Anyone can contribute a case.
Everyone plans it. The scoreboard decides who teaches.
Upload a de-identified CT and structure set and it becomes a challenge. Plan it in your own TPS, submit the dose, and get a DVH-based score in minutes — the same criteria for everyone, regardless of vendor.
- 6
- Published cases
- 0
- Scored plans
- 4
- Planners
A published case in 3D — drag to rotate
Building the 3D view…
Open Radiotherapy Case Arena
No vendor owns the leaderboard.
No sponsorship, no advertising, no analytics. Scoring criteria are versioned in public and every change carries a written reason. Each planning system is also ranked within itself, because comparing two machines is not comparing two planners.
The name
OrcaDose — Open Radiotherapy Case Arena
- OpenNo vendor owns the leaderboard.
- No sponsorship, no advertising, no analytics. Scoring criteria are versioned in public and every change carries a written reason. Each planning system is also ranked within itself, because comparing two machines is not comparing two planners.
- RadiotherapyReal DICOM, judged the way a plan is judged.
- A CT and structure set in, an RTDOSE out, scored from the dose–volume histogram against criteria you can read before you start. No proxy metric, no opinion.
- CaseThe case is the unit, and it outlives everything built on it.
- Anyone can contribute one; it passes a PHI gate no administrator can override. Competitions come and go over the top of it, and the case stays open to plan afterwards.
- ArenaThe ranking is the point.
- Unlimited attempts, one plan you choose to count, attempts shown rather than capped. Whoever is at the top earns the right to write the notebook for that case.
First, the machines
Radiation is delivered on very different machines
Each panel below is one of them, animated: a C-arm that swings around the couch, a helical unit that spins while the couch slides through, a robotic arm, a gamma helmet, an afterloader, a proton nozzle. Watch how differently each one paints the dose — the same case is a different problem on every machine. That is why plans here are scored from the delivered dose, not the technique, and ranked within your own machine.
C-arm linac
Varian TrueBeam · Elekta Versa · Siemensdose + beamsWhat you are looking at. The gantry swings around a couch that never moves, and the leaves slide to reshape the field at every control point. 24 control points from a real arc DAO solution, at the gantry's own 6°/s.
IMRT and VMAT. Aperture shape carries the modulation, and the gradient is bought with arc geometry and leaf travel.
Animated from real plan and log data.
Helical tomotherapy
Accuray Radixact · TomoTherapydose + beamsWhat you are looking at. The gantry never stops and the couch carries the patient through the bore, while 64 binary leaves snap open and shut. From a real delivery log: 26 rotations, 370 mm of couch travel.
Fan beam, continuous rotation, binary leaves. Modulation is open time; pitch and field width decide where the ripple lands.
Animated from real plan and log data.
Robotic radiosurgery
Accuray CyberKnifedose + beamsWhat you are looking at. A jointed arm carries the linac from node to node around the patient and fires only while parked — the traverse between nodes is most of the treatment time. Beams already delivered stay drawn, so the star of non-shared entry paths builds up.
Hundreds of non-coplanar beams and no flattening filter. Prescribed to a low isodose on purpose — the heterogeneity is where the gradient comes from.
Schematic. Drawn to show the geometry, not measured from a delivery.
Gimballed head
Hitachi OXRAY · ZAP-X · Verodose + beamsWhat you are looking at. The head hangs on two orthogonal gimbals and tilts the whole beam onto a tumour moving with respiration. The aperture is never reshaped; only the axis points somewhere else. Swing angle is shown, and drawn to scale — a couple of degrees is the whole range, which is why the 1.8° symmetry limit is a real constraint.
Pan and tilt instead of a moving couch or a tracking MLC. The distribution keeps its shape, which is the point, and the swing angle is what it costs.
Schematic. Drawn to show the geometry, not measured from a delivery.
Upright
Leo Cancer Caredose onlyWhat you are looking at. Two arrangements side by side. On the left the gantry swings around a patient lying still; on the right the beam is bolted down and the patient stands and turns instead. Same target, same entry angles — the difference is which object the angle belongs to.
