ARTICLE IN PRESS

ORIGINALARBEIT

Normal tissue dose-effect models in biological dose optimisation$

Markus Alber
Sektion für Biomedizinische Physik,Uniklinik für Radioonkologie Tübingen, Hoppe-Seyler-Strasse 3, D-72076 Tübingen

Received 10 May 2007; accepted 8 August 2007

Abstract Dosis-Effekt-Modelle für Normalgewebe für

die biologische Dosisoptimierung
Sophisticated radiotherapy techniques like intensity mo-
dulated radiotherapy with photons and protons rely on Zusammenfassung
numerical dose optimisation. The evaluation of normal
tissue dose distributions that deviate significantly from
Moderne Strahlentherapietechniken wie die intensi-
the common clinical routine and also the mathematical
tätsmodulierte Strahlentherapie mit Photonen oder Pro-
expression of desirable properties of a dose distribution
tonen sind abhängig von numerischen Dosisoptimierungs-
is difficult. In essence, a dose evaluation model for nor-
algorithmen. Sowohl die Bewertung von Dosisverteilun-
mal tissues has to express the tissue specific volume
gen in Normalgeweben, welche stark von der üblichen
effect. A formalism of local dose effect measures is pre-
klinischen Routine abweichen, als auch die mathemati-
sented, which can be applied to serial and parallel res-
sche Beschreibung der wünschenswerten Eigenschaften
ponding tissues as well as target volumes and physical
einer Dosisverteilung ist schwierig. Im Wesentlichen
dose penalties. These models allow a transparent descr-
muss ein Dosisevaluationsmodell für Normalgewebe
iption of the volume effect and an efficient control over
den spezifischen Volumeneffekt erfassen. Es wird ein
the optimum dose distribution. They can be linked to
Formalismus für lokale Dosis-Effekt-Maße vorgestellt,
normal tissue complication probability models and the
welcher sowohl auf seriell als auch auf parallel reagie-
equivalent uniform dose concept. In clinical applications,
rende Normalgewebe sowie auch auf Zielvolumina und
they provide a means to standardize normal tissue doses
physikalische Penaltyfunktionen angewendet werden
in the face of inevitable anatomical differences between
kann. Diese Modelle erlauben eine transparente
patients and a vastly increased freedom to shape the do-
Beschreibung des Volumeneffektes und eine effiziente
se, without being overly limiting like sets of dose-volume
Kontrolle über die Optimierung der Dosisverteilung. Die
constraints.
Modelle stehen in Zusammenhang mit Normalgewebe-
Komplikationswahrscheinlichkeiten und dem Equivalent-
Uniform-Dose-Konzept. In der klinischen Anwendung
bieten die Modelle eine Möglichkeit, die Normalgewe-
bedosen zu standardisieren trotz der unvermeidlichen
anatomischen Unterschiede zwischen Patienten und der
großen Freiheit, die Dosisverteilung zu gestalten, ohne
diese jedoch zu sehr einzuschränken.

Schlüsselwörter: Biologische Optimierung, intensitäts-

Keywords: Biological optimisation, IMRT modulierte Strahlentherapie

$
Wissenschaftspreis der DGMP 2006
 Tel.: +7071 2986055; Fax: +7071 294820.
E-mail: markus.alber@med.uni-tuebingen.de
1. Introduction
The advancement of dose delivery technology for in-
tensity modulated radiotherapy with photons or protons
(IMRT, IMPT) has made numerical dose optimisation a

