with SlicerRT load input/breast_plan/. Show 95%, 100% and 107% isodose lines of the 50 Gy prescription on the axial slice through the isocenter with the standard colors from input/isodose_colors.txt, hide other structures except PTV, and save the scene and a screenshot output/isodose.png.

Platforms
mac · macOS win · Windows
Applications
3D Slicer · SlicerRT
Interfaces
desk · Desktop GUI
GUI need · task setting
G2 · GUI-essential
GUI need · ordinary access
CL · CLI-sufficient
Interaction primitives
click · Point & click vis · Visual judgment
Work product
WP03 · Visual layout
Conditions
None declared
Proposed difficulty
2 / 5

Proposed verifier

Saved .mrml: isodose levels exactly 47.5/50/53.5 Gy with specified colors; axial slice offset equals isocenter z (+-1 slice); screenshot pixel colors along isodose contours match specified RGB. Slicer's Python Interactor is disabled and the launcher rejects --python-script/--python-code; the VM has no shell, so display settings must be made in Slicer's GUI modules.

Proposed partial credit

40% isodose levels; 20% colors; 25% slice position; 15% visibility settings

R09.02.01.01-T13 Find and fix four planted RTSTRUCT defects in 3D Slicer

Original instruction

input/rtstruct_qa/ holds a CT and RTSTRUCT with 4 planted defects: a hole in the bladder contour, a structure named 'Rect' instead of 'Rectum', a femoral head contour on one stray slice 4 cm away, and a self-intersecting PTV contour. Fix them in 3D Slicer and export output/rtstruct_fixed.dcm plus output/qa_log.json listing each defect.

Platforms
mac · macOS win · Windows
Applications
3D Slicer · SlicerRT
Interfaces
desk · Desktop GUI
GUI need · task setting
HY · Hybrid
GUI need · ordinary access
G2 · GUI-essential
Interaction primitives
path · Continuous path vis · Visual judgment click · Point & click
Work product
WP10 · Annotations
Conditions
None declared
Proposed difficulty
3 / 5

Proposed verifier

Exported structures: bladder has no holes and Dice>=0.98 vs gold; name exact; stray slice removed; PTV contours valid polygons; other structures unchanged (Dice>=0.99); qa_log names all 4 defects.

Proposed partial credit

80% four defects fixed (equal share each); 10% other structures unchanged; 10% QA log

R09.02.01.01-T14 Quantify dose-grid resolution sensitivity in matRad

Original instruction

Recompute the dose of input/brain_plan.mat in matRad at dose grids of 1, 2.5 and 5 mm with identical fluence, and report PTV D2%, D98% and brainstem D0.03cc for each grid in output/grid.json together with the percent difference relative to 1 mm.

Platforms
mac · macOS win · Windows lin · Linux
Applications
matRad · MATLAB
Interfaces
term · Terminal
GUI need · task setting
CL · CLI-sufficient
GUI need · ordinary access
CL · CLI-sufficient
Interaction primitives
code · Code & commands
Work product
WP07 · Simulation result
Conditions
None declared
Proposed difficulty
3 / 5

Proposed verifier

Checker reruns the script: 3 dose cubes with the stated resolutions; metrics within 1% of reference; fluence identical across runs (weights hash).

Proposed partial credit

30% three recomputations; 50% metrics; 20% percent differences

R09.02.01.01-T15 Set up isocenter and VMAT arc geometry in RayStation

Original instruction

In RayStation research mode (synthetic prostate case PRO-R-11) create a plan with isocenter at the PTV centre of mass, add 2 full VMAT arcs (181-179 CW and 179-181 CCW) with collimator 30 and 330 degrees, couch 0, 6 MV, 2 degree control-point spacing. Save and export RTPLAN to output/.

Platforms
win · Windows
Applications
RayStation
Interfaces
desk · Desktop GUI
GUI need · task setting
G2 · GUI-essential
GUI need · ordinary access
CL · CLI-sufficient
Interaction primitives
click · Point & click type · Text & forms
Work product
WP08 · App/system state
Conditions
licensed · Licensed pro app
Proposed difficulty
3 / 5

Proposed verifier

Parsed RTPLAN: isocenter within 1 mm of PTV centroid; 2 arcs with exact start/stop/direction, collimator angles, couch 0, energy and control-point spacing. RayStation scripting (Python console and Script Management) is disabled for the research user and the VM has no shell, so geometry must be set in the RayStation GUI.

Proposed partial credit

30% isocenter; 40% arc geometry; 30% beam parameters

R09.02.01.01-T16 Compute lung NTCP from the DVH using the LKB model

Original instruction

Load input/lung_plan/ in 3D Slicer with SlicerRT, export the total-lung-minus-GTV differential DVH, convert each bin to EQD2 (alpha/beta 3; the plan delivers 15 fractions), then compute LKB NTCP for radiation pneumonitis with TD50=30.8 Gy, m=0.37, n=0.99 and report it in output/ntcp.json along with mean lung dose.

Platforms
mac · macOS win · Windows
Applications
3D Slicer · SlicerRT · Python
Interfaces
desk · Desktop GUI term · Terminal
GUI need · task setting
HY · Hybrid
GUI need · ordinary access
CL · CLI-sufficient
Interaction primitives
click · Point & click code · Code & commands
Work product
WP07 · Simulation result
Conditions
multiapp · Multi-app
Proposed difficulty
3 / 5

Proposed verifier

Checker recomputes from RTDOSE/RTSTRUCT: EQD2 mean lung dose within 0.3 Gy and NTCP within 1 percentage point; the exported DVH must come from SlicerRT (format header).

