The British Journal of Radiology, 72 (1999), 781±786 E 1999 The British Institute of Radiology

Relative electron density calibration of CT scanners for

radiotherapy treatment planning
S J THOMAS, MA, MSc
Medical Physics Department, Addenbrooke's Hospital, Hills Road, Cambridge CB2 2QQ, UK

Abstract. Several authors have reported data on the variation of Houns®eld numbers with
electron density in CT scanners. The data can be ®tted with a double straight line approach. For
non-bone tissues (or phantom materials with similar atomic numbers) the data from all authors
can be ®tted to a single straight line. For bone-like materials the line varies between authors. The
method used to measure electron density has a greater effect than the differences between
scanners, or the kilovoltage used on a given scanner. The effect of variation of these slopes on the
accuracy of radiotherapy treatment planning is analysed. For typical radiotherapy beams, to
produce a 1% error in dosimetry would require errors of over 8% in bone electron density. Using
a single pair of calibration lines for all the scanners reported would give dosimetric errors of
under 0.8%. A formula is recommended as a default for use in planning systems in circumstances
where no data are available for a particular scanner.

CT is routinely used in radiotherapy treatment
planning to determine both clinical and physical
information. The principal physical information
comprises the size, shape and location of outlines
and inhomogeneities and, via the Houns®eld
Units (HU), the relative electron density of
tissues.
HU are de®ned as:
k{k 
w
HU~1000 …1†
kw
where m is the attenuation coef®cient, and mw is
the attenuation coef®cient of water. Attenuation
coef®cients depend on electron density, atomic
number (Z) and on the quality of the beam used
in the CT scanner.
The relationship between relative electron
density, re, and the CT number in HU has been
measured by several authors [1±5]. For materials
with an average atomic number similar to that of
water, the HU lie on or near a straight line that
passes through HU521000 for air and HU50
for water. Bone substitute materials give HU
values that lie above this line, with different
results being observed for different scanners.
These differences lead to the standard recommen-
dation that the relationship between re and HU
be measured for each scanner used for treatment
planning. However, in many circumstances, large
uncertainties in relative electron density give only
small uncertainties in dosimetry.
The aim of this work is to analyse the
Received 1 December 1998 and in revised form 3 March
1999, accepted 8 April 1999.
dosimetric errors that will result from uncertain-
ties in the HU calibration, to quantify the amount
of variation between scanners that is acceptable
for treatment planning purposes. This will be
illustrated by an analysis of published data for a
number of scanners, together with data that have
been measured on four machines used to provide
CT input to treatment planning at Addenbrooke's
Hospital.
Method
In radiotherapy treatment planning, the relative
electron density at a point in the patient is used as
the input into a calculation of the dose. Treatment
planning systems use a large variety of algorithms
to correct for heterogeneities, ranging from one-
dimensional methods such as the effective depth
(ratio of tissue maximum ratios (TMR)) algo-
rithm and the power law algorithm, to methods
based on separation of primary and scatter, kernel
convolution or Monte Carlo methods.
The following analysis is based on the ratio of
TMR method. A water-equivalent effective depth
is calculated, based on the product of depth and
relative electron density. The ratio of (TMR at
effective depth) to (TMR at true depth) is used to
calculate a correction factor CF, de®ned as the
ratio of the (dose corrected for heterogeneities) to
the (dose at same place if all material were water).
As will be shown in the discussion, the error in
dosimetry produced for a given error in relative
electron density is similar for more complex
methods [6, 7].
Table 1 gives a typical HU, typical thick-
ness and typical relative electron density for

