The British Journal of Radiology, 73 (2000), 11781184 E 2000 The British Institute of Radiology

CT assessment of tumour response to treatment:

comparison of linear, cross-sectional and volumetric
measures of tumour size
1
S A SOHAIB, MRCP, FRCR, 1B TURNER, MSc, 1J A HANSON, 3M FARQUHARSON, PhD,
2
R T D OLIVER, FRCP and 1R H REZNEK, FRCP, FRCR

Departments of 1Diagnostic Imaging and 2Oncology, St Bartholomew's Hospital, West Smitheld, London
EC1A 7BE and 3Department of Radiography, City University, Charterhouse Square, London EC1, UK

Abstract. Changes in cross-sectional area are currently used to assess tumour response to
treatment. The aims of this study were to validate a helical CT technique for volume
determination using a series of phantoms and to compare tumour responses indicated by one-,
two- and three-dimensional measures of tumour size change in patients treated for germ cell
cancer or lymphoma. All studies were performed on an IGE HiSpeed Advantage helical CT
scanner with an Advantage Windows workstation. Phantom volumes were calculated using
volume reconstruction software and compared with reference volumes determined by water
displacement. 20 lymph node masses were studied on serial CT scans in 16 patients treated with
chemotherapy for germ cell cancer or lymphoma. For each lesion the maximum diameter,
maximum cross-sectional area and volume were determined before and after treatment. Tumour
response was assessed using the standard World Health Organisation criteria (i.e. changes in
cross-sectional area) and the newly proposed unidimensional response evaluation criteria in solid
tumour (RECIST). The CT volume measurement error was 1.05.1% for regularly shaped
phantoms larger than 35 cm3. In the assessment of treatment response there was 90% agreement
between one-dimensional (1D) and two-dimensional (2D) measurements and 100% agreement
between 2D and three-dimensional (3D) measurements. CT volume measurements are accurate
and reproducible, particularly for larger structures. Assessment of tumour response using 1D, 2D
and 3D measures had limited inuence on the classication of treatment response. However, the
impact of CT assessment of tumour response using 1D, 2D and 3D measurements on clinical
decisions and patient outcome remains to be determined.

The standard method for assessing the response
of a tumour to treatment is to determine its
change in maximum cross-sectional area [1, 2].
Using World Health Organisation (WHO) cri-
teria, reductions in cross-sectional area of at least
50% and of 100% (i.e. no measurable disease)
dene partial and complete responses, respectively
[2]. However, two-dimensional (2D) measure-
ments may misrepresent change in tumour size
by disregarding alteration to the third dimension.
More accurate determination of tumour size
change may be obtained using three-dimensional
(3D) volume measurements, with consequent
Received 2 November 1999 and in revised form 6 June
2000, accepted 27 June 2000.
Address correspondence to Dr S A Sohaib, Academic
Department of Radiology, Dominion House, 59
Bartholomew's Close, St Bartholomew's Hospital,
West Smitheld, London EC1A 7BE, UK.
S A Sohaib was supported by the Joint Research
Board, St Bartholomew's Hospital, London.
implications for assessment of response. Until
recently, tumour volume measurements were not
easily achieved. With the advent of helical CT,
volumetric data acquisition within a single breath-
hold can be now obtained. Tumour volume
measurements have become easier using new
software, and in the future may be automated.
Volume measurement techniques have been
reported for both conventional (non-helical) and
helical CT [35].
Despite the fact that volume measurement
potentially offers more accurate representation
of change in size of a tumour following treatment,
many cancer research organizations are now
proposing response evaluation criteria in solid
tumour (RECIST) guidelines based on unidimen-
sional (1D) measurements, i.e. change in the
maximum diameter of the tumour (see Table 1)
[6, 7]. The aims of this study were to validate a
helical CT technique for volume measurement
using a series of phantoms and to compare
tumour responses determined by 1D, 2D and

