Radiotherapy and Oncology 62 (2002) 299307

www.elsevier.com/locate/radonline

Margins for geometric uncertainty around organs at risk in radiotherapy

Alan McKenzie a,*, Marcel van Herk b, Ben Mijnheer b
a

Medical Physics Department, Bristol Oncology Centre, Horfield Road, Bristol BS2 8ED, UK
Radiotherapy Department, The Netherlands Cancer Institute, Plesmanlaan 121, 1066 CX Amsterdam, The Netherlands
Received 5 September 2001; received in revised form 24 December 2001; accepted 5 February 2002

Abstract
Background and purpose: ICRU Report 62 suggests drawing margins around organs at risk (ORs) to produce planning organ at risk
volumes (PRVs) to account for geometric uncertainty in the radiotherapy treatment process. This paper proposes an algorithm for drawing
such margins, and compares the recommended margin widths with examples from clinical practice and discusses the limitations of the
approach.
Method: The use of the PRV defined in this way is that, despite the geometric uncertainties, the dose calculated within the PRV by the
treatment planning system can be used to represent the dose in the OR with a certain confidence level. A suitable level is where, in the
majority of cases (90%), the dose-volume histogram of the PRV will not under-represent the high-dose components in the OR. In order to
provide guidelines on how to do this in clinical practice, this paper distinguishes types of OR in terms of the tolerance doses relative to the
prescription dose and suggests appropriate margins for serial-structure and parallel-structure ORs.
Results: In some instances of large and parallel ORs, the clinician may judge that the complication risk in omitting a margin is acceptable.
Otherwise, for all types of OR, systematic, treatment preparation uncertainties may be accommodated by an OR ! PRV margin width of
1.3S. Here, S is the standard deviation of the combined systematic (treatment preparation) uncertainties. In the case of serial ORs or small,
parallel ORs, the effects of blurring caused by daily treatment execution errors (set-up and organ motion) should be taken into account. Near a
region of high dose, blurring tends to shift the isodoses away from the unblurred edge as shown on the treatment planning system by an
amount that may be represented by 0.5s. This margin may be used either to increase or to decrease the margin already calculated for
systematic uncertainties, depending upon the size of the tolerance dose relative to the detailed planned dose distribution. Where the detailed
distribution is unknown before the OR is delineated, then the overall margin for serial or small parallel ORs should be 1.3S 1 0.5s.
Examples are given where the application of this algorithm leads to margin widths around ORs similar to those in use clinically.
Conclusions: Using PRVs is appropriate both for forward and inverse planning. Dose-volume histograms of PRVs for serial- and parallelstructure ORs require careful interpretation. Nevertheless, use of the proposed algorithms for drawing margins around both serial and parallel
ORs can alert the dosimetrist/radiation oncologist to the possibility of high-dose complications in individual treatment plans, which might be
missed if no such margins were drawn. q 2002 Elsevier Science Ireland Ltd. All rights reserved.
Keywords: Radiotherapy; Organs at risk; Geometric uncertainty; Margins; Treatment planning

1. Introduction
Many workers have addressed the problem of geometric
uncertainty in radiotherapy by determining the widths of
margins to draw around clinical target volumes (CTVs)
[13,6,7,9,11,1315]. The converse problem is how to
deal with organs at risk (ORs). ICRU Report 62 [5] suggests
drawing margins around organs at risk (ORs) to produce
planning organ at risk volumes (PRVs) to account for
geometric uncertainty in the radiotherapy treatment process.
In that report it is recommended that an integrated margin be
added to the OR to compensate for any movements of the
OR during the treatment, as well as uncertainties in the set-

up during the whole treatment course. Some clinical examples (treatment of breast cancer, cancer of the prostate and
lung cancer) are given in which PRVs of various organs are
shown.
While a few workers have considered the effect of
geometric uncertainty on the dose to ORs (for example,
refs [4,14]), there is still a scarcity of advice in the literature
dealing with the question of how to draw the margins
recommended by the ICRU. Even in the clinical examples
given in ICRU Report 62, no details are given about the
origin of the margins drawn around the ORs. This paper
suggests an approach to that problem which effectively
mirrors that used in constructing margins around the CTV.
The need to draw margins around a CTV arises because

* Corresponding author.
0167-8140/02/$ - see front matter q 2002 Elsevier Science Ireland Ltd. All rights reserved.
PII: S 0167-814 0(02)00015-4

