Acta Ophthalmologica 2018

Tumour thickness, diameter, area or volume? The

prognostic significance of conventional versus
digital image analysis-based size estimation
methods in uveal melanoma
Thorsteinn Snaebj€ornsson Arnljots,1 Zaid Al-Sharbaty,1 Emma Lardner,1 Charlotta All-Eriksson,1,2
Stefan Seregard1,2 and Gustav St
alhammar1,2
1
Ophthalmic Pathology and Oncology Service, St. Erik Eye Hospital, Stockholm, Sweden
2
Department of Clinical Neuroscience, Karolinska Institutet, Stockholm, Sweden

ABSTRACT.
Purpose: The aim of this study was to compare conventional and novel size a later stage after which the outcome is
estimation methods’ ability to predict survival in uveal melanoma (UM). inevitably fatal (Kujala et al. 2003).
Identifying the patients that will
Methods: The study was designed as a retrospective consecutive chart review of
develop metastasis is therefore key in
patients with UM, enucleated between the years 1984 and 1993. Area and
UM diagnostics. Several methods have
volume were estimated based on the largest histopathological cross-section, the been proposed and implemented,
second centroid theorem of Pappus and digital image analysis, correlated to among which gene expression assays
overall and relative survival. have shown excellent prognostic utility
Results: Of 168 patients analysed, 20 (12%) of tumours were categorized as T1, (Onken et al. 2004, 2012; Gill & Char
47 (28%) as T2, 67 (40%) as T3 and 19 (11%) as T4 (15 N/a). A total of 91 2012; Staby et al. 2017). However, this
tumours with complete survival and measurement data were included and may not be universally available in
recategorized into small, medium and large volume groups. Increased separation clinical routine, resulting in a need for
of overall survival was seen compared with current American Joint Committee alternative methods. This holds espe-
on Cancer T categories. Difference between the large and small volume groups cially true for the large portion of
was 8.6 years (p = 0.001), compared to a difference of 5.6 years (p = 0.091) patients with UM not undergoing enu-
between T1 and T4. Hazard ratio for all-cause mortality in the large versus small cleation as primary therapy. In these
volume group was 2.6 compared to 1.9 for T4 versus T1. Relative survival rates cases, tumour material might not
for small, medium and large volumes were 62, 44 and 31% at 10 years, versus always be available (Singh et al. 2011).
As in most other solid tumours,
50, 45, 56 and 0% for T1, T2, T3 and T4.
including lymph node-negative breast
Conclusion: This study provides evidence that a novel UM volume estimation
cancer, tumour size is one of the
method might offer a practical and cost-efficient alternative to improve the strongest prognostic factors regarding
prognostic value intrinsic to a tumour’s size. survival in UM (Cianfrocca & Gold-
stein 2004; Shields et al. 2013; Horto-
Key words: classification – digital image analysis – size – uveal melanoma – volume bagyi et al. 2017; Kivel€ a et al. 2017).
Current methods for size estimations
Acta Ophthalmol. 2018: 96: 510–518 rely on manual measurement of the
ª 2018 Acta Ophthalmologica Scandinavica Foundation. Published by John Wiley & Sons Ltd tumour’s largest basal diameter (LBD)
doi: 10.1111/aos.13668
and thickness during ultrasound (US)
and gross inspection of resected speci-
cases per million inhabitants and year, mens in UM. In breast cancer, size
partially depending on geographic estimations rely on macro- and micro-
Introduction location and skin tone (Singh et al. scopically defined largest tumour diam-
Uveal melanoma (UM) is the most 2011; Jovanovic et al. 2013). Although eters. Although UM and breast tumours
common primary intraocular malig- less than four percent of patients have typically assume dome-like or spheroi-
nancy among adults (Singh et al. detectable metastases at diagnosis, dal shapes, respectively, outlines vary
2011). Incidence reaches two to eight roughly half will develop metastasis at significantly (Singh et al. 2007) (Fig. 1).