Z. Med. Phys. 18 (2008) 102–110

doi: 10.1016/j.zemedi.2007.08.002
http://www.elsevier.de/zemedi
 ARTICLE IN PRESS
M. Alber / Z. Med. Phys. 18 (2008) 102–110
necessity. The price for the new freedom in dose shaping
offered by these technologies has to be paid by a loss of
transparency of the treatment planning process, the de-
pendence on optimisation algorithms and the con-
sequential need to communicate the treatment intentions
to a computer.
A very large variety of formulations of the radiotherapy
optimisation problem with its inherent ambivalence be-
tween therapy success and harmful side-effects has been
conceived over the last 15 years (c.f. [1]). Despite this mul-
titude, all these approaches serve the same purpose of ex-
pressing the clinical treatment intentions in a concise, nu-
merically expedient, unambiguous and intuitive manner.
Hence, it should not come as a surprise that some basic
common traits can be identified.
Optimisation equals evaluation and exploration. There
are two distinct layers of evaluation of a therapeutic dose
distribution. Radiotherapy commonly affects both target
volumes (TV) and normal tissues (NT) alike. The indivi-
dual effects in each NT should be mild or at least in a
justifiable relation to a gain in target effect. Once
quantified, effects may readily be prioritized in decision
making. However, for each NT, a value has to be as-
signed to a non-homogeneous dose distribution that
reflects faithfully the intuitive, implicit rules of clinical ex-
perience. Henceforth, dose evaluation on a per-organ ba-
sis with a focus on biological effectiveness shall be termed
level one (L1), while the search for the right balance be-
tween a set of TV and NT effects shall be termed level two
(L2) evaluation.
Despite the high degree of non-explicit clinical know-
ledge and the difficulty to extract solid data from clinical
studies, L1 dose evaluation is best left to the computer
and taken out of the hands of the clinician. The dose dis-
tribution in a NT volume is a massive amount of
data, even if compacted into a dose volume histogram
(DVH). Further, it is usually impossible to alter one
aspect of a dose distribution without affecting it as a
whole, including all other NT and TV. From this inabi-
lity to control a dose distribution in all its facets, the need
for dose-effect models arises. These models serve both as
the knobs and levers to steer the dose distribution into a
desired direction and as magnifying glasses to raise the
awareness for undesirable features in the optimisation
process.
In contrast, L2 dose evaluation should not, at the time
being, be delegated to software. Especially when treat-
ment intentions are mutually exclusive, like sufficient tar-
get dose and safe organ sparing, this conflict has to be
resolved by the clinician, either in the direction of a safe,
less promising or a more risky, aggressive treatment. The
best support the computer can give in the frequent case of
conflicts like this is to provide for easy navigation of the
solution space [2–4], exploration of alternative solutions
[5,6] or offer navigation aids [7].
103
This paper is devoted to a specific class of L1 dose eva-
luation models, their connection to biological dose-effect
and normal tissue complication probability (NTCP) mo-
dels, and their efficacy both numerically and in controlling
the dose distribution in normal tissues. These models are
termed local dose-effect measures (LDEM). The field of
L2 dose evaluation is of interest only to put the LDEM
dose evaluation in a general context. To start, a ‘‘standard
model’’ of biological dose optimisation is introduced with
the intent to provide a clean canvas for the following de-
velopment. The LDEM class of models is a superset of the
equivalent uniform dose formalism [8]. Its properties are
discussed in the context of NT volume effects and associ-
ated NTCP models. Each specific implementation of an
LDEM effects control over the dose distribution in a parti-
cular manner, which is elucidated for two standard
types of LDEMs in clinical applications. In an appendix
(Table 1), a table of typical model parameter values for a
number of normal tissues is given.
2. Methods
2.1. The standard model of biological optimisation and
local dose-effect measures
For dose optimisation, high- and low-dose volumes in the
same normal tissue or target volume need to be balanced
(L1), and the total effects of various NT and TV need to
be weighed against each other (L2). In other words, there
have to be mathematical functions that assess the meaning
of dose in a subvolume relative to the total volume of one
organ, and other functions that poise the total effect of the
dose distribution in one organ against the others, inclu-
ding the targets.
It has been suggested to express the total effect in a
TV(NT), denoted as F (G), by a sum over all volume ele-
ments belonging to the volume in question [9,10]
1X N
F¼ f ðDi Þ, ð1Þ
N i¼1
1X N
G¼ gðDi Þ, ð2Þ
N i¼1
where Di is the dose to volume element i and f ðDÞ and
gðDÞ are the local effect densities for a TV or NT res-
pectively. By convention, f ; gX0 and the smaller the value
the more preferable. Per this definition, the integral effect
of a dose distribution in an organ is the sum of the inde-
pendent effects in all subvolumes. If the dose in one sub-
volume changes, the effect densities in its neighbourhood are
unaffected. Although this assumption of locality seems
overly restrictive, it stems from a very general approxima-
tion of non-local biological effects (analogous to mean-
field methods in solid state physics [11]), allows for very
 ARTICLE IN PRESS
104 M. Alber / Z. Med. Phys. 18 (2008) 102–110
fast dose evaluation- and optimisation algorithms and
makes the effect density practically independent of the
voxel size of the chosen patient model. The limitations
with respect to long-range, non-local biological effects are
discussed in the final section of this paper.
Having obtained the total effects of n target volumes
and m normal tissues, a particular dose distribution can be
quantified by the vector
ðF 1 ; . . . F n ; G 1 ; . . . ; Gm Þ 2 ½0; 1½nþm .
Clearly, it is not possible to minimize all of these effects
simultaneously or else the problem would have the trivial