Rotate the lighter object. Every convention that begins with a supine patient on a couch has to be restated, starting with which frame an angle is measured in.
Schematic. Drawn to show the geometry, not measured from a delivery.
Gamma unit
Elekta Leksell Gamma Knifedose + beamsWhat you are looking at. Fixed sources around a helmet, all collimated to one focal point. Shots are placed one at a time, each a sphere of dose, and conformity to the irregular outline is composed out of them.
Fixed sources, shots placed by the planner. Conformity is composed from spheres rather than shaped by an aperture.
Schematic. Drawn to show the geometry, not measured from a delivery.
MR-guided linac
Elekta Unity · ViewRay MRIdiandose + beamsWhat you are looking at. The same fraction handled two ways after imaging on the couch. On the left the plan is shifted onto today's anatomy; on the right it is reshaped around it, pulling the isodose off the organ that moved. The organ moves again every fraction.
Image on the couch, then decide whether to move the plan or rebuild it. The difference is roughly half an hour with the patient lying there.
Schematic. Drawn to show the geometry, not measured from a delivery.
HDR brachytherapy
Elekta Flexitron · Varian BRAVOSdose onlyWhat you are looking at. A single source steps through dwell positions inside three catheters, pausing at each. Dose comes from inside the patient and falls as roughly 1/r², which is why the high-dose region hugs the applicator.
Dose from dwell positions inside the patient. The DVH is judged on D90 and V100 rather than coverage against a flat prescription.
Schematic. Drawn to show the geometry, not measured from a delivery.
Intraoperative
Carl Zeiss Intrabeam · Sordina LIAC · Mobetrondose onlyWhat you are looking at. Depth against dose, from an applicator laid on the tumour bed. Both curves are most of the way gone within two centimetres — that steepness is the technique, and it is why the prescription is written at the applicator surface rather than to a volume.
One fraction, in theatre, on tissue that was under a surgeon's hands a minute earlier. The target is a bed rather than a contour.
Schematic. Drawn to show the geometry, not measured from a delivery.
Proton and carbon ion
IBA · Hitachi · Toshiba · Sumitomodose + beamsWhat you are looking at. Depth against dose. Single Bragg peaks are added one by one and summed into the flat spread-out peak a target needs — then the beam stops. The dashed curve is a 6 MV photon for comparison; it never does.
Pencil-beam scanning, and a distal edge instead of an exit dose. Range uncertainty is a planning constraint no photon plan has.
Curves computed from Bragg–Kleeman with range straggling; the spread-out peak is solved distal to proximal. Not a transport calculation.
Not drawn, and scored the same way. External electron therapy, permanent seed implants and radiopharmaceutical therapy all end in a dose distribution over a structure set, which is all the scorer needs — the metrics a case asks for are part of the case, so an implant judged on D90 and V100 is the same machinery as a lung plan judged on V20.
Intraoperative treatment is the exception, and it is worth being exact about why. Classic applicator-based IORT is prescribed from applicator size and a depth-dose table, in a theatre, on anatomy that has just been operated on. There is frequently no planning CT, no structure set and no exported RTDOSE — so there is nothing for a dose-based scorer to read, and no amount of platform would change that. It is on this page because it is part of the field, not because it can be ranked. Where an intraoperative electron plan is planned on a CT with a dose grid, it scores like anything else.
What the labels mean. Scoring reads the RTDOSE and the structure set, so it works the same for every card — that part is not modality specific. The plan summary is separate: it reads beam geometry and monitor units from the RTPLAN, which carries photon and ion beams but not applicator setups, so a brachytherapy plan is scored on its dose and reports no beams. Nothing is inferred to fill the gap.
The loop
How it works
One case goes in; a leaderboard and a write-up come out. Every step is open to anyone with a planning system.
- 1
Contribute
Upload a de-identified CT and structure set. Pick a site template and the scoring criteria are generated with it.
- 2
Validated
Every file is scanned for patient identifiers and checked for geometry — a case carrying them cannot be published, and you see exactly which tags.
- 3
Plan
Download the case and plan it in your own TPS. For helical machines the platform hands you a ready-made script.