Proposed partial credit

25% DVH export; 25% EQD2 conversion; 20% mean lung dose; 30% NTCP

R09.02.01.01-T17 Plan lung SBRT meeting RTOG 0915 conformity and gradient limits

Original instruction

In matRad plan synthetic peripheral lung SBRT case input/sbrt_lung.mat: 48 Gy in 4 fractions, 9-11 non-coplanar or coplanar fields of your choice, prescribed so 95% of PTV receives 100%. Meet the RTOG 0915 ratio limits (R100%, R50%, D2cm) and OAR limits in input/rtog0915.csv. Save plan and output/sbrt_metrics.json.

Platforms
mac · macOS win · Windows lin · Linux
Applications
matRad · MATLAB
Interfaces
desk · Desktop GUI term · Terminal
GUI need · task setting
HY · Hybrid
GUI need · ordinary access
CL · CLI-sufficient
Interaction primitives
code · Code & commands vis · Visual judgment
Work product
WP07 · Simulation result
Conditions
None declared
Proposed difficulty
5 / 5

Proposed verifier

Checker recomputes from the dose cube: PTV V100=95%+-1%; R100<1.2, R50 and D2cm within table limits for the PTV volume; chest wall, cord and bronchus limits met; reported metrics within 2%.

Proposed partial credit

20% prescription normalization; 40% conformity/gradient metrics; 30% OAR limits; 10% reporting

R09.02.01.01-T18 Adaptive replan after anatomical change on a repeat CT

Original instruction

A head-and-neck patient (synthetic) lost weight; input/adapt/ has the planning CT with structures, plan, and a repeat CT. In 3D Slicer propagate contours to the repeat CT with deformable registration and review/edit them, recompute the original plan's dose there, then re-optimize in matRad and report original-vs-adapted metrics in output/adapt.json.

Platforms
mac · macOS win · Windows lin · Linux
Applications
3D Slicer · SlicerRT · matRad
Interfaces
desk · Desktop GUI term · Terminal
GUI need · task setting
HY · Hybrid
GUI need · ordinary access
HY · Hybrid
Interaction primitives
path · Continuous path code · Code & commands multi · Cross-app
Work product
WP07 · Simulation result
Conditions
multiapp · Multi-app long · Long-horizon
Proposed difficulty
5 / 5

Proposed verifier

Propagated contours Dice>=0.8 vs gold repeat-CT contours; recomputed original plan shows the planted parotid overdose; adapted plan restores PTV D95>=95% and parotid mean <=26 Gy; JSON metrics within 2% of recomputation.

Proposed partial credit

30% contour propagation; 20% original-plan recalculation; 35% adapted plan meets goals; 15% report

R09.02.01.01-T19 Compute cumulative EQD2 for a re-irradiated spinal cord

Original instruction

For input/reirr_spine/ (course 1: 30 Gy/10 fx; course 2: 20 Gy/5 fx; doses registered on one grid), convert each course dose voxelwise to EQD2 with alpha/beta 2 for cord, sum them, and report the cumulative cord EQD2 Dmax and D0.1cc in output/eqd2.json; save output/eqd2_sum.nrrd.

Platforms
mac · macOS win · Windows lin · Linux
Applications
Python · 3D Slicer
Interfaces
term · Terminal desk · Desktop GUI
GUI need · task setting
HY · Hybrid
GUI need · ordinary access
CL · CLI-sufficient
Interaction primitives
code · Code & commands
Work product
WP07 · Simulation result
Conditions
None declared
Proposed difficulty
4 / 5

Proposed verifier

Checker recomputes voxelwise EQD2 using fraction counts: Dmax and D0.1cc within 2%; saved volume matches within 0.5 Gy at 1000 random cord voxels.

Proposed partial credit

30% voxelwise conversion; 20% summation; 50% reported metrics

R09.02.01.01-T20 Export a matRad plan to DICOM-RT and verify DVHs in SlicerRT

Original instruction

Export the optimized plan in input/cervix_plan.mat from matRad as DICOM RTSTRUCT, RTPLAN and RTDOSE (with the CT), import them in 3D Slicer/SlicerRT, compute DVHs there, and write output/crosscheck.json listing D95 and Dmean per structure from both tools and their differences.

Platforms
mac · macOS win · Windows lin · Linux
Applications
matRad · 3D Slicer · SlicerRT
Interfaces
desk · Desktop GUI term · Terminal
GUI need · task setting
HY · Hybrid
GUI need · ordinary access
CL · CLI-sufficient
Interaction primitives
code · Code & commands click · Point & click multi · Cross-app
Work product
WP07 · Simulation result
Conditions
multiapp · Multi-app
Proposed difficulty
3 / 5

Proposed verifier

DICOM files validate (dciodvfy-clean for required tags) and import in Slicer; SlicerRT metrics within 1% of matRad for all structures; reported values match checker recomputation.

Proposed partial credit

30% valid DICOM export; 20% import; 30% metrics agree; 20% report

Mapped research references

Existing benchmark-to-class links from the research catalog. They are not record-level annotations, executions or product endorsements.