1178 The British Journal of Radiology, November 2000

CT assessment of tumour response to treatment
Table 1. Criteria for tumour response
WHO [2]
Complete response (CR) Complete disappearance of all known disease
Partial response (PR) At least 50% reduction in tumour sizea
No change (NC) Neither PR nor PD
Progressive disease (PD) Greater than 25% increase in sizea of at least
one lesion (or new lesion)
a
Quantied as the product of the longest diameter and the greatest perpendicular diameter.
b
Quantied as the longest diameter.
3D measurements of tumour size in patients
treated for germ cell cancer or lymphoma.
Materials and methods
All scans were performed on an IGE Hispeed
Advantage scanner (GE Medical Systems,
Milwaukee, WI). Image analysis was performed
using Advantage Windows version 1.2 software
on an IGE/Sun Microsystems computer work-
station (Sun Microsystems Inc, California, CA).
Phantom study
Ten ovoid shapes with smooth, regular outlines
ranging from approximately 1 cm to 10 cm in
diameter were formed from PlasticineH, an easily
deformable, non-compressible and non-absorbent
material with an approximate CT attenuation
value of 1250 Hounseld units (HU). The true
volume of each PlasticineH phantom was deter-
mined by a water displacement method.
To determine the CT volume, each phantom
was scanned in a vessel containing dilute contrast
media (CT attenuation approximately 1200 HU).
The dilute contrast media was used to reduce the
density difference between the phantom and its
surrounding in order to simulate the in vivo
environment [3]. Scans were performed at 1 cm
collimation, pitch 1.0:1 with axial image recon-
struction at 1 cm intervals using a soft tissue
algorithm.
3D reconstruction and volume determinations
were performed at the workstation by tracing out
each axial image with a mouse-operated cursor.
The software then generated a 3D model
and directly calculated the phantom volume
(Figure 1). This interactive process was repeated
three times and the mean phantom CT volume
was obtained. The coefcient of variation for the
CT volume measurements of each phantom was
calculated. The coefcient of variation was used
to compare the reproducibility of measurements.
Each phantom was reformed into an elongated,
irregularly outlined shape and the process of CT
volume determination was repeated.
The CT volume measurement error was
The British Journal of Radiology, November 2000
RECIST [6]
Complete resolution of all target lesions
At least 30% reduction in tumour sizeb
Neither PR nor PD
Greater than 20% increase in sizeb
calculated as follows [3]:
CT volume measurement error (%)~
(CT volume true volume)|100
true volume
Patient study
16 patients (15 males, 1 female; mean age 42
years, range 2464 years) with germ cell tumours
or lymphoma who had undergone abdominal
pelvic helical CT before and after chemotherapy
were studied. 20 measurable lesions were identi-
ed on the pre-treatment scans. All lesions
consisted of enlarged lymph node masses, which
were all ovoid in shape. As for the phantom
study, scans were performed at 1 cm collimation,
pitch 1.0:1 with axial image reconstruction at
1 cm intervals using a soft tissue algorithm and
displayed at abdominal soft tissue window
settings (level 40 HU, width 400 HU).
The maximum diameter of each lesion on pre-
and post-treatment scans was determined at the
workstation using the electronic cursor. This
measurement corresponds to the new unidimen-
sional measure (RECIST guideline) proposed for
assessing treatment response [6].
The maximum cross-sectional area of each
lesion on pre- and post-treatment scans was
determined at the workstation using the electronic
cursor (maximum diameter multiplied by perpen-
dicular bisector) (Figure 1), representing the
conventional 2D measure (WHO criteria) of
tumour size [2].
3D reconstruction and CT volume determina-
tion were performed for each lesion (Figure 1)
on pre- and post-treatment scans using the
Advantage Windows workstation as described
above.
The fractional size change (%) of each lesion
with treatment was determined with respect to
unidimensional (1D), cross-sectional area (2D)
and volumetric (3D) measurements. Fractional
change (%) in CT tumour size was calculated as:
Fractional change (%)~
New tumour size{initial tumour size|100
Initial tumour size
1179
 S A Sohaib, B Turner, J A Hanson et al

(a) (b)

Figure 1. 60-year-old man with enlarged right iliac

lymph node mass from a testicular teratoma.
(a) Contrast enhanced CT scan shows nodal mass, with
two-dimensional measurements indicated. (b) The out-
line of the nodal mass is traced on the workstation,
from which the software calculates the volume. (c) A
three-dimensional reconstruction of the nodal mass is
generated following outlining of the mass in (b).

