IMRT Inverse Planning

Question 1

How does forward planning work?

Answer 1

Planner specifies beams
TPS calculates dose distribution
Planner assesses plan and alters beams in iterative process

Question 2

How does inverse planning work?

Answer 2

Planner specifies dose distrubtion
TPS calculates beam profiles, iterates using cost function
Planner assesses plan and alters dose limits/ constraints if necessary

Question 3

What is a cost function?

Answer 3

Summarise the merit of the plan into a single figure used to drive optimisation. Usually uses dose, interest on basing it on biological things like TCP

Question 4

When is a cost function zero?

Answer 4

For a plan where all objectives are met

Question 5

How long do you let TPS iterate?

Answer 5

Could set x number of iterations
Could stop when iterations are barely changing cost function

Question 6

What are objectives vs constraints?

Answer 6

Constraints must be achieved
Objectives are things that would be good

Remember, don’t need to use clinical goals, could change dose instead of penalt.

Question 7

What are examples of physical constraints?

Answer 7

Minimum MU, maximum leaf speed

Question 8

Example of cost function equation

Answer 8

C = sum Pi (Di (x) - Dp)^2

Pi is penalty, Di and Dp are calculated and prescribed doses

Question 9

What kind of optimisation algorithms can be used?

Answer 9

Analytical techniques (eg inverse CT)
Iterative techniques (1. deterministic methods, 2 stochastic methods)

Question 10

What is deterministic method and an example of it?

Answer 10

A deterministic method gives you the same result for the same input
Output determined by parameter values and initial conditions
An example is the gradient method

Question 11

How does gradient method work?

Answer 11

Follows path of gradients to minima, but can end up in local minima.

Question 12

What are stochastic methods and give an example

Answer 12

A method where the same result may not come from the same input, eg simulated annealing.

Question 13

How does simulated annealing work?

Answer 13

Step size and direction are random, explore the whole space to find global minima. Requires more steps. Start with large ones and get smaller, steps can increase the cost function with certain probability. Size of step and probability of accepting increase cost function decrease over time

Question 14

Beamlet based optimisation

Answer 14

Field split into beamlets with fluence and corresponding dose distributions computed
Beamlet weights optimised to produce optimal fluence map
Optimal fluence map translated into deliverable segments (S+S) or leaf trajectory (dynamic)
Final dose distribution, slightly degraded vs optimal

Question 15

Aperture based optimisation

Answer 15

Leaf segmentation step is eliminated
Segment shapes and weights are optimised together
Takes MLC limitations into account during optimisation

Question 16

Fewer vs more levels in optimisation

Answer 16

Fewer, less segments, greater efficiency
More, closer to optimum but less efficient (slower to deliver)

Question 17

Why might we not just want biological optimisation?

Answer 17

TPS can do things we aren’t happy with, give extremely high dose to the PTV, don’t have clinical evidence that this is not harmful

Question 18

Things to consider when planning

Answer 18

Beam arrangement
Help/tuning volumes
Dose prescription
Planning with class solutions

Question 19

Planning with class solutions?

Answer 19

Set of inverse planning parameters used for patients with disease at certain staging
EG, outline these volumes, use this number of beams at this orientation and optimise using this

Question 20

Help/tuning volumes

Answer 20

What are overlap regions assigned to?
Clip PTV from skin
Extent fluence for breathing motion
Dummy volumes to prevent hotspots
Subvolumes to deal with known conflicts

Question 21

Dose prescription planning considerations

Answer 21

Normalise to volume rather than point
Report near min and near max not absolute

Question 22

Beam arrangement planning considerations

Answer 22

Odd number of evenly spaced beams avoids overlap
Beam direction less important as beam numbers increased
Coming in through an OAR may give a better plan
Choice of beam energy has little impact on optimisation - above 10MV rarely used

Question 23

What are some advanced planning tools?

Answer 23

Multi criteria optimisation
Knowledge based planning
Robust optimisation
Arc duplication

Question 24

When is a treatment plan pareto optimal?

Answer 24

When there is no other plan that is at least as good in all objective functions and strictly better in one objective function

Question 25

What is the process of MCO?

Answer 25

A set of pareto optimised solutions are produced - different goal for each, optimise dose to OAR or PTV
An interactive plan navigation tool used to explore optimal trade off between planning goals for patient

Question 26

What is knowledge based planning?

Answer 26

An approach directly utilising prior knowledge or experience to predict an achievable dose in a new patient of a similar population or to derive a better starting point for further optimisation

Atlas based or model based

Question 27

How does knowledge based planning work?

Answer 27

Creates models using good quality TPs for particular site
Uses model for new patients to predict optimal dose distribution

(Atlas based selects closest matching patient from databse and gives starting point from there)

Question 28

What do results of KBP depend on?

Answer 28

Quality of input data
How good model is
Whether new patient is consistent with training population

Question 29

What is robust optimisation?

Answer 29

A plan that considers the effect of possible errors into account, should produce a better plan under error conditions.
Note that this may not be as good as non-robust plan under normal conditions

Question 30

What is arc duplication?

Answer 30

Allows the targeting of PTV sectors to improve OAR sparing - splits PTV into multiple sections and does arc per section.

Question 31

How does model-based KBP work?

Answer 31

Use a large number of previous high quality plans to model the relationship between the patient anatomy and the achievable dose distribution, which can be used to predict the achievable dose distribution for a new patient in the same population.