300

A. McKenzie et al. / Radiotherapy and Oncology 62 (2002) 299307

of geometric uncertainties in the target shape, position and

so on. One option is to draw the margins wide enough such
that in, say, 90% of treatments planned using the margins,
the dose in the CTV will be everywhere greater than a given
minimum, say 95% of the prescribed dose [15]. Another
option is to calculate how wide the margins should be for
the 95% isodose to enclose, on average, at least, say, 99% of
the CTV [14]. Both of these approaches suggest similar
margin widths, although the underlying philosophies are
different.
Conversely, depending upon whether the OR is of a
serial or parallel structure, the OR is supposed to receive
less than a given maximum dose (serial structure) or a distribution of dose characterised by a dose-volume histogram
containing an acceptably small proportion of high-dose
components (parallel structure). The functionality of a serial
OR is compromised when even a small part of it is exposed
to a dose above a critical level: hot spots are not well
tolerated. Hence, for example, doses above 45 Gy are to
be avoided in the spinal cord, even over short lengths. In
contrast, a parallel OR will tolerate small volumes raised
above a critical level of dose, although loss of functionality
is progressive with increase in the volume that is irradiated
above this critical level. An illustration of this might be the
specification of a limit of no more than 200 cm 3 of lung to be
raised above 30 Gy in 25 fractions during radiotherapy of
the breast. In practice, organs lie at intermediate points
along the serial-parallel axis, bracketed by spinal cord and
lung at the extreme ends.
The dose volume histogram (DVH) is the most convenient tool for incorporating such constraints into treatment
plans. With the further implementation of intensity modulated radiotherapy (IMRT) or other conformal radiotherapy
techniques, these dose-volume histograms of the actual dose
given to the ORs will play an important role in the treatment
optimisation process. The use of PRVs might be a tool to
help the treatment planner in designing these optimised
plans.
One should be aware that the size of any type of margin
will differ between treatment techniques and institutions.
Before implementing any form of conformal radiotherapy
one should know the size of the errors in ones own department and reduce them as much as possible.

margins around the OR. The need for a margin around the
OR depends on the types of uncertainties and the dose
constraints for the OR. One should differentiate between
random and systematic geometric errors. Random errors
blur the dose distribution: high isodose lines move towards
the target, whilst low isodose lines move away from the
target. Systematic errors (i.e. errors that are fixed for a single
patient over all fractions but that are stochastic over a group
of patients) shift the dose distribution either closer to or
further away from an OR.
If the organ is parallel and large and if the risk is acceptable, the clinician may decide that the effect of geometric
errors can be ignored. The organ can be considered large
for these purposes if its dimensions considerably exceed the
scale of the geometric uncertainties. For dose levels that
cause unacceptable complications, in particular for serial
organs, the situation is different and geometric errors must
be taken into account. In these patients unacceptable
complications will occur if the tolerance isodose is close
to the OR. We suggest using a margin for the OR in the
latter situation, i.e. when dose levels are used close to an OR
that would cause unacceptable complications. This margin
is then used to draw the PRV. We suggest a definition of the
PRV such that, in 90% of cases, the DVH of the PRV will
not underestimate the contribution of the high-dose components to the OR.
The approach we shall take is.
1. Establish a margin which extends the OR (as defined at
the time of localisation) to encompass completely (at
least, in 90% of cases) the mean position of the OR.
This accounts for the systematic, treatment preparation
uncertainties introduced at localisation.
2. Examine the effect of random, treatment execution
uncertainties on the DVH of the OR.
3. Decide whether, in order to account for random uncertainties, a further margin in addition to the systematic
margin is necessary, or whether the systematic margin
itself can define the PRV, or even whether a margin
reduction may be applied.
4. Highlight the case of critical, serial ORs such as spinal
cord, optic nerve and brain stem, and calculate an appropriate margin to draw around the systematic margin in
order to produce the PRV.

2. Purpose
The aim of this paper is to propose an expression for
drawing an OR ! PRV margin, to compare the proposed
margin widths with examples from clinical practice and to
discuss the limitations of such an approach.
3. Theory
As in the case of the CTV, geometric uncertainties in the
position of the OR may be accommodated by drawing

3.1. Systematic, treatment preparation uncertainties

The treatment of systematic errors in the radiotherapy
process is described in detail, for example, in van Herk et
al. [15], but, for convenience, the approach is summarised
here. Systematic geometric errors arise generally during
treatment preparation such as errors in the CT aiming lasers
or errors in the indicated position in the treatment plan due
to organ motion during imaging. For a given patient, the
combination of these errors is frozen into the image, the

