1

Errors and margins in image guided radiation therapy

Marcel van Herk, David Jaffray*, Anja Betgen, Peter Remeijer, Jan-Jakob Sonke, Roel Steenbakkers, Monique Smitsmans, Marnix Witte and Joos Lebesque

Netherlands Cancer Institute, Amsterdam, The Netherlands

*Princess Margaret Hospital, Toronto, Canada

This work was sponsored by the Dutch Cancer Foundation, NIH and Elekta Oncology Systems

Classic radiotherapy procedure

Tattoo, align and scan patient

Draw target and plan treatment on RTP

Align patient on machine on tattoos and treat (many days)

In principle this procedure should be accurate but

Problems in radiotherapy:The patient is nervous, did not sleep the night before and lay

wriggling on the CT scanner

The physician was in a rush when drawing the target volume

The patients belly flopped from day to day, letting the skin marks move all over the place

The patient was breathing

And the tumor shrank away from its original place

Motion counts: prostate trial data (digital image analysis of 660 patients)

Risk+ rectal distention Crevoisier et al, IJROBP 2004Heemsbergen et al, IJROBP 2006

How can we solve this problem ?

1. Use large margins, irradiating too much healthy tissues

2. Use small margins, and risk missing the target

3. Or: use image guided radiotherapy

Image Guided Radiotherapy

The gain of image guidance depends on the magnitude of geometrical uncertainties and variations that are and aren t addressed by the image guidance protocol

Image guidance will introduce its own uncertainties and variations

Let s look at geometrical errors in RTImaging errorsPlanning errorsTreatment errorsImage guidance errors

Increase precision by imaging target and/or healthy tissues just prior to treatment

2

NomenclatureGross error: mistakes, transcription errors, software faults:

must be caught by QA, not in this lecture

Error: difference between planned measurand and its true value during treatment, however small

Uncertainty: the fact that unpredictable errors occur quantified by standard deviations

Variation: the fact that predictable or periodic errors occur quantified by amplitude or standard deviations

Nomenclature

Systematic error: average difference between planned and executed treatment

Patient group errorsInter-patient errors

Random error: uncertainty and variation in difference between planned and executed treatment

Inter-fraction errorsIntra-fraction errors

Did you do a good job planning the treatment?

Imaging errors

CT scan is just a random snapshot of a changing patient

Organ motion and setup error are frozen in arbitrary position

Interference between motion and imaging distorts image contents

The beams will be pointed to the target in this image systematic error !

Better imaging in adaptive RT: use multiple scans to get better estimate of prostate position

Day 1 Day 2

Day 3 Composite

Match up to 10 scans

Paint each prostate

Automatic generation:

- composite

- mean

Yan et al; IJROBP 96

Better imaging: respiration correlated CT & PET

Free-breathing What it should be!

Allows determination of correct shape, SUV, mean position and trajectory of tumor

Fused 4DCT and 4DPET: Wolthaus et al, PMB 2005

3

Main planning error: GTV/CTV delineation

- 11 observers from 5 institutions, 22 patients- newly developed delineation software (runs from CD)- delineation on CT + (one year later) CT+PET

Steenbakkers et al, IJROBP 2005

CT (T2N2)

SD 7.5 mm

CT + PET (T2N1)

SD 3.5 mm

Delineation variation: CT versus CT + PET

The beams will be pointed to the target the physician draws !

4.210.2Total

8.214.6Lymph nodes

4.819.1Tumor atelectasis

3.74.0Tumor chest wall

4.47.4Tumor mediastinum

3.35.9Tumor lung

CT + PETSD (mm)

CTSD (mm)

Interface / region

Delineation variation in lung cancer

Steenbakkers et al, IJROBP 2005

Main errors in image guided RTImaging (planning CT) and planning errors

Systematic error not solved by image guidance

Observer errors in image guidanceRandom and systematic

Short-term (intra-fraction) motionNegligible for bony anatomyRandom and systematic soft tissue

Inadequacy of surrogate for tumor position

Machine calibrationGroup systematic error

This is what IGRT solves: setup errors -measured with CBCT at NKI

Van Herk et al EPI 2K4, Borst et al, IJROBP 2007

Elekta Synergy system Bony setup error: 3 mm SD

Are you an accurate observer ?