solution ðF 1 ; . . . F n ; G1 ; . . . ; G m Þ ¼ 0. Obviously, some
combinations of effects cannot be obtained from physical
dose distributions. This reflects the practical experience
that improving the dose distribution in one volume usu-
ally makes it worse in others.
The physics of dose deposition introduce a boundary in
the solution space that separates feasible from unphysical
dose distributions. The optimum solution for a given pa-
tient, expressed in its particular combination of achievable
F and G values, has to be found on this boundary. Since
greater values of F or G mean worse dose distribution
properties, any dose distribution whose total effect values
lie above the boundary are not optimum, while physical
dose distributions cannot lie below.
In order to arrive at a point on this boundary manifold,
the standard approach is to identify the effect functions F,
G as cost functions of an optimisation problem, and com-
bine them to a Lagrange function L by virtue of
X
n nX
þm
L¼ li F i þ li G in ,
i¼1 i¼nþ1
P ment of some normal tissue by the applied dose, it is clear
with Lagrange multipliers li 2 ½0; 1, nþm i¼1 li ¼ 1. By con-
vention, the optimum dose is obtained by minimizing L.
By varying the values of ðli Þ1::nþm , any possible combina-
tion of total effects can be obtained. Under very general
assumptions and for most practical purposes, there exists
a one-to-one relationship between the n þ m  1 dimen-
sional hyperprism of possible ðli Þ1::nþm vectors and the
entire solution space of the problem. Romeijn [12] has
shown that virtually all so far suggested approaches to L2
evaluation can be reduced to the form of Eq. (4). All spe-
cializations result in a mapping of the multiplier space
onto itself, which may be numerically expedient, but does
not alter the solution space or the optimum solution.
Recently, the terminology of Pareto optimality at-
tracted attention in this context [5,6]. The aforementioned
boundary is identical to the Pareto frontier P and can be
obtained in this framework as a manifold in the n þ m di-
mensional solution space parametrized by ðli Þ1::nþm
PðF ; GÞ ¼ ðF 1 ; . . . F n ; G 1 ; . . . ; G m Þðl1 ; . . . ; lnþm Þ.
Once a solution is found for a particular set of Lagrange
multipliers, these may be altered incrementally to navigate
on the Pareto frontier to other, more preferable solutions
with a different balance of NT and TV effects.
Naturally, there is a tight antagonistic coupling be-
tween NT effects and TV effects. If no NT effect was do-
se limiting to the targets, the solution to the optimisation
problem would be trivial. Hence, it is also possible to ex-
press the best TV effects as a function of the NT effects.
(3) This results in a different parametrization of the Pareto
frontier in terms of the NT effects. In practice, one could
choose li ¼ 1; i ¼ 1; . . . ; n, assuming that all target vol-
ume effects are normalized identically, and then minimize
the sum of all target effects such that Gi pC i ; i ¼ 1 . . . m.
The target effects can be controlled by altering the con-
straint limits C i . This constraint optimisation approach
requires a more sophisticated optimisation algorithm [4],
but offers support for the navigation on the Pareto fron-
tier [7]. A further exploration of this subject of L2 evalua-
tion is beyond the scope of this paper, yet the relationship
between apparently incongruous approaches to L2 op-
timisation may lend some generality to the following de-
velopment, which occurs in the framework of Eq. (4).
2.2. The volume effect and equivalent uniform dose
Although the form of Eqs. (1), (2) can be motivated by a
mean-field approximation of non-local biological effects,
the true benefit of this formalism lies in the transparent
and effective way by which these effect functions offer
control over the shape of the dose-volume histogram of a
(4) NT volume. Starting from the notion that the local effect
densities g reflect the damage wrought in a volume ele-
that g is a continuous, monotonically increasing function
of dose. It is also obvious that g is bounded from above as
the maximum damage in a tissue is limited by its total ob-
literation. So in general, g is of a sigmoidal shape. How-
ever, this upper bound to the NT effect may not become
apparent in clinical dose regions.
The salient feature of g is the representation of the typi-
cal volume effect of a normal tissue dose response, i.e. its
deviation from a step function (no volume effect). The
volume effect details how much tissue volume has to be
spared in order to gain some added tolerance in a high-
dose subvolume. With a small volume effect, the return on
volume sparing is low. The ideal DVH in this case is a
steep drop at the tolerance dose. This kind of behaviour is
usually preferred for organs showing a serial, chain-like
complication mechanism: just as the chain breaks when
one link gives way, the complication occurs if a small
subvolume of the organ is destroyed. In contrast, with a
large volume effect small volumes may receive very high
(5) doses if at the same time large volumes are effectively
 ARTICLE IN PRESS
M. Alber / Z. Med. Phys. 18 (2008) 102–110
spared. The ideal DVH in this case shows a steep drop at
low to intermediate doses, and levels off to a shallow tail
towards high doses. This is often associated with a paral-
lel, rope-like complication mechanism: a fraction of the
strands may tear, yet the rope may still hold.
In mathematical terms, the volume effect answers the
question: if the homogeneous irradiation of a fractional
volume v with a dose D results in a certain level of toxicity,
how does a change of fractional volume dv alter the dose
dD at the same toxicity? This is captured in the function
Qðv; DÞ40 in the following differential equation
dv ¼ Qðv; DÞdD. (6)
It is easy to see, for G ¼ const., that
vg0
Q¼ (7)
g For typical parameters, e.g. k ¼ 12; s ¼ 0:15, the gEUD
in the local effect formalism. By virtue of this relationship,
it is straightforward to construct an efficient cost function
for optimisation from NTCP models. At the core of any
relevant NTCP model lies a representation of the volume
effect, which can be laid bare by Eq. (6). One NTCP mo-
del has recently attracted interest in the context of dose
optimisation. The particular choice of gðDÞ ¼ Dk ; kX1