510

Fig. 1. Current American Joint Committee on Cancer (AJCC) T category size parameters in the
measurement of uveal melanomas (UMs) take the largest basal tumour diameter (LBD) and
thickness into account (Kivel€a et al. 2017). Tumours show a wide array of different shapes,
including (A) dome-shape, (B) mushroom-shape and (C) marked asymmetry. (D) Ultrasound
examinations of UMs offer an opportunity to measure tumour size in vivo. Green lines symbolize
LBD and thickness in each tumour. (E) Current UM T category classification as defined by the
AJCC staging manual, 8th Edition.
Previously published reports have
evaluated different methods of estimat-
ing tumour volumes (Guthoff 1980;
Gass 1985; Char et al. 1997; Li et al.
2003; Richtig et al. 2004). These studies
base tumour volume calculations on the
assumption that tumours have either a
spheroidal (Guthoff 1980; Li et al. 2003)
or ellipsoidal shape (Gass 1985; Char
et al. 1997; Richtig et al. 2004). Richtig
et al. (2004) found tumour volume, as
calculated with their method, to com-
pare favourably to LBD and thickness.
However, when performed on a larger
patient material, these results could not
be repeated (Kujala & Toivonen 2006).
On the whole, these methods share a
common limitation. As UMs are known
to show great variation in growth pat-
tern, uniformly assigning all tumours a
spheroidal or ellipsoidal shape is a
compromise (Gass 1985; Char et al.
1997; Li et al. 2003; Richtig et al. 2004;
Singh et al. 2007).
In this study, we evaluate a novel
method of estimating tumour volume
in an attempt to adapt size estimations
to each tumour’s specific shape. Using
digital image analysis (DIA) and a
specific calculating formula, we esti-
mate volumes of enucleated UM and
compare to current size measurement
methods, correlated to overall survival
(OS) and relative survival rates (RSR).
Additionally, the findings are validated
in smaller cohorts of UMs in vivo based
on US images and digitally scanned
breast cancer slides.
Materials and Methods
The study adhered to the tenets of the
Declaration of Helsinki and the proto-
col was approved by the regional eth-
ical review board in Stockholm.
Patient selection
Cohort 1
Consecutive clinical records and
archives of haematoxylin–eosin stained
tissue sections from all patients that
underwent enucleation for UM (chor-
oid, ciliary body or iris) at the Oph-
thalmic Pathology and Oncology
Service at St. Erik Eye Hospital in
Stockholm between the years 1984 and
1993 were reviewed retrospectively.
Mortality data were retrieved from the
Swedish population register. Patients
were excluded if reduced quality of
tissue sections prevented proper estima-
tion of the tumour’s outline. A total of
168 patients enucleated between April
1984 and February 1993 were analysed.
Glass slides with tissue sections from
respective tumours were photographed
using a digital camera (Infinity 1–5 5.0
Megapixel USB 2.0 Microscopy Cam-
era; Lumenera Corporation, Roper
Technologies Inc., Ottawa, Canada)
mounted on a stereo microscope (Nikon
Acta Ophthalmologica 2018
SMZ460 Zoom Stereomicroscope;
Nikon Corporation, Shinagawa,
Tokyo, Japan).
The following clinicopathological
factors were analysed as follows: age at
diagnosis, gender, date of diagnosis,
date of death, location of tumour epi-
centre (choroid, ciliary body or iris), cell
type (spindle cell, mixed or epitheloid),
treatment prior to enucleation (pretreat-
ment) as well as tumour LBD and
thickness (mm) as measured after enu-
cleation. The latter measurements were
used to assign each tumour a T category
as defined by the American Joint Com-
mittee on Cancer (AJCC) staging man-
ual, 8th Edition (Kivel€ a et al. 2017) as
well as a tumour thickness subcategory.
The thickness groups were defined as 1:
thickness 0–6.0 mm, 2: 6.1–9.0 mm, 3:
9.1–15.0 mm and 4: >15.0 mm.
Pretreatment consisted of episcleral
plaque brachytherapy (n = 57),
transpupillary thermotherapy (n = 1)
or cryotherapy (n = 1). Tumours that
had undergone pretreatment before
enucleation (n = 59) were excluded
from most statistical analyses out of
concern of effect on tumour size. No
significant differences were seen on OS
between smaller and larger pretreated
tumours, regardless of measurement
method (Fig. S1). Patient demograph-
ics are listed in Table 1.
Overall survival was selected over
disease-specific survival as the primary
outcome, as registries of the latter have
been previously found to be inaccurate
(Bergman et al. 2003). Instead, 5- and
10-year RSR were added as a surrogate
for the disease-specific measure. In brief,
generalized data from the Swedish life
tables were used to assign each patient a
remaining life expectancy, based on
year-per-year statistics for the general
population with the same sex and age
(Statistiska Centralbyr an 1992; Statis-
tiska Centralbyr an 2001). This allowed
for the estimation of a reference survival
when grouping patients by various
tumour size categories: 5- and 10-year
RSR = observed cumulative propor-
tion of the patients in a specific UM size
category surviving at the end of year 5
and 10 after diagnosis∕expected cumu-
lative proportion surviving in the gen-
eral population (Bergman et al. 2003).
Cohort 2
As material from the primary tumour of
UMs is only available if the eye has been
enucleated or the tumour biopsied, we
511
Acta Ophthalmologica 2018

Table 1. Patient demographics and clinical features. TNM size classification as defined by the American Joint Committee on Cancer staging manual,
8th edition (Kivel€
a et al. 2017).