A. McKenzie et al. / Radiotherapy and Oncology 62 (2002) 299307

treatment plan and the treatment itself. For a given patient,

therefore, this error is constant throughout treatment, and,
hence, is a systematic error for that patient. However, the
size of systematic errors arising from the same source (for
example, organ motion or set-up error at the time of
imaging) will be different in different patients.
Errors arising during transfer of the patient parameters
through the system of plan preparation and treatment (as
distinct from those arising from delineation uncertainties
or organ motion) may be investigated by analysing, over a
period of time, the portal images of phantoms that have been
imaged, planned and treated. The scatter of the discrepancies between the intended and the delivered treatment will
indicate the standard deviation of the transfer error (and will
include errors arising from the drift and adjustment of CT
and linac lasers over extended periods of time, ill-matched
immobilisation devices, and differences in couch sag of CT
and linac). The analysis will also reveal any residual (that is,
constant) discrepancies in the process, and, as is consistent
with good radiotherapy practice, action should be taken to
minimise such errors (for example, producing an insert for
the CT couch).
Once the residual error has been minimised, then with the
exception of breathing motion (e.g. refs [8,10]), the combination of all of the positional uncertainties, including transfer error, organ motion and set-up error is then generally
described adequately by a gaussian distribution. The standard deviation of the combined errors is then called S. This
approach to combining the systematic errors is described
more fully in van Herk et al. [15].
3.2. Margin for systematic uncertainties
The quantity S determines the probability that the mean
position of the OR boundary lies within a given distance of
the OR boundary as displayed on the CT scan. Where the
dose distribution within and surrounding the OR is a relatively homogeneous plateau, uncertainty of position does
not affect the accuracy with which the treatment planning
system (TPS) displays the dose in the OR. It is only when a
boundary of the OR lies close to the edge of a high-dose
region, such as a treatment beam or the CTV, that positional
uncertainty can lead to inaccurate portrayal of the dose in
the OR.
In the process of defining treatment margins to account
for systematic uncertainties in the position of the CTV, all
boundaries of the target volume are potentially vulnerable to
underdosing. (This depends upon the treatment technique.
For example, with AP-PA treatments, the boundaries are
more prone to underdosing in the lateral directions than in
the antero-posterior directions.) Therefore the three-dimensional form of the gaussian distribution is used to calculate
the margin width [14,15]. In general, with an OR, a onedimensional solution is more appropriate. In the majority of
cases, the threat of a high dose to the OR comes broadly
from one direction that of the high-dose region of an

301

adjacent treatment beam or target volume. Movement in

any other direction will lower the OR dose or keep it the
same (when the movement is along an isodose).
To determine the margin width in the one-dimensional
case, the area of the gaussian distribution of the positional
uncertainty of the OR is calculated. This shows that there is
only a 10% chance that the boundary of the OR (when in its
mean position) lies closer to the high-dose region than a
distance of 1.28 S from its position as shown in the CT
scan. (Note that the margin width of 1.28 S results from a
one-tailed distribution. This is in contrast with drawing
margins around the CTV for the one-dimensional case
when a two-tailed distribution is appropriate [15]). Hence,
starting from the CT image of the OR and drawing a margin
around it 1.3S in width (to two significant figures) will
ensure that, in any single direction, the mean position of
the edge of the OR in this direction will be encompassed
by this margin in 90% of treatment plans. This margin is
used to construct the PRV.
However, sometimes the one-dimensional solution for
drawing margins around ORs is not enough. For example,
the dose distribution in conformal techniques may be
horse-shoe shaped, and surround an OR such as spinal
cord or rectum. In such cases, a two-dimensional approach
may be required. IMRT techniques may be expected to
produce problems which require consideration of three
dimensions.
In these cases, it is appropriate to consider increasing the
systematic margin from 1.3 S in the one-dimensional case
up to a limit of 2.2 S and 2.5 S in the two- and threedimensional cases respectively. This may be deduced, for
example, by inspection of Table 2 of ref. [15]. It should be
kept in mind, when referring to that table, that it was
designed for a different purpose and, hence, the one-dimensional value given in that table is different from that
proposed in this paper.
3.3. Random, treatment execution uncertainties
Random, treatment execution uncertainties introduced by
organ motion and patient set-up errors are characterised by
standard deviations which may be summed in quadrature to
yield a combined standard deviation, s. Although, in reality,
it is the OR which moves from day-to-day relative to the
room co-ordinates, it is more convenient to consider the OR
as being stationary and the dose distribution to be blurred
[7]. In practice, this means blurring the edges of high-dose
regions including beam edges and the edges of the CTV.
3.4. Role of DVHs in serial and parallel ORs
It is appropriate here to question the utility of margins for
parallel-structure ORs. With serial-structure ORs, the DVH
will show whether any portion of the PRV is irradiated
above the tolerance (critical) dose. If such is the case,
then the treatment plan will need to be adjusted. However,
with a parallel-structure OR, the question is not whether the