The British Journal of Radiology, August 1999 781


Table 1. Data for typical inhomogeneities encountered in treatment planning. The ®nal column shows the change
in water equivalent thickness caused by the inhomogeneity
Structure Typical HU Typical
thickness (t)
Lung 2740 8 cm
Fat 250 4 cm
Liver +50 8 cm
Rib +600 1 cm
Humerus +700 2.5 cm
Cranium +900 1.5 cm
inhomogeneities found in the human body. The
relative electron densities shown in this table are
those published by Duck [8]. The ®nal column in
Table 1 shows the change in water equivalent
thickness caused by an inhomogeneity of this
thickness and relative electron density.
Data from BJR Supplement 25 [9] were used to
calculate the error in the water equivalent depths
that would result in a 1% error in dose for a
10 cm610 cm ®eld, and depths of 5 cm and
10 cm, for a range of photon energies. These
results were combined with the ®nal column of
Table 1, to give values of the error in relative
electron density that would produce 1% error in
dose calculation at a clinically relevant depth
(10 cm for all except cranium, where 5 cm has
been used).
Measurement of relationship between CT
number and relative electron density
With one exception, all measurements included
in this analysis were obtained by imaging
phantoms containing various tissue substitute
materials, and tabulating the CT number obtained
against relative electron density.
For low Z materials (everything except bone
substitute phantoms), similar materials were used
by all authors. Air and water were universally
used. Other materials included plastics (PMMA,
polystyrene), wood, organic liquids and resin-
based tissue substitutes. As will be seen in
the results section, the choice of materials
did not lead to systematic differences between
results.
For bone substitute materials, the phantom
materials fall into two distinct sets. Several
authors used aqueous solutions of CaCl2.
Others used resin mix, to which CaCO3 or
CaHPO4?2H2O had been added. As will be seen
in the results section, the choice of material did
lead to systematic differences between authors
results.
The results of Schneider et al [5] were calculated
from the known chemical composition of various
782
S J Thomas
Relative electron Product of t and
density (re) (re21)
0.26 25.92
0.95 20.20
1.05 0.40
1.34 0.34
1.39 0.98
1.51 0.77
tissues, using a stochiometric calibration tech-
nique. Table 2 summarises the sources of data,
listing in each case the scanner, tube potential
(kV) and the bone substitute phantom material
used.
Results
Table 3 shows the errors in relative electron
density needed to give ¡1% error in dose.
Greater accuracy is required for thick structures
with electron densities near to that of water (e.g.
liver) than for thinner structures of high density
(e.g. cranium). The higher the energy of X-rays
used for radiotherapy, the lower the accuracy
required in the values of relative electron density.
The values for 6 MV from Table 3 have been
used in Figures 1 and 2.
Figure 1 shows the relationship between CT
number and relative electron density for low Z
materials (i.e. all tissues and phantom materials
other than bone or bone substitutes). All the data
fall on or near a single straight line:
HU
oe ~ z1:00 …2†
1000
This result is not unexpected; for materials of a
similar atomic number the attenuation coef®cient
is proportional to electron density, hence this
result will follow from Equation (1). The bars
shown on the graph at HU of 2740, 250 and +50
show the range of relative electron densities
corresponding to ¡1% error in 6 MV dosimetry
for lung, fat and liver. In all cases the points lie
within the range.
Figure 2 shows the relationship between HU
and relative electron density for bone substitute
materials. The ®lled symbols are all for measure-
ments with aqueous solutions of CaCl2, the
hollow symbols are all for resin-based solid
phantoms. The crosses are for a stochiometric
calibration. In contrast to the low Z situation,
there is considerable difference between data sets
for different machines and for different measure-
ment techniques.
The British Journal of Radiology, August 1999
Electron density calibration of CT scanners for treatment planning

Table 2. Sources of the data used in Figures 1 and 2, giving in each case the scanner, kV, bone substitute phan-
tom material used and symbols used in Figures 1 and 2

Source of data Scanner kV Bone substitute Symbol

[1] EMI CT-5005 Aqueous CaCl2 X

[2] GE 9800 140 Aqueous CaCl2 {
[3] Siemens Somatom 2 125 CB4+CaHPO4 s
[4] Siemens DR3 125 CB2+CaCO3 or SB3 n
[5] GE9000 120 SR1, SR4 ,
[5] GE9000 120 Stochiometric calculations 6
J
N
This paper GE high-speed adv. 120 Aqueous CaCl2
This paper Siemens Som. Plus 120 Aqueous CaCl2
This paper Siemens Som. Plus 140 Aqueous CaCl2 m
This paper Nucletron Simulix 120 Aqueous CaCl2 .