4

Reference Localization

IGRT software: automatic bone localization IGRT software: automatic bone localization

Reference Localization

Cone beam CT series matched to planning CT on bone

Estimated match accuracy << 1 mm SD

-10.0

-8.0

-6.0

-4.0

-2.0

0.0

2.0

4.0

6.0

8.0

10.0

-10.0 -5.0 0.0 5.0 10.0

Correlation CBCT EPID for lung

-10.0

-8.0

-6.0

-4.0

-2.0

0.0

2.0

4.0

6.0

8.0

10.0

-10.0 -5.0 0.0 5.0 10.0

-10.0

-8.0

-6.0

-4.0

-2.0

0.0

2.0

4.0

6.0

8.0

10.0

-10.0 -5.0 0.0 5.0 10.0

Left-Right Sup-Inf Ant-Post

Setup error measured with cone beam CT

EPI

D

EPI

D

EPI

D

Observer error in CBCT is negligible large observer error in EPID such as a clear under-estimation of AP shift (lung)

Slope: 0.81 .. 1.06 0.66 .. 0.88 0.30 .. 0.66 (95% CI)

Borst et al, IJROBP 2007

Difference 0.2 3.7 mm

LR

Difference 2.7 3.9 mm

SI

Difference 1.6 3.1 mm

AP

Observer error in US prostate localization

Comparison of marker and US based prostate localization:

Langen et al, IJROBP 2003Van den Heuvel et al, IJROBP 2003

Automatic prostate localization in CBCT (30 s)

Cone beam CT

Planning CT contours placed automatically

10 CBCT scans: automatic bone match

10 CBCT scans: automatic prostate matchhelp line (GTV+3.6 mm)

Smitsmans et al., IJROBP 2004, 2005

LR (mm)

CC (mm)

AP (mm)

Mean 0.2 -0.4 -0.9

SD 1.0 2.4 2.3

Observer error:(calcifications)

5

Does the tumor move after imaging ?

Intra-fraction patient motion (bone) negligible example:

6 bladder cancer patients35 x 2 CBCT scans10 minutes between post- and pre-scanmeasured set-up changes:

0.3

SD

anterior-posterior (mm)

cranio-caudal(mm)

left-right (mm)

0.5-0.100.40post-pre

SDmeanmeanSDmean

A. Betgen et al, ESTRO 2005

Short-term prostate motion (1 h)

Data courtesy of Jaffray and Gilhezan, Beaumont

Main problem for any prostate IGRT: moving gas

cone-beam CT scanProjection images

Moving gas reduces image quality and introduces short term motion

Are you using a good surrogate for the tumor

position?

The tumor changes: repeated 4D cone beam CT

6

Baseline motion: 4D scans taken within one week and matched on bone, displayed in same phase

Imagine treating this patient with gating and a small margin, without 4D cone-beam CT!

Mean Tumor Position Variability (4DCBCT)

Mean SD

0 0.5 cm (10 Pts)

0.5 1.0 cm (8 Pts)

1.0 1.5 cm (9 Pts)

1.5 2.0 cm (5 Pts)

Amplitude:

Jan-Jakob Sonke et al, IJROBP 2007On average 3 mm SD

What should the margin be ?

Analysis of motion(random and systematic errors)

mean =M

RMS = SD =

Intra-fraction

0.0

0.3

0.4

0.1

0.3

_________

Mean = 0.2

RMS of SD = f

patient 1 patient 2 patient 3 patient 4fraction 1 0.5 0.0 0.2 0.7fraction 2 0.6 -0.5 0.3 0.2fraction 3 0.9 0.2 0.2 -0.4fraction 4 1.3 -1.1 0.3 -0.1

mean 0.8 -0.4 0.3 0.1sd 0.3 0.6 0.1 0.5

van Herk et al, Sem Rad Onc 2004

M = group systematic error (equipment)= standard deviation of the systematic (preparation) error= standard deviation of the random (execution) error

f = standard deviation of the intra-fraction motion{

Definitions (sloppy)

CTV: Clinical Target VolumeThe region that needs to be treated (visible plus suspected tumor)

PTV: Planning Target VolumeThe region that is given a high dose to allow for errors in the position of the CTV

PTV margin: distance between CTV and PTV

Don t even think of using an ITV! (SD adds quadratically)

Demonstration errors in RT

Margin between CTV and PTV: 10 mm

Errors:Setup error:

4 mm SD (x, y)

Organ motion: 3 mm SD (x, y)10 mm respiration

Delineation error: optional

7

What is the effect of geometrical errors on the CTV dose ?