- 4
Scored
Submit the dose. The DVH is computed and graded against the published criteria — the same ones for everyone, whatever vendor you plan on.
- 5
Teach
Everyone who planned the case can write it up, and the top three lead the page. Rank is the billing; the write-up is what everyone else came for.
The point of the scoreboard
Ranking is not the reward.
It is the credential.
Plan quality has always been taught by apprenticeship: you learn it from whoever happens to sit next to you. Papers publish the result, textbooks publish the principle, and the part in between — what a good planner actually did, in what order, and why — is never written down.
Here anyone who planned the case can write it up, with their own attempts beside them: every script revision, every score, every plan that did not work. Often the best read is not the winner’s — the planner who finished eleventh after four tries had to fix something, and that is the part you can use.
Every attempt is kept
Not just the plan that won. The chain of what was tried is the lesson.
Numbers are attached
A claim in a notebook links to the score card that backs it, on the same case anyone can download.
Failures are included
Breaching a hard constraint takes the plan off the board rather than costing it points. Your score card still shows exactly which limit and by how much, and resubmitting is free.
The rules
How it stays fair
A leaderboard is only worth entering if the way it is kept is written down. This is that list.
PHI is a hard gate
Birth date, institution, referring physician, operator, station, accession — any value fails the case outright. The contributor sees the offending tags per file and can have the server strip them.
Criteria published first
Scoring criteria are visible before you plan and do not change mid-challenge. If one turns out to be wrong, every affected plan is rescored and everyone is told.
Ranked within your TPS
A robotic plan against a coplanar arc mostly measures the machine. Boards split by delivery geometry — measured from your plan’s own couch angles — so the comparison is between planners.
Unlimited attempts, one final plan
Iteration is the point, so resubmitting is free. You choose which plan counts, and your attempt count sits next to your rank.
The plan script writes itself
The case's scoring criteria become optimizer objectives. Optimize in Precision, upload the dose, and the next revision reacts to whatever missed.
Failures stay visible
A plan that breaches a hard constraint is marked and unranked, not hidden. It is usually the most instructive row on the board.
What you write with
A planning technique is a citation, not an opinion
Every skill states what it does, which machines it applies to, and what it is based on. Where the platform has a recipe for your planning system it gives you the objectives to type in — and marks the ones nobody has verified on a real installation as exactly that.
A head that swings to follow the tumour
Two gimbals tilt the whole beam onto a moving target instead of moving the leaves or the couch. The tilt is not free.
Adapt-to-position or adapt-to-shape
One reuses the plan's shapes against today's position; the other re-contours and re-optimizes. The difference is roughly half an hour.
Avoidance sectors through a serial OAR
Stop the beam entering through the organ you are trying to spare, rather than asking the optimizer to undo it afterwards.
Changing one thing per optimization
Two changes at once and you have learned nothing about either. The chain of versions is the write-up.
Choosing how hot the target gets
A long tail on the target DVH is a decision, not an accident. The prescription isodose line sets it and the ring constraints hold it.
Choosing the prescription isodose line
A robotic plan is prescribed far lower than a C-arm plan on purpose — the heterogeneity is where the gradient comes from.
Open cases
All cases →Prostate-Anatomical-Edge-Cases · Prostate-AEC-002
78 Gy / 39 fx
Prostate-Anatomical-Edge-Cases · Prostate-AEC-003
78 Gy / 39 fx
Prostate-Anatomical-Edge-Cases · Prostate-AEC-001
78 Gy / 39 fx
4D-Lung · 101_HM10395
54 Gy / 3 fx
4D-Lung · 102_HM10395
54 Gy / 3 fx
4D-Lung · 100_HM10395
54 Gy / 3 fx
You have a case
Upload a de-identified CT and structure set. Identifiers are checked before anything is published, and no administrator can wave that check through. Pick a site template and the scoring criteria come with it.
Contribute a caseYou have a planning system
Download a case, plan it the way you would plan it at work, and submit the dose. Resubmitting is free and you choose which plan counts. Once it is scored, the case is yours to write up.
Browse open cases