302

A. McKenzie et al. / Radiotherapy and Oncology 62 (2002) 299307

critical dose is exceeded, but what volume is raised above

the critical dose. By drawing a margin all around the OR, the
DVH of the PRV is liable to overestimate the total volume
of the OR that is irradiated above the critical dose. This is
because the PRV represents the total volume likely to
contain the mean position of the OR, which is, of course,
greater than the volume of the OR itself. It would have been
better if the treatment planning system could have tested the
dose distribution by moving the complete OR by 1.3S
towards the high-dose region. However, treatment planning
systems are currently not yet designed to do this automatically, whereas they can grow margins around volumes with
ease.
Hence, when a DVH shows that a relatively high volume
of a parallel-structure PRV will be irradiated above the
critical dose, then that DVH must be regarded only as indicative rather than definitive of organ overdose. Nevertheless, as experience of using such PRVs accumulates,
interpretation of DVHs for parallel-structure ORs will
become easier.

broadening caused by the random, treatment execution

errors. The dotted line in Fig. 1(c) illustrates the effect of
the blurring on the DVH, which is to move high-dose
components to lower-dose levels.
The OR will be at less risk of complications from a dose

3.5. Margin for random uncertainties

In order to visualise the effect of random uncertainties it
is convenient to consider an OR in the direct neighbourhood
of a region of high dose (Fig. 1(a)) and the corresponding
DVH (Figs. 1(b,c)). In Fig. 1(a), a stylised OR is represented
by an elongated rectangular slab of tissue, most of which is
assumed to receive a uniform dose from a pair of parallel
and opposed beams (beams 1 and 2). A small section of this
OR penetrates into the high dose region formed by the intersection of the first pair of parallel and opposed beams with
an orthogonal second pair (beams 3 and 4).
It should be appreciated that the DVH of this stylised
arrangement will be identical to the dose profile in the direction of the long axis of the OR provided that:
1. the dose changes only in the direction of the long axis;
and
2. the dose changes only monotonically that is, with
increasing distance along the OR in the direction away
from the high dose region, the dose either decreases or
remains static: it never increases.
These requirements are not crucial to the argument, but they
make it easier to relate the DVH to the isodose distribution.
To illustrate the theory, we have assumed in Fig. 1(a) that
about 20% of the OR penetrates the high-dose region. The
effect of this is apparent in the step in the DVH (solid line in
Fig. 1(b)). The sharpness of the step reflects the narrowness
of the penumbra of the edge of the high dose. All of the OR
is shown in Fig. 1(b) to receive at least 50% of the prescription dose (assuming that, in the idealised four-beam configuration, each of the two pairs of beams contributes a
uniform dose which is half the prescription dose).
What the TPS cannot show is the effect of penumbra

Fig. 1. (a) OR partially penetrating the high-dose region in a four-field beam

configuration. In this stylised representation, dose gradients are assumed
flat everywhere except at the penumbra of the high-dose region. (b) Dose
profile along the OR in 1(a). It will be seen that, for example, 20% of the
length/volume of the OR receives a dose greater than D%. In this very
stylised illustration, we have assumed that the dose never increases with
distance along the OR away from the region of high dose. Therefore, over
the region enclosed by the OR, the dose profile along the OR will be
identical to the DVH of the OR. (c) Dose profile and DVH corresponding
to 1(a) and (b). The effect of blurring caused by random, treatment execution errors is illustrated by the dotted line. The dose profile is shown
extended beyond the boundary of the OR into the high-dose region.

