Intensity-Modulated Radiotherapy and Inverse Planning

C-M Charlie Ma, Ph.D.

Department of Radiation Oncology

Fox Chase Cancer Center

Philadelphia, PA 19111, USA

Outline

Rationale for intensity-modulated radiotherapy

The IMRT process

Elements of an inverse planning system

Concepts of inverse planning

Inverse planning algorithms

Clinical Rationale for IMRT

To improve local-regional control through dose escalation

to improve overall survival

To reduce normal tissue complications to improve quality of life

To reduce treatment time/cost

IMRT Plan for Vertebral Body Tumor

Prostate and

Nodes

In-house hypofractionated protocol

70.2 Gy - 2.7Gy fx

Rectum 65Gy <15%, 31Gy <40%

Posterior margin: 4mm, the rest 8mm

Prostate AcceptanceProstate Acceptance

IMRT - A Complex Process

Planning

Plan Verification

Position Verification

target localization

Treatment Delivery

and Verification

Delivery

Structure Segmentation

Treatment Optimization

Patient Immobilization

Bite-block Head Holder

ImmobilizationAquaplast

Breath Control Spirometer

Vacuum frame

CT/MRI/PET Image Acquisition

Structure Segmentation

Tumor

Node #1

Tumor

Node #2

Tumor

Image Guidance (PET/CT)

Treatment Delivery: Multileaf Collimator

Beam Delivery with a MLC

Beam Delivery with a MLC

0.03 MU Dose Delivery

2.5 mm spatial longitudinal

5.0mm spatial lateral

Target Localization

CT-on-rails

BAT

Intensity Modulation and Treatment Optimization

Beam Intensity Modulation

1 cm

1 cm

fluence

Does intensity modulation improve the dose distribution?

Intensity Modulated Radiotherapy

It works!

Conventional treatment planning starts with a set of beam weights and obtains aplan by a trial-and-error process.

This procedure won’t work for IMRT since there are too many unknowns (>2000 beamlet weights).

How can we determine the individual

beamlet weights for IMRT ?

70%

80%

90%

!!

!

40%

Conventional Treatment PlanningForward Planning

70%

80%

90%

??

?

40%

!

IMRT Treatment PlanningInverse Planning

What is in an Inverse Planning System?

Dose calculation

Interface with R&V

Optimization

Patient data

Leaf sequencing

Di= Cij Wj

-- Weight for beamlet jwj

Cij-- dose contribution in voxel i from beamlet j in an open beam

i

j

Dose Calculation for IMRTTotal dose in voxel i

Or dose in any voxel in a more generic form

jijnj

i WCD

1

i

j

CWD

What’s Inverse Planning ?

Assume D0 is the desired dose

and W0 the required beamlet

weights and we have

00 CWD

1000 CD/CDW

However,Unfortunately this inverse process does not work

in most, if not all, realistic treatment cases.

Practically, what we want is a set of beamlet weights

that will give us the best available dose distribution !

is an exact mathematical expression of inversely derived beamlet

weights for a desired dose distribution D0

100 CDW

What’s Our Solution ? Assume D0 is the desired dose and W0 the required

beamlet weights

What we want is to derive Db the “best achievable” dose and Wb the corresponding beamlet weights

00 CWD

bb CWD

The question is how do we know Db is good enough compared with D0?

Ideal but may be a pie in the sky!

Not ideal but achievable

What’s an Objective Function ?

• An objective function is a mathematical evaluation of a treatment dose distribution (wrt. the desired dose distribution).

• The question now is how to “optimize” a given objective function.

),()( 00 bb DDforDDffunctionObjective

• Ideally, it should include all of our knowledge of radiotherapy: physical as well as biological dosimetric requirements.

A Sample Objective Function • A simple dose-based objective function takes the form

...)}2()2({)}1()1({ 20

20 bb DDDDO

Objective function

Iteration step20 40 60 80

0.10

0.20

0

0.30

There are many ways to optimize a treatment plan for a given objective function (forward,

backward, hybrid, etc)!

An inverse planning system may use any optimization algorithms (more likely it is a forward

planning, or a hybrid, process).

There are many ways to build an objective function (everybody wants his/her own)!

Parallel vector method (?)

Gradient method (Helios, Pinnacle)

Computer simulated annealing (Corvus)

Optimization of a Multi-Dimensional Objective Function

Other iterative methods (CMS)

Filtered back-projection (Konrad)

Major differences between optimization systems are the construction of the objective function and the methods for search directions and step-length

Most optimization methods use an iterative approach, one way or another

Computer Simulated Annealing

Global minimumSmall perturbation to avoid local minima

100

50 2575

A random walk

Other iterative methods

Global minimum

100

50 2575

Not a random walk

Gradient Method

Global minimum

100

50 2575

local downhill gradient [ -grad f(wi)].

Parallel Vector Method

Global minimum

100

50 2575

Independence and local minimum avoidance

How Inverse Planning Is Done?

Inside a computer

W is generally not optimal

Optimal Input WOutput D0

Compute C

Change Wb

Compute Db=CWb

Evaluate O=f(Db-D0)

Experience is gold!

Beam orientation

Number of beams

Factors Affecting Optimization Results

Optimization algorithm and objective function

Optimization parameter and dose constraints

Regions for forced dose gradient

Region1

Region2 Region

3

Region4

Region5

Region6

PTV

CTV

Prostate Plan with 5 intensity levels, 7 beam directions

60%

PTV

100%

50%

49 segments ~ 11.8 min (6MV)

38 segments ~ 9.1 min (6MV)

PTV

100%

50%

Prostate Plan with 5 intensity levels, 6 beam directions(using a forced dose gradient method)

Conclusions

An inverse planning system does not give an optimal plan, but a customized plan

Inverse planning generally works but it is not magic!

It works better for you if you know how it works

Conclusions (cont.)

If it does not work, it’s more likely due to the complexity of the situation…

Conclusions (cont.)

If it does not work, maybe the situation is too simple ...

Conclusions (cont.)

Fortunately, we are very familiar with the situation and we also learn from each other. Therefore, we reach more or less the same goal ...

Conclusions (cont.)

Treatment optimization is an integral part of IMRT

Much more work is needed for the clinical implementation of IMRT

Much more effort is needed to keep it running smoothly and keep pace with upgrades and future enhancements

Thank You