The dashed line shown on Figure 2 is for a
straight line with the equation of:
HU
oe ~ z1:00 …3†
1950
The bars shown on the graph at HU of +700 and
+900 show the range in relative electron densities
corresponding to ¡1% error in 6 MV dosimetry
for humerus and cranium. Notwithstanding the
considerable differences between the data sets,
they all lie within these bars.
Discussion
With the widespread introduction of CT, a
radiotherapy centre frequently ®nds itself using
CT data from a wide range of scanners, spread
over the geographical area from which referrals to
the centre are made. With the implementation of
cross-manufacturer standards such as DICOM, it
is increasingly feasible to make use of this data for
radiotherapy treatment planning. Whilst it is
clearly desirable for full electron density calibra-
tion curves to be available for every scanner used
for treatment planning, this is not always
practicable. Faced with a choice between using
this CT data, or planning using a non-CT method
(manual outlines and no inhomogeneity data), it
is better to have a means of using the CT data.
Even when a centre uses only a few CT
scanners, and physics staff have ample access to
these scanners for the imaging of phantoms,
uncertainties remain. The data in Figure 2 show
greater variation by phantom material than by
manufacturer or by kV. For relative electron
densities between 1.1 and 1.3, the highest CT
numbers occur for phantoms consisting of aque-
ous solutions of calcium chloride solution. Since
these are solutions in water, they all tend back to
an HU of 0.0 at a relative electron density of 1.0,
but then follow slopes with dHU/dre52370¡260.
The slopes (dHU/dre) for solid phantoms are
similar to those for CaCl2 solutions. Solid
phantoms tend to be based on a high density
resin; the resin with no added calcium always lies
on or near the straight line of Figure 1, typically
with an HU of 100 at a relative electron density of
1.1. Since the CaCl2 solutions typically have an

Table 3. Percentage change in tissue maximum ratio (TMR) per cm of depth and the change in depth required to
produce a 1% change in TMR. Data are taken from BJR Supplement 25 [9]. These are used to calculate the
range of relative electron densities that would give a dose correction factor within ¡1% of those calculated for
the typical inhomogeneities in Table 1

Cobalt 6 MV 10 MV 21 MV

% Change in TMR per cm of depth

5 cm 4.0 3.0 2.4 0.7
10 cm 5.4 3.6 3.0 2.3
Change in depth to produce 1% change in TMR
5 cm 2.5 mm 3.3 mm 4.2 mm 14.3 mm
10 cm 1.9 mm 2.8 mm 3.3 mm 4.3 mm
Range of relative electron density for ¡1% in dose
Lung 0.24±0.28 0.23±0.30 0.22±0.30 0.21±0.31
Fat 0.90±1.00 0.88±1.02 0.87±1.03 0.84±1.06
Liver 1.03±1.07 1.02±1.08 1.01±1.09 1.00±1.10
Humerus 1.31±1.47 1.28±1.50 1.26±1.52 1.22±1.56
Cranium 1.34±1.68 1.23±1.79 1.23±1.79 0.56±2.46