A. McKenzie et al. / Radiotherapy and Oncology 62 (2002) 299307

303

distribution represented by the blurred DVH than one represented by the unblurred DVH. In other words, where the OR
penetrates the high-dose region, the TPS slightly exaggerates the hazard to the OR because it uses the unblurred dose
distribution to calculate the DVH. Hence, in this circumstance, there is no need to add a further margin to the 1.3S
systematic margin. On the other hand, diminishing the
margin around the OR would also be wrong, because blurring will not reduce the maximum dose in the OR if it has
penetrated sufficiently into the high-dose region.
Fig. 2(a) shows an OR which does not quite extend to the
edge of the high dose region and Fig. 2(b) shows the corresponding DVH. In this case, blurring (dotted line) adds a
small high-dose component to the DVH. However, in the
case of large, parallel ORs, the comparatively small extra
high-dose component will be clinically acceptable, and so,
again, the 1.3S systematic margin is sufficient.
On the other hand, in the case of serial ORs, or where the
broadening of the penumbra due to blurring is a significant
fraction of the size of small, parallel ORs, the effect of the
blurring may be clinically significant (see Fig. 3). An

Fig. 3. (a) As in Fig. 2(a) but with a small OR. (b) Dose profile and DVH of
3(a). The dose profile is shown extended beyond the boundary of the OR
into the high-dose region. The effect of blurring caused by random, treatment execution errors is illustrated by the dotted line. Note that the blurred
high-dose component within the DVH contributes significantly to the DVH
because the extent of the OR is comparable with that of the blurring.

Fig. 2. (a) OR does not quite penetrate the high-dose region. (b) Dose profile
and DVH corresponding to 2(a). Again, the DVH may be understood in
terms of the dose profile along the OR. The dose profile is shown extended
beyond the boundary of the OR into the high-dose region. The effect of
blurring caused by random, treatment execution errors is illustrated by the
dotted line.

obvious candidate for such a case is the spinal cord,

where the diameter may be equal to only a few standard
deviations of the set-up error (there is no organ motion to
be considered in the case of the spinal cord). Also, any
overdosing of the cord can be clinically critical. Examples
of similar critical ORs are the optic nerve, the brain stem,
the lens of the eye and the left anterior descending coronary
artery. The DVH of the OR (with its systematic margin) will
be computed from the unblurred dose distribution. This
DVH will not show the additional high-dose components
resulting from blurring, so that, in the case of these organs,
the hazard may be underestimated.
Hence, in the case of the spinal cord, it may be necessary
to add a margin to the systematic margin around the OR to
account for the blurring arising from random, treatment
execution uncertainties.
As we shall discuss in the next section, there may also be
cases where blurring allows us to reduce the margin width.
This arises as a result of the high-isodose lines shrinking
away from the OR as noted earlier. We shall indicate in the
next section how to decide whether to increase the systematic margin to account for blurring or whether to reduce it. In
either case, the change in margin size resulting from blurring should be in the order of 0.5s.
This expression is a somewhat approximate practical
substitute for an explicit blurring of the dose distribution.
However, as may be seen from Fig. 4, this is a reasonable

304

A. McKenzie et al. / Radiotherapy and Oncology 62 (2002) 299307

tice to base clinical decisions on analysis of DVHs for delineated ORs, any such differences should be kept in mind.
This was discussed in detail in Section 3.4 in the context of
parallel-structure organs.
4.1. Spinal cord

Fig. 4. The effect of random, treatment execution errors is to blur the total
dose distribution, which is equivalent to blurring the dose of each beam. In
this figure, the edge of a single beam is considered. Due to beams from
other directions there will in general be a plateau dose, which is 50% in this
example. The 50% intensity of each particular beam will remain in its place
(at X 0), even when the dose is blurred. In the left of this figure (X , 0)
the dose increases with increased blurring, while in the right (X . 0) it
decreases. The shift of the isodose lines depends both on the location in
the dose profile and on the magnitude of the random errors. Typically, the
shift is zero at the block edge of the beam (75% of the total dose in this
example). The shift is around 0.5s with an increase in dose in the left hand
of the figure. The shift is around 0.5s with a decrease in dose at the right
hand of the figure. The arrows showing the extent of blurring are drawn
0.5s long, illustrating that the practical approximation of using a single
value, 0.5s, is a reasonable representation of the blurring over a useful
proportion of the profile.

representation of the size of the blurring over much of the

profile.

