Skills

The MLC is 30–40% faster and slightly less conformal — which matters depends entirely on the case.

Three numbers decide most of a helical plan. There is a published Pareto front and a rule for pitch — use them instead of the defaults.

Two different tools with the same name. One removes entry and exit, the other only removes entry.

Two settings most planners leave at default that measurably change the plan, especially with multiple targets.

No gantry, no couch, no MLC. Two coupled gimbals, a wheel of circular collimators, and beams from a few discrete nodes.

Two gimbals tilt the whole beam onto a moving target instead of moving the leaves or the couch. The tilt is not free.

The beam is fixed and the patient turns. Every convention that starts from a supine patient on a couch has to be restated.

One reuses the plan's shapes against today's position; the other re-contours and re-optimizes. The difference is roughly half an hour.

A 4D-CT gives you ten breathing phases and one decision: which of them the target is drawn on. That decision sets the margin, not the other way round.

Below the cord there is no cord. What you contour instead changes the constraint, and contouring the nerve roots individually is not reproducible.

Dice tells you almost nothing about whether a contour will change the plan. Review where the dose gradient is, not where the volume is.

Ask for a structure in words and get a 3D mask. Useful precisely for the organs a challenge's structure set left out.

A peripheral lesion sits against ribs. The dose that reaches them is the most common late toxicity of lung SABR, and it is entirely a planning decision.

Constrictors and larynx can be spared at no measurable cost to target coverage or parotid dose. Most head and neck plans simply never ask.

Everyone is scored on the same structure set, so an error in it moves every plan the same way — and no leaderboard can show that.

A shell a few millimetres outside the PTV, pushed hard, to force the gradient down instead of hoping for it.

Five concentric shells at different distances, each doing a different job — conformity, fall-off, then spillage.

Contour the hotspot itself, then constrain it. Faster and more local than raising the global max-dose weight.

When the target overlaps something you must spare, split the target rather than asking the optimizer to satisfy both in the same voxels.

Drawing a structure along the entry path and constraining it hard is not a style — it is what you do when your machine cannot remove the direction geometrically.

Stop the beam entering through the organ you are trying to spare, rather than asking the optimizer to undo it afterwards.

A disqualifying limit is not one objective among many. Spend everything else before you spend it.

Two changes at once and you have learned nothing about either. The chain of versions is the write-up.

A model predicts achievable DVHs from your anatomy. It is only as good as the plans it was trained on — including their bad habits.

Compute a set of Pareto-optimal plans once, then slide between them in real time. The trade-offs become visible instead of guessed.

Three questions, in order, that tell you which knob to turn — and stop you turning all of them.

A robotic plan is prescribed far lower than a C-arm plan on purpose — the heterogeneity is where the gradient comes from.

Conformity says how well the prescription hugs the target. These say how fast the dose leaves, which is the part that decides whether a SABR plan is good.

MU per Gy is the crude measure. MCS is the one with a published correlation to QA failure — and the correlation is only moderate.

In lung, a pencil-beam and a Monte Carlo calculation of the same plan disagree by up to 10%. Nothing about the plan changed.

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.