leads to the generalized equivalent dose model (gEUD)
[8] which is the core of the well established Lyman-Kutc-
her-Burman model [13,14].
Under the above conditions on g, it is always possible to
form
EUD ¼ g1 ðGÞ. (8)
The EUD represents the dose that causes the same effect if
applied homogeneously to the entire organ volume. The
advantage of expressing the effect in terms of a dose is that
the numerical value becomes more meaningful, and can be
related to clinical experience.
It has been shown (c.f [15,16]) that the gEUD - power
law model provides a good approximation to the ob-
servable volume effect of serial complications, where the
volume effect is small. For optimisation algorithm pur-
poses, it provides the additional benefit that it is dose-
scale invariant [17]. However, it appears as if the power
law arises only as an approximation to a more fundamen-
tal dose-response in the limit of low doses/low complica-
tion rates [15]. The serial reconstruction unit (sRU) model
suggests an exponential local effect density g ¼ expðsDÞ,
with s being typically in the range of  0:15 Gy1 .
It may appear as a contradiction to the previous state-
ment about the boundedness of g that both presented
effect densities tend to 1 with dose. This is a consequence
of the fact that for most serial complications, the point of
inflexion of the sigmoid is well above the clinical dose
range so that these functions still provide a faithful ap-
proximation for the convex branch of the sigmoidal dose
response [15]. If clinical doses reach the saturation level of
the effect density, this means that a partial obliteration of
105
the tissue is acceptable. Hence, the clinical endpoint is not
tied to the integrity of the entire organ as in serial
complications, but rather to the integrity of a critical
subvolume of the organ. Requiring the entire sigmoidal
shape of the local effect density is the hallmark of parallel
complication mechanisms. Thus, the full sigmoidal dose
effect needs to be described. For the gEUD model, this
extension to the full range results in g ¼ 1=ð1 þ ðD0 =DÞk Þ
(pEUD), the reconstruction unit model yields g ¼ 1=ð1 þ
expðsðD  D0 ÞÞÞ (pRU) [15]. The reference dose para-
meter D0 marks the point of inflexion.
2.3. Serial complication mechanisms
and sRU models show a steeply increasing, strongly cur-
ved local effect density, and an associated small volume
effect. For example, a dose increase by 10 per cent ne-
cessitates a reduction of irradiated volume by a factor 0:3.
This strongly non-linear behaviour has two consequences.
Firstly, the total effect will always be dominated by the
volume elements receiving the highest doses. This effect is
so strong that volume elements with less than 80 per cent
of the maximum dose in this organ contribute virtually
nothing to the total effect. Secondly, the cost of increasing
the dose to an already hot subvolume rises tremen-
duously. Since by virtue of Eq. (4) all effects are con-
nected, this rise has to be compensated by a drop in
one of the effect functions of other NT, or more likely, a
target volume. This ensures that both the risk is spread
equally among NT, and that cold spots in a TV are tightly
coupled to hot spots in a NT. In consequence, these low-
volume-effect cost functions act strictly dose limiting by
themselves, which makes the adjustment of the associated
Lagrange multiplier less critical. As a corollary, the
power-law dependence of the total effect on basically the
maximum dose may even serve as a guide to adjust the
magnitude of Lagrange multipliers should the total effect
of this NT be too high [11].
This discussion already emphasizes the aspect of con-
trolling the shape of the DVH by the choice of local effect
density. The dose distribution of each NT and TV is partly
a consequence of the chosen local effect density and the
interplay of different NT and TV, mediated by patient
geometry and the physics of dose deposition. Total con-
trol of the dose distribution in one volume usually means
loss of control in all other volumes. Hence, it is important
to realize which aspects of a dose distribution are control-
led by a particular effect density function, and to which
aspects this function is blind.
During dose optimisation, each cost function will
reward a redistribution of dose such that the total effect
is diminished. Hence, it is illustrative to look at the drop of
local effect density per dose level, dg=dD. In Figure 1 the
 ARTICLE IN PRESS
106 M. Alber / Z. Med. Phys. 18 (2008) 102–110
Figure 1. A typical DVH for a serial organ, e.g. rectum. The
salient feature of a cost function for dose optimisation in the lo-
cal dose-effect measure formalism is the way in which it affects
the shape of the DVH. The length of the arrow signifies the con-
trol over each dose bin exacted by the serial Reconstruction Unit
(or gEUD) model. The last arrow is truncated. Arrow lengths
correspond to realistic values for rectum. The cost of increasing
the high dose bins outweighs the cost for the lower dose bins by
far, so that the volume receiving these high doses is controlled
most tightly while the volume receiving low or intermediate
doses is not controlled and hence depends on the circumstances
of patient geometry.
length of the arrows corresponds to the derivative of the
local effect density with respect to dose, signifying the
reduction of total effect if this dose bin in the DVH was
reduced. In other words, the length of arrows shows the
level of control which is exercised by this effect density
over the dose distribution. Increasing the value of k or s
will shift the importance further towards high doses, de-
creasing it will distribute it more evenly. In general, serial-
type cost functions are not very powerful to control the
mid- and low-dose range. In the gEUD formalism, the
lowest value for the exponent k ¼ 1, resulting in the eva-
luation of the mean dose. Here, all dose bins are control-
led with equal power. If the low- and mid-dose regions do
need to be controlled, the complication mechanism is usu-
ally parallel.
2.4. Parallel complication mechanisms
Applying the same reasoning to the full sigmoidal local
effect density for a parallel complication mechanism, it
obtains that the derivative dg=dD is peaked around the