4. Clinical examples
Before applying the proposed algorithm in the clinic, it is
necessary to have knowledge about the current practice. The
delineation of the OR should be done in a uniform way for
example, is the spinal canal drawn or is the spinal cord
considered to be the dose limiting organ? One has to be
sure that the clinician or dosimetrist has not already
included organ motion or set-up error in the delineation of
the OR. Quantitative information on the various factors
influencing the systematic and random uncertainties may
be derived locally and from the literature. In this section
we will compare margins for some ORs predicted by the
algorithm outlined in this paper with margin widths used in
clinical practice.
It should be recognised that, despite the efforts to allow
for geometric uncertainty by using the algorithm in the
previous section, the DVH for the PRV will, in general,
differ from that of the OR due to the blurring and shifting
effect of geometric errors. Therefore, if it is common prac-

For the spinal cord, systematic uncertainties in delineation, organ motion and change of organ size and shape are
negligible. The standard deviations of the set-up uncertainty
and that in transferring the image from the CT scanner to the
TPS (systematic errors) are each typically in the order of 2
mm so that S 2.8 mm, and so the margin for systematic
error is 1.3 2.8 3.6 mm. Only set-up error will contribute to the random, treatment execution uncertainty, and so
the corresponding margin for blurring is 0.5 2.0 1.0 mm
(where a value similar to that for the systematic set-up
uncertainty has been used for the treatment execution standard deviation).
Let us consider in more detail the effect of the blurring
(Fig. 4). Suppose that a centre considers the tolerance dose
of the spinal cord to be 45 Gy. Assume also that the
prescribed dose is 80 Gy, and that the cord receives a
plateau dose of 50% of this dose, i.e. 40 Gy. The bending
point of the edge of the high-dose region lies at 60 Gy.
The dose indicated by the TPS is the unblurred dose. If
the dose at the edge of the cord indicated by the TPS is less
than 45 Gy, this would mean that the cord is on the low-dose
side of the bending point, so that blurring will increase the
dose to the cord. Hence, a margin of 0.5s has to be added in
order for the TPS to indicate the increased dose due to
blurring. This may be seen by inspection of Fig. 4 where
the difference between the unblurred penumbra of the highdose region (i.e. as indicated by the TPS) and the blurred
dose may be represented by 0.5s approximately.
If, on the other hand, the prescribed dose is 50 Gy, the
bending point will be at 37.5 Gy. In that case, if the edge of
the cord moves past the bending point, close enough to the
high-dose region to receive the tolerance dose of 45 Gy
according to the TPS, then blurring will decrease the dose
to below 45 Gy. In that case, the margin may be reduced by
0.5s in order for the TPS to indicate the reduction in dose
due to blurring.
More generally, if the plateau dose and the prescribed
dose are known at the time of planning, and if the tolerance
dose for a serial OR is about halfway between the plateau
dose and prescribed dose, then no margin for random errors
is required. If the tolerance dose is lower, the overall margin
should be 1.3S 1 0.5s, while it may be 1.3S 2 0.5s if the
tolerance dose is higher. This margin defines the PRV.
If it is inconvenient to estimate the plateau dose before
delineating the PRV, or if the final prescribed dose is not
known at the time of planning, then it is prudent to assume
the worst case, and to add the margin for blurring. In this
example, the overall margin for systematic and blurring
uncertainties would be 3.6 1 1.0 4.6 mm. This is of the

A. McKenzie et al. / Radiotherapy and Oncology 62 (2002) 299307

same order as the margin widths drawn around the spinal

cord in clinical practice, which can, for instance, be deduced
from Fig. A.3.e in ICRU Report 62.
It is important to note that a margin for the spinal cord
may already have been added as a matter of local protocol,
for example by delineating the spinal canal rather than the
cord.
4.2. Rectum
The situation for the rectum is more complicated than for
the spinal cord because changes of organ size and shape are
no longer negligible. The systematic and treatment execution errors were estimated to be, respectively, S 4 mm
and s 3 mm. These values were deduced from inspection
of repeated CT scans of the rectum over a treatment period
of about 6 weeks, combined with data on set-up and delineation uncertainties taken from an overview of uncertainties in prostate treatment [12].
Typically, the plateau dose on the rectum is about 30% of
the prescribed dose. Since the tolerance dose for the rectum
is close to the prescribed dose, a negative margin for random
deviations can be used, provided that, as has been pointed
out, the OR does not already overlap into the high-dose
region. With that proviso, the algorithm predicts a margin
width of 1.3 4 2 0.5 3 3.7 mm, which is similar in
size to that indicated in Fig. A.2. of ICRU Report 62. It
should be noted that this margin might be different in the
various directions or parts of the rectum, as can be deduced
for instance from Fig. A.2.
5. Discussion
The need for geometric margins depends upon the significance of radiation damage to an OR. Large, parallel ORs
may sustain a degree of damage without compromising their