1 (pretreated patients
Cohort no. 1 excluded, n = 59) 2 3

Type of material Stereo microscope photos Stereo microscope photos In vivo ultrasound images Digitally scanned
of uveal melanoma of uveal melanoma of uveal melanoma breast cancer
glass slides glass slides glass slides
n 168 109 21 20
Gender, n (%)
Male 91 (54) 55 (50) 15 (71)
Female 76 (45) 53 (49) 6 (29) 20 (100)
N/a 1 (1) 1 (1)
Mean age at diagnosis, years (range) 64 (23–92) 66 (27–92) 70 (48–85) 58 (N/a)
Mean tumour diameter, mm (range) 13 (5–20) 13 (6–20) 16 (10–26) 13 (3–19)
Mean tumour height, mm (range) 8 (2–19) 9 (2–19) 10 (2–25) N/a
TNM size classification, n (%)
1 20 (12) 10 (9) 1 (5)
1a 3 (15)
1b 6 (30)
1c 11 (55)
2 47 (28) 27 (25) 5 (24)
3 67 (40) 49 (45) 10 (47)
4 19 (11) 10 (9) 5 (24)
N/a 15 (9) 13 (12) 0 (0) 0 (0)
Cell type, n (%)
Spindle cell 87 (52) 55 (50) 11 (52)
Mixed 33 (20) 24 (22) 7 (33)
Epitheloid 42 (25) 28 (26) 2 (10)
N/a 6 (3) 2 (2) 1 (5) 20 (100)
Primary site, n (%)
Choroid 157 (93) 101 (93) 20 (95)
Ciliary body 8 (5) 5 (4) 1 (5)
Iris 0 (0) 0 (0) 0 (0)
N/a 3 (2) 3 (3) 0 (0)
Material resected, dates 1984–1993 1984–1993 2011–2014 1994–1996
5-year overall survival 59% (after enucleation) 56% (after enucleation) 31% (after 79% (after
ultrasonography) tumour resection)

sought to validate the volume estima-
tion method in an in vivo setting. There-
fore, we assembled a smaller cohort of
21 patients diagnosed with UM and
examined by US, at the Ophthalmic
Pathology and Oncology Service, St.
Erik Eye Hospital between the years
2011 and 2014. Ultrasound still images
of the tumours’ largest cross-sectional
areas were printed and saved to each
patient’s data journal. Subsequently,
volume estimation was performed.
Cohort 3
In an attempt to validate the feasibility
of the volume estimation method as
presented in this study, outside of the
UM setting, we also collected a small
sample population of 20 patients with
lymph node-negative stage I breast
cancers (largest tumour diameter
≤20 mm) (Hortobagyi et al. 2017).
These were surgically removed, paraf-
fin-embedded, sectioned and stained at
the accredited clinical laboratory of the
Department of Clinical Pathology,
Karolinska University Hospital, Stock-
holm between the years 1994 and 1996.
All glass slides were digitally scanned
at 920, using a Nano Zoomer 2.0 HT
(Hamamatsu Photonics K.K., Hama-
matsu, Japan). Last, volume estima-
tions were performed.
Volume estimation method
Volume estimations were performed
with the freely accessible, web-based
public domain software IMAGEJ
(Abramoff et al. 2004) developed at
the National Institutes of Health.
For each UM specimen, the glass
slide with the largest cross-sectional
tumour area was singled out for further
analysis. Largest basal diameter (LBD)
and thickness were measured manually
at the time of histopathological prepa-
ration by an experienced pathologist’s
assistant. In cases where no recorded
data regarding tumour dimensions were
available, largest thickness was mea-
sured directly from glass microslides.
Supporting this method, measurement
of LBD and thickness from glass micro-
slides has previously been shown to
correlate well with direct measurements
obtained the time of sectioning (Folberg
et al. 1985). Lastly, Pappus’ second
centroid theorem was applied to esti-
mate volume (Weisstein 2017) (Fig. 2).
Volume estimations were performed by
the first and last author.
The first two steps were performed
in IMAGEJ:
(1) Firstly, the software applies default
arbitrary values for length. To retrieve
results in millimetres, distance was
calibrated based on already known
largest thickness.
(2) Accounting for normally occurring
asymmetry, each half of the tumour
was estimated separately: tumour out-
line was drawn manually whereby
tumour area and position of the cen-
troid were calculated by IMAGEJ. To