point of inflexion D0 . Figure 2 shows the corresponding
weight control arrows in a typical DVH. It becomes appa-
rent that this effect density type acts like a watershed. As
the gain of reducing the dose in volume elements close to
D0 is greatest, the dose is redistributed towards the extre-
mes, where dose does not mean much harm yet or cannot
cause any more damage than already wrought. In con-
Figure 2. A typical DVH for a parallel organ, e.g. lung. The
length of the arrows signifies the control over each dose bin ex-
acted by the parallel Reconstruction Unit (or parallel gEUD)
model. Arrow lengths correspond to realistic values for lung with
the longest arrow being at 20 Gy. The cost of increasing the high
dose bins is relatively low, reflecting the fact that it does not
cause any more harm to increase the dose above a certain thres-
hold which is assumed to obliterate organ function. Since the
volume receiving at most this threshold dose is most relevant for
conserving tissue function, the cost function assigns the greatest
weight to the dose bins at this threshold.
sequence, this type of effect density acts as a fuzzy dose-
volume constraint for parallel complications. Typical ex-
amples are lung, liver, kidney and parotids, where the or-
gan function can be maintained by a sufficiently large
critical volume while the remaining volume is sacrificed.
Notice, that if g is normalized such that g ¼ 1 as D ! 1,
the total effect corresponds to the integral loss of function
in this organ. It is more meaningful to refer to the total
effect in this case rather than the EUD, as parallel res-
ponding organs are rarely irradiated with a homogeneous
dose so that clinical experience pertains to partial irradia-
tions. The fact that the total effect of such a cost function
can be interpreted as the fraction of the functional volume
that is obliterated highlights the similarity to dose-volume
constraints.
3. Results
These LDEM models may also be defined for targets, e.g.
by means of f ¼ expðaDÞ and can be combined with
physical cost functions such as quadratic penalties, which
fit into the same formalism. The routine IMRT treatments
at University of Tübingen have made use of this for-
malism since September 2001, and to date in excess of
800 patients have been treated with dose distributions op-
timized on these principles. In combination with the use of
hard constraints for normal tissue objectives, this has
resulted in very homogenous dose distributions in terms
of EUD across the patient populations of various case
 ARTICLE IN PRESS
M. Alber / Z. Med. Phys. 18 (2008) 102–110
Figure 3. DVHs for rectum and prostate PTV for 8 randomly
chosen patients. The EUD in the rectum was constrained to 65
Gy. A slight gain in high dose tolerance can be made out for the
patient with the smallest overlap between rectum and PTV be-
cause the rest of the rectum was spared exceptionally well. In
most other patients, the overlap volume is so large that the full
prescription dose cannot be reached and the PTV is not covered
perfectly. The spread of rectum DVHs at the same nominal EUD
(or NTCP) displays the freedom offered by this type of cost func-
tion to adapt the dose distribution to the particular circumstan-
ces of the patient geometry.
classes. The remaining variability of DVHs in a popula-
tion is due to inevitable differences in patient anatomy.
The central benefit of expressing the tissue specific volume
effects by a LDEM is that they offer greater freedom than
dose-volume constraints to adjust the optimum dose dis-
tribution to the specific patient anatomy, but allow higher
doses if and only if the remaining volume can be spared
better than required. At the same time, the dose distribu-
tion is standardized by enforcing a maximum limit on the
EUD, and by stipulation of an explicit volume effect.
Figure 3 shows the DVHs for prostate-PTV and rectum
of eight randomly chosen patients. In this case class, the
dominant conflict arises from the overlap of rectum and
PTV, which can only be irradiated with a sufficient dose if
the overlap volume is small. Hence, an unfavourable ana-
tomy results in either underdosage at the periphery of the
PTV, or an increase in rectum toxicity. Since the rectum
EUD for all patients is constrained to the same number
(65 Gy), all patients should have the same risk of rectum
complication if the volume effect is estimated correctly
(k ¼ 12). It can be observed that the patient with the
smallest overlap volume has the highest dose in the rectum
and the best target coverage. If the overlap volume is lar-
ger, the dose in this volume is reduced.
Owing to the short follow-up and initially slow accrual
into clinical IMRT in Tübingen, the analysis of side effects
is still underway. If the assumptions about the magnitude
of the volume effects are correct, it should not be possible
to identify properties of dose distributions (like cut-off
doses or dose-volume points) that predict for complica-
tions: whenever it was necessary to generate an unfavou-
rable dose distribution in one aspect like maximum dose,
107
this should have been balanced by another aspect, like
V95, such that the overall complication probability stays
constant. Hence, if this compensation was correct, no
particular aspect of the dose distributions in the popula-
tion should correlate with complications.
Moving from an empirically founded and dose-volume
based prescription to EUD constraints is not a difficult
step. Even highly standardized treatments with conformal
radiotherapy introduce a large variability of dose distri-
butions in a patient population. If previously applied dose
distributions are evaluated retrospectively with a LDEM,
a range of EUD values obtains. For example, in a group
of more than 300 prostate patients, the range for the
rectum was 60:1274:8 Gy. Rectal bleeding was clearly
associated with higher EUD values [16]. Even if no out-
come analysis is available, the EUD prescription can be
calibrated to the past clinical practice by defining the past
mean EUD as the limit for future dose optimisations. This
is a definite advantage of EUD-like concepts over NTCP
for dose optimisation: in order to determine the past
NTCP of a clinical protocol, a cumbersome outcome ana-
lysis is required.