Treatment execution (random) errors blur the dose distribution

Preparation (systematic) errors shift the dose distribution

dose

CTV

CTV

Error distributionsCentral limit theorem:

the distribution of the sum of an increasing number of errors with arbitrary distribution will approach a Normal (Gaussian) distribution

Large errors happen sometimes if all or most of the small sub-errors are in the same direction

Normal distribution:

-3 0 30

200

400

600

800

1,000

1,200

1,400

mean = 0s.d. = 1N = 10000

-2..2 = 95%

Analysis of CTV dose probabilityBlur planned dose distribution with all execution (random) errors to estimate the cumulative dose distribution

For a given dose level:

Find region of space where the cumulative dose exceeds the given level

Compute probability that the CTV is in this region

Computation of the dose probability for a small CTV in 1D

x

dose

x

P

..and compute the probability that the average CTV position is in this area

In the cumulative (blurred) dose, find where the dose > 95%

98%

95%

average CTV position

What should the margin be ?

0 100minimum CTV Dose (%)

prob

abili

ty (

%)

0

100

0 mm

6 mm

9 mm

12 mm

How to choose the PTV margin

Express required CTV dose for a specified fraction of patients. For example: 90% of the patients must get a minimum CTV dose of 95% or more

Add first margin so that 90% of the preparation (systematic) errors are covered

Add margin for penumbra and execution (random) variation so that CTV + first margin lies within the 95% isodose

8

Simplified PTV margin recipe for dose - probability

To cover the CTV for 90% of the patients with the 95%isodose (analytical solution) :

PTV margin = 2.5 0.7

quadratic sum of SD of all preparation (systematic) errors quadratic sum of SD of all execution (random) errors

van Herk et al, IJROBP 47: 1121-1135, 2000)

*For a big CTV with smooth shape, penumbra 5 mm

What about respiration ?

0 0.5 cm (10 Pts)

0.5 1.0 cm (8 Pts)

1.0 1.5 cm (9 Pts)

1.5 2.0 cm (5 Pts)

Amplitude:

Jan-Jakob Sonke et al, IJROBP 2007SD = 0.35 peak-peak

Computing margins

A=10 mm 3.3 mmTreatment respiration

22 mmMargin M

Lung classicError (SD)

4 mmDelineation

4 mmTreatment setup

3 mmTreatment organ motion

A=10 mm 3.3 mmImaging snapshot respiration

3 mmImaging snapshot organ

4 mmImaging snapshot setup

7.05.2M

van Herk et al IJROBP 2000

2.5 + 0.7 is a gross simplificationDose gradients ( penumbra = p) very shallow in lung small margins for random errors

Number of fractions is small in hypofractionationBUT: beam on time is very long respiration only causes dose blurring

Dose prescription at 80% instead of 95%

Respiration is not gaussian asymmetry

222 64.1)(64.15.2 ppM

222 84.0)(84.05.2 ppM

Margins in lung hypo (3 x 18 Gy)

0 mm7 mm +Margin A=10 mm

7 mm +

3 mm SD

1.5 mm SD

1.5 mm SD

2 mm SD

Systematic

2 mmMargin A=20 mm

2.2 mm SDTotal

1.5 mm SDIntra-fraction motion

1.5 mm SDRegistration/couch shift

-Delineation

Random

222 84.0)(84.05.2 ppM p 7.8 mm

Ensures 80% isodose encompasses GTV 90% of time in lung

Future developments

9

Incorporation of remaining geometrical errors in radiotherapy planning

blurred dose

CTV

PTV

planned dose

Correct - blur for random and test all systematic errors

Wrong - compute DVH and TCP for PTV

Uncertainty management: Conventional IMRT planning with margin

CTV PTV

Inverseoptimization

Objective functionsPoisson cell kill, EUD,

DVH points, ...

Dose distribution

90% prob. ofD 95% Dprescribed

in CTV

M = 2.5 +0.7

OAR

Witte et al, IJROBP 2007

Uncertainty management: Probabilistic biological IMRT planning without margin

CTV

Inverseoptimization

Objective functionswith simulated errors

TCP , NTCP

Dose distribution

Maximum TCPfor given

OAR NTCP

OAR

,

no PTV margin!

Witte et al, IJROBP 2007

Best current practice versusProbabilistic planning

74.1 Gy64.6 Gy39.0 Gy

78.0 Gy

SIB Probabilistic (PTVs not used)

CTVCTV CTVCTV

RectumRectumRectumRectum axial

Same control, complications down with 50%Witte et al, IJROBP 2007

ConclusionsThere are many error sources in radiotherapy

Determine what these error sources are and what their impact is in your department

Focus on correcting remaining systematic errorsDo not forget the doctor s error delineation4D CT and portal imaging can half the margin for lung

IGRT does not eliminate all errors; carefully consider the margins to be used

IGRT introduces new errors and makes old errors more important (where is the CTV?)

Margin recipes assume that you know ALL ERRORSUSE AT YOUR OWN RISK

Thank you for your attention!

IGRT+IMRT

US