(c)

Comparison was made between the response
categories according to 1D, 2D and 3D assess-
ment of tumour response. For 1D measurements,
tumour response was categorized using the new
RECIST guidelines (Table 1) [6, 7]. For 2D
measurements, tumour response was categorized
using the established WHO criteria for adult
patients with cancer (Table 1) [2]. For volumetric
3D measurement of tumour response, we used a
volume increase greater than 40% for progressive
disease and a volume reduction of at least 65% for
partial response. These thresholds were used as
they correspond to existing response categories
for 2D measures of response, assuming the
tumour changes uniformly (Table 2).
Similarly, the coefcient of variation exceeded
10% for the smaller and irregularly shaped
phantoms.
Patient study
Initial tumour volume as determined by CT
ranged from 26 cm3 to 911 cm3; post-treatment
tumour volume ranged from 0 cm3 to 433 cm3.
The relationship between changes in volume and
cross-sectional area is shown in Figure 2. Table 4
shows the treatment response categories according
Table 2. Relationship between change in diameter,
cross-sectional area and volume, assuming a uniform
change in size

Results Diameter Area Volume

Phantom study Response Decrease Decrease Decrease

30% 50% 65%
Results of the phantom study are shown in 50% 75% 87%
Table 3. The CT volume measurement error was Progression Increase Increase Increase
small (5%) for regularly shaped phantoms larger 12% 25% 40%
than 38 cm3 (i.e. diameter . 4 cm, phantom Nos 20% 44% 73%
25% 56% 95%
17 in Table 3). Errors increased as the phantoms 30% 69% 120%
became either smaller or irregular in shape.

1180 The British Journal of Radiology, November 2000

CT assessment of tumour response to treatment

Table 3. Phantom study

Phantom Reference CT volume (cm3) CT measurement error (%) Coefcient of variation (%)
volume (cm3)
Regular Irregular Regular Irregular Regular Irregular

1 575 598 589 4.07 2.47 1.6 4.1

2 397 402 389 1.26 1.99 1.8 4.4
3 268 265 298 1.04 11.19 2.8 10.6
4 136 132 153 2.65 12.79 6.1 12.2
5 102 107 111 4.41 8.53 11.9 9.3
6 65 68 74 5.08 14.15 0.89 2.82
7 38 39.7 44.8 4.47 17.89 10.2 10.2
8 19.5 21.6 23.7 10.77 21.54 3.0 6.9
9 11.5 10 12.4 13.04 7.83 37 9.2
10 3.5 2.5 4.3 28.6 22.9 14 19

to 1D, 2D and 3D measures of change in tumour Discussion

load. No patient had progressive disease. There
was complete agreement between 2D and 3D
assessment of response. However, in 2 (10%) out
of 20 lesions there was a difference in the response
classication between the 1D and 2D assessment
of response. In one lesion, the longest diameter
decreased by 25% and the area by 58%, and in
another the diameter decreased by 34% and the
area by 49%.
This study conrms that volume measurements
determined from helical CT data with appropriate
image processing software are accurate and
reproducible for large and regularly shaped
objects. The CT volume measurement error was
5% or less for regular phantoms of volume 38 cm3
or larger. This compares well with previous
studies, which have reported accuracy of approxi-
mately 5% for CT volume measurements using

Figure 2. Relationship between fractional change (decrease) in volume and fractional change (decrease) in cross-
sectional area of the tumour following treatment. Points represent individual lesions. The continuous line indicates
the expected change in volume for a given change in cross-sectional area, assuming that the mass shrinks
uniformly.