305

overall function. In such instances, a simple extension of the

OR by a margin of 1.3S will be appropriate, as has been
described above. (Since systematic errors may both increase
or decrease the volume of the parallel organ receiving a high
dose, the clinician may decide not to use a margin at all,
since the net effect of geometrical errors for a population of
patients is small.)
Where the OR is a relatively small parallel structure or
where it is of a serial structure, then blurring caused by
random, treatment execution errors will affect the risk of
complications. On the low-dose side of the bending point
of the penumbra of a high-dose region, blurring will
increase the dose indicated by the TPS and, conversely,
on the other side of the bending point, blurring will reduce
the dose. Provided that the tolerance dose is known in relation to the plateau and maximum dose, a margin for blurring
can be added to the systematic margin or subtracted from it
(or none applied at all) as summarised in Table 1. Where the
plateau dose or the prescribed dose is unknown at the time
of planning, then the prudent course, in the case of serial
structures, is to add a margin for blurring.
We have recommended that the margin to be used for
blurring should be 0.5s. It will be seen by inspecting Fig.
4 that the distance between a given blurred dose and the
same level of unblurred dose is zero at the bending point
and increases as dose is either raised or reduced above or
below the bending point dose. Calculation of the dose difference at any point, X, between the blurred dose and the dose
indicated by the unblurred penumbra shows that using a
margin of 0.5s is a satisfactory representation of this difference. That is, in practice, using such a representative margin
for blurring, the dose indicated by the TPS will generally not
underestimate the dose to the OR by more than 5%, which is
clinically acceptable.
We have defined the OR ! PRV margin such that, in
90% of cases, the DVH of the PRV displayed by the treat-

Table 1
Summary of suggested OR margins a
Situation

Dose constraint

Suggested OR margin

When the risk of ignoring geometric uncertainties may be deemed acceptable

(as in the case of some large, parallel structures)

Any

When a simple margin that accounts for the bulk of the uncertainties is required
(as in the case of some large, parallel structures)

Any

1.3S

When the benefit of a simple margin is outweighed by the need for greater accuracy
(as, for example, in the case of serial ORs or small, parallel ORs)

,40% above plateau dose

1.3S 1 0.5s

4060% above plateau dose

.60% above plateau dose
.60% above plateau dose, s
varies between patients b

1.3S
1.3S0.5s
1.3S

Any

1.3S 1 0.5s

For serial ORs or small, parallel ORs, when the value of the plateau dose or the final
prescription dose is unknown at the time of delineation of the OR
a
b

Note that when multiple dose constraints are set on an OR it may be necessary to define multiple PRVs, one for each dose constraint.
When one cannot count on the dose blurring effect, one should not reduce the margin for it.

306

A. McKenzie et al. / Radiotherapy and Oncology 62 (2002) 299307

ment planning system will not under-represent the highdose components. The choice of 90% is pragmatic. To
achieve 100% certainty that the margin would encompass
the mean position of the edge of the OR, the margin width
would need to be indefinitely wide. Settling on the figure of
90% does not automatically condemn the remaining 10% of
ORs to radiation morbidity. The fraction of the volume of
the OR which breaches the margin limit will be relatively
small in most cases. An alternative to the 90% level may be
suggested for particular organs, but, as we have seen in the
case of the spinal cord, choosing the 90% level leads to
margins similar to those currently in clinical use.
The above derivation of an OR ! PRV margin is one
dimensional because, generally, positional uncertainty of
an OR is unimportant in all directions except for one, in
the direction of the nearest high-dose region. Rarely,
where a high-dose region bends around an OR to a significant extent, the problem can, in theory, become two-, or
even three-dimensional. We noted in Section 3.2 that the
systematic margin would change from 1.3 S in the onedimensional case to 2.2 S and 2.5 S in the two- and threedimensional cases respectively.
It is appropriate to emphasise that the margins suggested
by the model in this paper are of the same order as those
already used in clinical practice. The model, then, may be
viewed as a tool for achieving consistency in defining
margins, which is necessary for the widespread adoption
of the principles of ICRU Report 62 [5].