The British Journal of Radiology, August 1999 783


HU of 237 at the same electron density, it follows
that, for all relative electron densities over 1.1, a
solid phantom will systematically give a lower HU
than a water based phantom.
Results from stochiometric calibrations tend to
lie in between these two sets. However, perform-
ing a full stochiometric calibration of a large
number of CT scanners is beyond the resources of
most radiotherapy departments. It is not the
purpose of this paper to pass an opinion on which
phantom material gives the most clinically
applicable densities; this analysis presumes only
that the truth lies somewhere between the
extremes of the measured data.
In the region for HU,100, the bars on
Figure 1 show the accuracy required to achieve
dosimetric accuracy ,1%. The requirements for
reproducibility of CT number are also important
for diagnostic use. For this reason, most radi-
ology departments will have a system of quality
assurance in place to ensure at the very least that
air gives 21000 HU, and water gives 0 HU. As
shown by Figure 1, there is no great difference
between manufacturers. In this region, all
784
S J Thomas
Figure 1. The relationship between
Houns®eld Units (HU) and relative
electron density for material of low
atomic number. See Table 2 for the
key to the symbols. The horizontal
error bars show the uncertainty in
electron density required to give 1%
error in dosimetry for a 6 MV
beam passing through typical in-
homogeneities of lung, fat and
liver.
scanners give essentially the same line, showing
that the manufacturers all meet the required
standards; routine quality assurance (QA) for
diagnostic purposes should ensure that scanners
meet the requirements for treatment planning.
In the region for HU.100, there is considerable
variation both between machines and between
calibration techniques. However, in this region
quite large errors in electron density calibration
are required before a signi®cant error in dosim-
etry occurs. The higher the energy of radiation
used for therapy the greater the permissible
uncertainty in electron density. If the dashed
line in Figure 2 were used for all machines, the
greatest error in dosimetry that would occur at
6 MV would be 0.8%. At 21 MV this would
reduce to 0.5%. Even at 60Co the error in
dosimetry would only be 1.3%.
To illustrate that conclusions drawn from an
analysis based on effective depth (TMR ratio) are
applicable to other methods of inhomogeneity
calculation, the effect of changing density is
calculated for a lung and a humerus by the
effective depth method, a modi®ed power-law
Figure 2. The relationship between
Houns®eld Units (HU) and relative
electron density for bone substitute
materials. See Table 2 for the key
to the symbols. The solid symbols
are all for aqueous solutions of
CaCl2, the hollow symbols for
resin-based solid phantoms. The
horizontal error bars show the
uncertainty in electron density
required to give 1% error in
dosimetry for a 6 MV beam
passing through typical bone
inhomogeneities.
The British Journal of Radiology, August 1999
Electron density calibration of CT scanners for treatment planning
Table 4. A comparison between the effective depth, modi®ed power law and ETAR methods of calculating the
correction factor for a 6 MV beam of X-rays passing through an inhomogeneity. Data were calculated using
beam data from a Varian 600C linear accelerator
rew CF value
Eff. depth Power law ETAR
Lung
0.234 1.221 1.213 1.203
0.260 1.214 1.205 1.196
0.286 1.207 1.197 1.189
Humerus
1.251 0.977 0.980 0.982
1.390 0.965 0.969 0.973
1.529 0.952 0.958 0.963
method [6] and the equivalent tissue±air ratio
method (ETAR) [7]. Although the correction
factors produced by the three methods are not
identical, the dosimetric changes produced by
small changes in density are of the same order.
Calculations are given for the lung and
humerus situations considered above, for 6 MV
X-rays. In each case, the correction factor (dose
with inhomogeneity/dose in water) was calculated
10 cm deep for the density given in Table 1, and
for densities 10% higher and lower. For the lung
case the region was assumed to extend from a
depth of 1 cm to a depth of 9 cm. For the
humerus it was assumed to extend from a depth
of 4 cm to a depth of 6.5 cm. In each case a
10 cm610 cm beam at 100 cm source-to-surface
distance was assumed.
Table 4 shows the correction factors calculated
in each case, both as raw numbers and as
percentage changes in dose caused by a 10%
change in electron density. Although different
algorithms give different CF values, the change in
CF value caused by a 10% change in electron
density is very similar. This shows that the
analysis in this paper, while based on effective
depth, is equally applicable to other methods of
calculating the dose. The only signi®cant differ-
ence is seen when going to a very high density on
humerus, where the effective depth algorithm
overestimates by 0.4% the effect of increasing
density by 10%. This strengthens the arguments of
the paper that large errors in electron density are
required before measurable errors in dosimetry
are observed.
Conclusions
The same conversion between CT numbers and
relative electron density can be used for all
scanners reported in the literature. For
HU,100, Equation (2) should be used; for
HU¢100, Equation (3) should be used.
This gives a discontinuity at HU5100. This is
The British Journal of Radiology, August 1999
% change in CF from 10% change in re
Eff. depth Power law ETAR
0.6% 0.7% 0.6%
0.6% 0.7% 0.6%
1.2% 1.1% 1.0%
1.4% 1.1% 1.0%
not clinically signi®cant, as the highest density for
low Z tissues lies under 100 HU (e.g. skin at
HU578) whilst the lowest density bone is well
above this value (e.g. spongiosa at HU5260).
These equations should not be used for materials
with Z values signi®cantly different from those
occurring naturally in the human body. For
example, they should not be applied to scans
through metal hip prostheses.
In the region where Equation (2) holds, the
equation agrees with experimental data within
measurement uncertainties for all CT scanners
reported in the literature.
In the region where Equation (3) is recom-
mended, its use will not cause dosimetric errors in
excess of 1%. Using an accurate conversion curve
for an individual scanner will further reduce these
uncertainties, but care should be taken in using
measurements with simple bone substitute phan-
toms; the difference from Equation (3) is more a
function of the phantom material chosen than it is
of the scanner.
Acknowledgments
I am grateful to Tim Fryer, Stuart Yates and
Tony Geater for performing relative electron
density calibration measurements on the
Somatom, Simulix and GE scanners.
References
1. Parker RP, Hobday PA, Cassell KJ. The direct use
of CT numbers in radiotherapy dosage calculations
for inhomogeneous media. Phys Med Biol
1979;24:802±9.
2. McCullough EC, Holmes TW. Acceptance testing
computerized radiation therapy treatment planning
systems: direct utilization of CT data. Med Phys
1985;12:237±42.
3. KnoÈoÈs T, Nilsson M, Ahlgren L. A method of
conversion of Houns®eld numbers to electron density
and prediction of macroscopic pair production cross
sections. Radiother Oncol 1986;5:337±45.
785