The British Journal of Radiology, November 2000 1181


Table 4. Two-dimensional and three-dimensional assessment of tumour response category
Assessment of response Unidimensional (1D)
CR PR
Cross-sectional area (2D) CR 2 0
PR 0 7
NC 0 1
CR, complete response; PR, partial response; NC, no change, as dened in Table 2.
both conventional and helical techniques [1, 3, 4].
In addition, our intraobserver error rate for 3D
measurements is comparable with previous
reports, again dependent on lesion size [8, 9].
We did not assess interobserver variation but
others have reported interobserver error rates of
1015% [8, 9].
Limited accuracy and reproducibility for small
objects is due to a combination of partial volume
effect and measurement error. For a given slice
thickness and pitch, the partial volume effect
is greater for smaller objects, and in smaller
objects the measurement errors are proportion-
ately larger. The effect of object shape on volume
measurement accuracy is clearly important, as
biological lesions may be irregular. Larger
measurement error and limited reproducibility
for the irregular phantoms is due to lower
accuracy of tracing the outline of irregular
objects. This point has not been emphasized
previously. Other factors known to inuence
measurement accuracy on CT include viewing
window settings, slice thickness, intravenous
contrast medium and observer variation
[3, 10]. In our study these parameters were
kept constant to minimize potential sources of
error.
Management of patients with cancer is inu-
enced by the response of the tumour to treatment.
Criteria for tumour response to treatment were
developed following two meetings of WHO
during the 1970s (Table 1) [2] and have been
applied to CT techniques for assessing response to
treatment. More recently, a new set of guidelines
(RECIST) has been proposed to evaluate solid
tumour response to treatment [6, 7, 11]. The
rationale behind the change is that it would allow
a simpler and more rapid assessment of tumour
response to treatment. The major change in the
new set of guidelines is the use of 1D measure of
tumour size (i.e. change in maximum diameter) to
classify response. The denition of complete
response is essentially the same for both systems.
For partial response, WHO requires a 50%
decrease in maximum area, whereas RECIST
requires a 30% decrease in the longest diameter of
the tumour. These criteria for partial response are
almost equivalent if one assumes a rounded
1182
S A Sohaib, B Turner, J A Hanson et al
Volumetric (3D)
NC CR PR NC
0 2 0 0
1 0 8 0
9 0 0 10
tumour shrinks uniformly. For progressive
disease, WHO requires a 25% increase in max-
imum area whilst RECIST requires a 20%
increase in maximum diameter. The criterion for
progressive disease is not equivalent between the
two systems, as a 20% increase in diameter
equates to 44% increase in area.
In our study we did not nd a notable
difference in tumour response classication
using 1D, 2D or 3D measures of tumour
response. In agreement, a previous report com-
paring 1D and 2D measurement of response
found high concordance between the two methods
of assessment [11]. A retrospective review using
RECIST (1D) and WHO (2D) criteria was
applied to data from the patients recruited in 14
different trials [6]. This showed that there was
good agreement between the two methods of
assessment, with only a small difference in partial
response, stable disease and progressive disease
rates between the WHO and RECIST criteria [6,
7]. Our study shows the discrepancies arise when
the size changes are near the boundaries for the
response categories.
If the percentage change in the longest diameter
of a tumour does not reect change in other
dimensions then measuring cross-sectional area or
volume would be more accurate. There are few
studies that have used CT volume measurements
to assess tumour response to treatment [1214],
and in only one of these was the response
determined by 2D and 3D measurements com-
pared [12]. This study found that up to 23% of
patients would have been classied into a different
response category if a 3D method of calculating
tumour size change had been used in place of
conventional 2D measurements. However, it is
interesting to note that of the re-classied
responses 48% were upgraded and 52% were
downgraded. Lack of a consistent pattern to this
re-classication suggests that it may have been
due to variability in the 2D and 3D measure-
ments. Furthermore, this group applied the WHO
criteria to both 2D and 3D measures of change in
tumour size. For any structure that alters in size,
the volume change is generally larger than the
cross-sectional area change. Therefore, it would
not be appropriate to classify the size change
The British Journal of Radiology, November 2000
CT assessment of tumour response to treatment
measured by a 3D technique with criteria devel-
oped for 2D planar data. It can be shown that
the difference between the fractional change
(%) in volume and change (%) in cross-sectional
area will always be less than 15% for regular
tumour masses that reduce uniformly in size
(Appendix 1). This difference is of similar
magnitude to the intraobserver and interobserver
error rates [8, 9] and is maximal around the size
of tumour reduction resulting in a partial
response.
We did not nd a meaningful difference in