6. Conclusion
As conformal therapy and IMRT become standard techniques in radiotherapy centres, it is increasingly urgent to
consider an issue highlighted by ICRU Report 62 that of
drawing margins around Organs at Risk. The uncertainty in
the position of the OR at the time of localisation can be
accounted for by drawing a margin 1.3S wide to account
for systematic uncertainties. The daily treatment execution
uncertainties of set-up and organ motion are best visualised
by considering the penumbras of adjacent high-dose regions
to be blurred by the uncertainties. For large, parallel ORs
this blurring is either of no significance (if the OR already
penetrates the high-dose region) or it may be ignored
because the blurring adds a relatively small high-dose
component to the DVH. Hence, for large, parallel ORs,
the PRV is produced by drawing a margin of 1.3S around
the OR. In some instances, the clinician may decide that the
risk of complications in omitting margins for uncertainties is
acceptable.
However, in the case of serial or small, parallel ORs the
blurring may contribute a relatively large component to the
DVH and cannot be ignored. This is particularly important
with ORs such as spinal cord, optic nerve and brain stem,
where overdosing may have a critical effect. The margin for
blurring is 0.5s where s is the standard deviation of the

treatment execution uncertainties. Where the plateau dose

and the prescription dose are known in relation to the tolerance dose, this margin may be either added to or subtracted
from the systematic margin, or no correction may be
needed. Table 1 summarises the options. Where the plateau
dose or the prescription dose are not known, the prudent
course is to add 0.5s to the systematic margin. The agreement of margins calculated using these algorithms with
margins illustrated in clinical examples gives confidence
that use of the algorithms will be consistent with best clinical practice.
The DVH calculated for the PRV determined using these
algorithms cannot be identical to that in the OR. Nevertheless, we believe that use of this PRV in a treatment planning
system to signal the risk of tissue complication is preferable
to using the bare OR.

Acknowledgements
The authors would like to express their gratitude to Drs
Harry Bartelink and Peter Remeijer for having useful
discussions about this topic.

References
[1] Austin-Seymour M, Kalet I, McDonald J, et al. Three dimensional
planning target volumes: a model and a software tool. Int J Radiat
Oncol Biol Phys 1995;33:10731080.
[2] Bel A, van Herk M, Lebesque JV. Target margins for random geometrical treatment uncertainties in conformal therapy. Med Phys
1996;23:15371545.
[3] Ekberg L, Holmberg O, Wittgern L, Bjelkengren G, Landberg T.
What margins should be added to the clinical target volume in radiotherapy treatment planning for lung cancer? Radiother Oncol
1998;48:7177.
[4] Fenwick JD, Nahum AE. Impact of dose-distribution uncertainties on
rectal ntcp modeling. 1: Uncertainty estimates. Med Phys
2001;28:550569.
[5] International Commission on Radiation Units and Measurements.
Prescribing, recording and reporting photon beam therapy (supplement to ICRU Report 50). Bethesda, MD: ICRU Report 62, 1999.
[6] Jones D, Hafermann MD, Rieke JW, Vermeulen SS. An estimate of
the margin required when defining blocks around the prostate. Br J
Radiol 1995;68:740746.
[7] Leong J. Implementation of random positioning error in computerised
radiation treatment planning systems as a result of fractionation. Phys
Med Biol 1987;32:327334.
[8] Lujan AE, Larsen EW, Balter JM, Ten Haken RK. A method for
incorporating organ motion due to breathing into 3D dose calculations. Med Phys 1999;26:715720.
[9] Mageras GS, Fuks Z, Leibel SA. Computerised design of target
margins for treatment uncertainties in conformal radiotherapy. Int J
Radiat Oncol Biol Phys 1999;43:437445.
[10] McKenzie AL. How should breathing motion be combined with other
errors when drawing margins around clinical target volumes? Br J
Radiol 2000;73:973977.
[11] McKenzie AL, van Herk M, Mijnheer B. The width of margins in
radiotherapy treatment plans. Phys Med Biol 2000;45:33313342.
[12] Remeijer P, Geerlof E, Ploeger L, Gilhuis K, van Herk M, Lebesque
JV. 3-D portal image analysis in clinical practice: an evaluation of 2-

A. McKenzie et al. / Radiotherapy and Oncology 62 (2002) 299307

D and 3-D analysis techniques as applied to 30 prostate cancer
patients. Int J Radiat Oncol Biol Phys 2000;46:12811290.
[13] Roach M, Pickett B, Rosenthal SA, Verhey L, Phillips TL. Defining
treatment margins for six field conformal irradiation of localised
prostate cancer. Int J Radiat Oncol Biol Phys 1994;28:267275.
[14] Stroom JC, de Boer HC, Huizenga H, Visser AG. Inclusion of geome-

307

trical uncertainties in radiotherapy treatment planning by means of

coverage probability. Int J Radiat Oncol Biol Phys 1999;43:905919.
[15] van Herk M, Remeijer P, Rasch C, Lebesque JV. The probability of
correct target dosage: dose-population histograms for deriving treatment margins in radiotherapy. Int J Radiat Oncol Biol Phys
2000;47:11211135.