4. Constantinou C, Harrington JC, DeWerd LA. An
electron density calibration phantom for CT-based
treatment planning computers. Med Phys 1992;
19:325±8.
5. Schneider U, Pedroni E, Lomax A. The calibration
of CT Houns®eld units for radiotherapy treatment
planning. Phys Med Biol 1996;41:111±24.
6. Thomas SJ. A modi®ed power-law formula for
inhomogeneity corrections in beams of high-energy
X rays. Med Phys 1991;18:719±23.
Book review
Computed Body Tomography with MRI
Correlation (3rd edn and CD-ROM). Ed. by J
K T Lee, S S Sagel, R J Stanley and J P Heiken,
1998 (Lippincott, Philadelphia), £165.00
ISBN 0781716691
This CD-ROM is the 3rd edition of the well-
known textbook of CT and MRI by Lee et al.
The CD-ROM requires 8 MB of RAM and 11 MB
of free hard disk space and will operate on
Windows systems as well as Apple Macintosh.
The CD-ROM contains the text, images and
references of the two volume CT/MRI textbook.
The whole body is covered systematically in
chapters. The text is comprehensive and is
accompanied by referenced CT/MRI illustrations
down the side which can be magni®ed when
786
S J Thomas
7. Sontag MR, Cunningham JR. The equivalent tissue±
air ratio method for making absorbed dose calcula-
tions in a heterogeneous medium. Radiology 1978;
129:787±94.
8. Duck FA. Physical properties of tissue. London:
Academic Press, 1990.
9. BIR/IPEMB Working Party. Central axis depth dose
data for use in radiotherapy: 1996, BJR Supplement
25. London: British Institute of Radiology, 1996.
viewing. The images are annotated and clearly
marked.
In the text, the references are numbered and
can be accessed immediately by clicking on the
numbers. In addition, there are facilities to print/
copy sections of the text, tables and images, which
should prove valuable as a teaching tool.
I have perused through several sections of this
CD-ROM. The search facility is particularly
helpful, especially when looking up a dif®cult
diagnostic problem. All aspects of CT/MRI of the
body are covered in detail making this a useful
reference of cross-sectional imaging CD-ROM for
personal and department libraries.
A K BANERJEE
The British Journal of Radiology, August 1999