In practice, LDEM model parameters can be derived
from published clinical trials, or planning studies. In the
absence of clinical evidence, the most important property
of a set of parameters is that it produces consistent
results in a patient population. Some parameter sets for
a number of organs/complications are given in the appen-
dix (Table 1).
4. Discussion and conclusion
Numerical optimisation relies on models for the evaluati-
on of the products of the process that epitomize com-
prehensively and faithfully the intended quality criteria
required of the optimum solution. In the context of ra-
diotherapy optimisation this means that dose evaluation
models do not have to be any more founded in biology
than the rules for the evaluation of conventional treat-
ment plans, which originate frequently from tradition
rather than solid data. However, if the essence of clinical
experience and data was distilled into some dose evalua-
tion model, this could liberate treatment planning from
the limitations of traditional technology, and bring mo-
dern technology like IMRT and IMPT to full fruition.
There is a risk associated with this chance: that the dose
distributions obtained from optimisation with such mo-
dels strain the range of applicability of these models be-
yond its limits. After all, the capacity to extrapolate does
not follow from the proven ability to interpolate. Hence, it
is essential that users understand the properties of dose
evaluation models, in particular for normal tissues, and
especially which traits of a dose distribution are not con-
trolled by the optimisation model.
 ARTICLE IN PRESS
108 M. Alber / Z. Med. Phys. 18 (2008) 102–110
The proposed LDEM is primarily geared towards this
transparency and numerical expediency. At the same time,
it is possible to express a wide range of normal tissue vol-
ume effects with these models. For dose optimisation, the
normal tissue volume effect is arguably the most impor-
tant property of any organ at risk, and is of paramount
importance if dose escalation is to be performed safely.
The LDEM compute a single number that represents the
total effect of a dose distribution in the volume to which
the model is applied. This number is a ‘‘biological’’ quan-
tity in that it considers the volume effect. It is possible to
map the total effect onto a normal tissue complication
probability with a one-to-one relation, so that any LDEM
can be incorporated in a NTCP model. Vice versa, almost
all NTCP models can be expressed in terms of a LDEM. It
follows that even approaches like the optimisation of the
probability of complication-free tumour control Pþ are
only a special case of Eq. (4).
The fundamental approximation and limitation of the
LDEM is the reliance on local effect densities, which can
constitute a problem if non-local biological effects played
a role. Evidence of such effects can be found in animal
experiments with ultra-small fields and very high doses
[18]. The reconstruction unit models introduced above as
LDEMs have been extended to take into account non-lo-
cal tissue repair effects [15]. In the context of radiotherapy
with photons, electrons and protons, the penumbra of the
dose distributions is usually so shallow that non-local re-
pair effects do not play a role and do not constitute a li-
mitation to the LDEM approach. Heavy ions may indeed
be an exception. In these cases, the LDEM would ove-
restimate the damage to normal tissue, and consequenti-
ally err to the safe side.
A practical limitation of LDEM is the undeniable spar-
city of clinical data to pinpoint the parameters required by
the model. For some organs, like rectum and lung, good
parameters sets exists. For others, where clinical compli-
cations are sparse, parameters have to be tried and tested
in planning studies and clinical routine. The parameters in
the appendix have been distilled from 6 years of clinical
application of the LDEM in Tübingen, but are by no
means final (Table 1). When in doubt, it is always prefer-
able to err to the safe side, which in the case of serial com-
plications means to assume a smaller volume effect, i.e.
greater k or s. The range of clinically applied total effects
can be assessed by computing the EUD for a set of already
applied dose distributions. In the end, nothing can replace
good judgement, for which it is helpful to know which
parts of the dose distribution are controlled by the
LDEM.
The cost for the detailed consideration of the volume
effect is that LDEM are non-linear functions which re-
quire more sophisticated optimisation algorithms than
e.g. linear cost functions. Of particular importance is the
existence of locally optimum solutions. In general, it can
be shown that if the local effect density is convex (curva-
ture like an upturned parabola), local minima cannot
occur. If the local effect density has a sigmoidal shape,
local minima cannot be excluded, but this is no guarantee
that they do occur. Their occurrence is frequently preven-
ted by the physics of dose deposition which does not per-
mit to change the dose distribution in isolated subvolumes
only. Further, the smaller the curvature of the local effect
density, the less likely is the existence of local minima.
In consequence, the pEUD/pRU models behave better
numerically than exact dose-volume constraints, and
allow some more flexibility in finding the optimum dose.
It is safe to assume that for all practical purposes, the
Lagrange function is convex so that the statements made
about the existence of a manifold of optimum solutions
as the boundary between physical and unphysical dose
distributions hold true.
The LDEM formalism encorporates biological model
assumptions, which will eventually be verified by clinical
studies. Its primary purpose is to provide a toolbox for
steering the dose optimisation. As with all tools, their
sheer existence does not prove their utility. However, their
design was guided by the intent to put dose optimisation
on a formal basis and to make control over the optimisa-
tion process more transparent, which are arguably worth-
while goals.
Acknowledgements
The author was awarded the Research Prize of the Ger-
man Association of Medical Physicists (DGMP) in 2006
for his research in the field of radiotherapy dose opti-
misation.
 ARTICLE IN PRESS