treatment response classication between 1D, 2D
and 3D measurements for several reasons. First,
the smallest change in tumour size recognized by
the response criteria is 25% [2, 6]. Second, we
studied only nodal tumour masses, which tend to
be regular ovoid structures and show proportional
changes in all dimensions. Also, the denitions for
partial response as determined by 1D, 2D and 3D
measurements have been adjusted to reect
similar change in size. However, for progressive
disease the RECIST criteria of a 20% increase in
longest diameter would correspond to a 44%
increase in cross-sectional area and a 73% increase
in volume. We did not have any patients with
progressive disease and were therefore unable to
assess this aspect. A larger study would be needed
to assess the impact of this RECIST criterion.
Furthermore, CT tumour volume measurement is
likely to be most contributory when assessing
irregularly shaped tumours that show non-uni-
form size changes. As our study population was
restricted to nodal tumour masses, further study
of irregular shaped tumours is required to assess
whether 1D, 2D or 3D measurement is best. For
our group of patients, the volumetric assessment
did not add further information over a conven-
tional 2D measurement technique, which should
remain the current method of choice for assessing
tumour response to treatment.
In summary, we have shown that, for helical
CT, volume measurements are accurate and
reproducible for larger regular lesions, but are
less reliable for small irregular lesions. 3D
measurements are time consuming to perform,
tracing one lesion over multiple slices can take up
to 2030 min compared with 12 min for 1D or
2D measurements. Tumour size changes following
treatment show close agreement between 1D, 2D
and 3D measurements. In our study, small
differences were found between the established
WHO criteria and the new RECIST criteria when
evaluating response. Larger studies are needed to
assess the new criteria. Most importantly, data to
show whether 1D, 2D or 3D assessment of
tumour response correlates more closely with
outcome are not currently available.
The British Journal of Radiology, November 2000
References
1. Husband JE. Monitoring tumour response. Eur
Radiol 1996;6:77585.
2. Miller AB, Hoogstraten B, Staquet M, Winkler A.
Reporting results of cancer treatment. Cancer
1981;47:20714.
3. Nawaratne S, Fabiny R, Brien JE, Zalcberg J,
Cosolo W, Whan A, et al. Accuracy of volume
measurement using helical CT. J Comput Assist
Tomogr 1997;21:4816.
4. Breiman RS, Beck JW, Korobkin M, Glenny R,
Akwari OE, Heaston DK, et al. Volume
determinations using computed tomography. AJR
1982;138:32933.
5. Van Hoe L, Haven F, Bellon E, Baert AL, Bosmans
H, Feron M, et al. Factors inuencing the accuracy
of volume measurements in spiral CT: a phantom
study. J Comput Assist Tomogr 1997;21:3328.
6. Therasse P, Arbuck SG, Eisenhauer EA, et al. New
guidelines to evaluate the response to treatment in
solid tumours. J Natl Cancer Inst 2000;92:20516.
7. Gehan EA, Tefft MC. Will there be resistance to
the RECIST? J Natl Cancer Inst 2000;92:17981.
8. Hopper KD, Kasales CJ, Van Slyke MA, Schwartz
TA, TenHave TR, Jozeak JA. Analysis of
interobserver and intraobserver variability in CT
tumour measurements. AJR 1996;167:8514.
9. Van Hoe L, Van Custem E, Vergote I, Baert AL,
Bellon E, Dupont P, et al. Size quantication of
liver metastases in patients undergoing cancer
treatment: reproducibility of one, two and three
dimensional measurements determined with spiral
CT. Radiology 1997;202:6713.
10. Koehler PR, Anderson RE, Baxter B. The effect
of computed tomography viewer controls on
anatomical measurements. Radiology 1979;130:
18994.
11. James K, Eisenhauer E, Christian M, et al.
Measuring response in solid tumours:
unidimensional versus bidimensional measurement.
J Natl Cancer Inst 1999;91:5238.
12. Hopper KD, Kasales CJ, Eggli KD, TenHave TR,
Belman NM, Potok PS, et al. The impact of
2D versus 3D quantitation of tumour bulk
determination on current methods of assessing
response to treatment. J Comput Assist Tomogr
1996;20:9307.
13. Quivey JM, Castro JR, Chen GT, Moss A, Marks
WM. Computerized tomography in the quantitative
assessment of tumour response. Br J Cancer
1980;41(Suppl. 4):304.
14. Friedman MA, Resser KJ, Marcus FS, Moss AA,
Cann CE. How accurate are computed tomographic
scans in assessment of changes in tumour size? Am
J Med 1983;75:1938.
Appendix
Consider a regular ovoid of diameters 2x1, 2y1,
2z1, initial cross-sectional area A1 and volume V1,
which changes uniformly in size by a factor f such
that
A2 ~fA1 A1
and
1183
 S A Sohaib, B Turner, J A Hanson et al

The difference D between fractional change in

x2 =x1 ~y2 =y1 ~z2 =z1
A2 =A1 ~nx2 y2 =nx1 y1
and
V2 =V1 ~4=3nx2 y2 z2 =4=3nx1 y1 z1 A4
Substituting Equations A1 and A2 into Equations
A3 and A4, then
V2 =V1 ~(A2 =A1 )3=2 or f 3=2
A2
volume and cross-sectional area is given by:
A3 D~f 3=2 {f
Using differential calculus, it can be shown
that the maximum value of D occurs when
3=2f 1=2 ~1 (for 0f1, i.e. reductions in size).
Therefore, the maximum difference between
fractional reduction in volume and cross-sectional
area of a regular ovoid is 15%, which occurs when
the cross-sectional area decreases by 44%.

1184 The British Journal of Radiology, November 2000