M. Alber / Z. Med. Phys. 18 (2008) 102–110 109

Table 1
Effect density types and parameters for selected organs/complications. Parameters are partially obtained from clinical data, partially
from planning studies. EUD and critical volume constraints should be lower than given values if possible.

Organ:Complication Type k s D0 EUD critVol

Rectum:Bleeding gEUD 12 N/A N/A 65 N/A

Rectum:Bleeding sRU N/A 0.13 N/A 65 N/A
Bladder gEUD 8 N/A N/A 56 N/A
Liver pEUD 4.5 N/A 30 N/A o40%
Liver pRU N/A 0.15 30 N/A 40%
Lung:Pneumonitis1 pEUD 3 N/A 20 N/A o60%
Lung:Pneumonitis pRU N/A 0.15 20 N/A o60%
Kidney1 pEUD 2.5 N/A 16 N/A o50%
Small bowel gEUD 4.0 N/A N/A 40 N/A
Heart gEUD 5.5 N/A N/A 32 N/A
Esophagus gEUD 8.0 N/A N/A 50 N/A
Spinal cord gEUD 12.0 N/A N/A 45 N/A
Brain Stem gEUD 14.0 N/A N/A 45 N/A
Optical Nerves gEUD 14.0 N/A N/A 48 N/A
Optical Chiasm gEUD 14.0 N/A N/A 48 N/A
Eyes gEUD 4.0 N/A N/A 30 N/A
Parotids gEUD 1.0 N/A N/A 30 N/A
Parotids pEUD 3 N/A 26 N/A o50%
Brain gEUD 6 N/A N/A 24 N/A

1: per side. Combined effect of both sides should be lower!

[4] Alber M, Reemtsen R. Intensity modulated radiation therapy

Appendix: LDEM parameters for selected planning by of a barrier-penalty multiplier method. Optimization,
organs Methods and Software (OMS) 2007;22(3):391–411.
[5] Thieke C, Bortfeld T, Niemierko A, Nill S. From physical dose con-
straints to equivalent uniform dose constraints in inverse radio-
List of mathematical symbols therapy planning. Med Phys 2003;30(9):2332–9.
[6] Craft D, Halabi T, Bortfeld T. Exploration of tradeoffs in
Di dose in volume element i intensity-modulated radiotherapy. Phys Med Biol 2005;50(24):
f ðDi Þ effect density for targets 5857–68.
gðDi Þ effect density for normal tissues [7] Alber M, Birkner M, Nüsslin F. Tools for the analysis of dose
F; G cost functions for targets (normal tissues) optimisation II: Sensitivity analysis. Phys Med Biol 2002;47N265–70.
[8] Niemierko A. A generalized concept of equivalent uniform dose
lj Lagrange multipliers (EUD) (abstract). Med Phys 1999;26:1100.
L Lagrange function [9] Raphael C. Mathematical modelling of objectives in radiation the-
v partial (organ/target) volume rapy treatment planning. Phys Med Biol 1992;37:1297–311.
k volume effect parameter in gEUD model, equals [10] Alber M, Nüsslin F. An objective function for radiation treatment
optimization based on local biological measures. Phys Med Biol
1=n of Lyman-Kutcher-Burman model 1999;44(2):479–93.
s radiation sensitivity of a reconstruction unit [11] Alber, M. A concept for the optimization of radiotherapy. PhD
D0 reference dose, e.g. TD50 thesis, Universität Tübingen, 2000. hhttp://w210.ub.uni-tuebingen.
de/dbt/volltexte/2001/221i.
[12] Romeijn HE, Dempsey JF, Li JG. A unifying framework for multi-
criteria fluence map optimization models. Phys Med Biol
2004;49(10):1991–2013.
[13] Kutcher GJ, Burman C, Brewster L, Goitein M, Mohan R. Histo-
References gram reduction method for calculating complications probabilities
for three-dimensional treatment planning evaluations. Int J Rad
[1] Webb S. Intensity-Modulated Radiation Therapy. Medical Science Oncol Biol Phys 1991;21:137–46.
Series. Bristol: IOP Publishing; 2000. [14] Emami B, Lyman J, Brown A, Coia L, Goitein M, Munzenrieder
[2] Jee K-W, McShan DL, Fraass BA. Lexicographic ordering: intui- JE, et al. Tolerance of normal tissue to therapeutic irradiation. Int J
tive multicriteria optimization for IMRT. Phys Med Biol 2007; Rad Oncol Biol Phys 1991;21:109–22.
52(7):1845–61. [15] Alber M, Belka C. A normal tissue dose response model of dynamic
[3] Wilkens JJ, Alaly JR, Zakarian K, Thorstad WL, Deasy JO. IMRT repair processes. Phys Med Biol 2006;51(1):153–72.
treatment planning based on prioritizing prescription goals. Phys [16] Söhn M, Meldolesi E, Vargas C, Yan D, Alber M. The inci-
Med Biol 2007;52(6):1675–92. dence of late rectal bleeding in high-dose conformal radiotherapy
 ARTICLE IN PRESS

110 M. Alber / Z. Med. Phys. 18 (2008) 102–110

of prostate cancer using EUD- and dose-volume based NTCP mo- [18] Bijl HP, van Luijk P, Coppes RP, Schippers JM, Konings AWT,
dels. Int J Radiat Oncol Biol Phys 2007;67(4):1066–73. van der Kogel AJ. Unexpected changes of rat cervical spinal cord
[17] Alber M, Nüsslin F. A representation of a NTCP function for local tolerance caused by inhomogeneous dose distributions. Int J Radiat
complication mechanisms. Phys Med Biol 2001;46:439–47. Oncol Biol Phys 2003;57(1):274–81.