512
 Acta Ophthalmologica 2018

with smaller tumours (7.9 years 95%

CI 5.3–10.4 versus 10.6 years, 95% CI
8.0–13.2, p = 0.13). In contrast, all
other size estimation methods did sep-
arate survival significantly: patients
with LBD 9 thickness above median
(>108.0 mm2) had significantly worse
OS of 7.1 years after diagnosis (95%
CI 4.7–9.6) versus the smaller tumours
at 11.6 years (95% CI 8.7–14.6. Differ-
ence: 4.5 years, p = 0.021). Patients
with largest cross-sectional tumour
Fig. 2. The second theorem of Pappus V = Ad states that the volume V of a solid revolution areas above median (>87 mm2) had
generated by rotating an area A around an external axis is equal to the product of that area and the significantly worse OS of 7.4 years
distance d travelled by its geometric centroid. This distance was calculated by having the digital image
(95% CI 5.3–9.5) compared to the
analysis software IMAGEJ that defines the distance r of the centroid in either half of the largest cross-
sectional area to the tumour’s central axis. The distance travelled by this half of the largest cross- smaller tumours of 11.3 years (95%
section d = pr is equal to the circumference of a semicircle with the radius r for an estimation of half CI 8.2–14.4. Difference: 3.9 years,
the total tumour volume. The volume of the other half of the largest tumour cross-sectional area p = 0.038), and patients with estimated
rotated in a second semicircle was then added for an estimation of the total tumour volume. Modified tumour volumes above median
from Weisstein (2017). Reproduced under a creative commons licence. (>793.2 mm3) had an OS of 6.8 years
(95% CI 1.1–4.7) versus 11.5 years for
mitigate the effect of achieving different followed a normal distribution, with an the smaller tumours (95% CI 8.6–14.4.
values between efforts, each side of the accumulation of cases in the intermedi- Difference: 4.7 years, p = 0.014)
tumour was drawn, measured and esti- ate T2–T3 categories (n = 24 and 38, (Fig. 3).
mated twice with a mean value respectively, or 77.5% of the total In other words, when separating
retrieved. cohort) and only a relatively small small and large tumours by the median,
(3) Lastly, volume was calculated number of cases classified as either T1 all methods provided meaningful prog-
according to the theorem for each of or T4 (n = 12 and 6, respectively, or nostic information except tumour
the halves of the tumour. The mean 22.5% of the total cohort). thickness. Estimations of tumour vol-
values of the separate sides were added The distribution of tumour thick- ume and LBD 9 thickness induced the
to represent the whole tumour. nesses, largest cross-sectional areas and largest separations of OS.
estimated volumes was increasingly left- Significant separations in survival
skewed (Fig. S2). Assuming size classi- between tumours larger and smaller
Statistical analysis
fications similar to the T categories in than the median were also seen when
For survival analysis, the Kaplan–Meier our sample, in the intervals of <the mean repeating the same approach for UM
method was used and for hazard of all- (l) 1 SD (r), l r, l + r and >l + r, US images and digitally scanned breast
cause mortality the Cox regression pro- would consequently induce an accumu- cancer sections (Fig. S3).
portional hazard analysis. The Log Rank lation of borderline cases in the lower
(Mantel-Cox) test of equality of survival and intermediate categories.
Survival according to T categories and
distributions for all levels of tumour sizes
tumour thickness
was used. When comparing more than Survival across small and large tumours
two ordered size groups, the Log-Rank T category T1–T4 tumours (n = 80) had
test for trend was used. Likelihood ratio For an estimation of each size mea- a mean OS after enucleation of 13.0
chi-square (LR v2) and change in LR v2 surement method’s prognostic capac- (95% CI 8.1–18.7), 10.3 (95% CI 6.8–
(LR-Dv2) were computed for estimation ity, patient survival was analysed based 13.2), 8.5 (95% CI 5.6–11.0) and 7.4
of the individual scoring methods’ prog- on tumour thickness (n = 91), (95% CI 0.7–4.9) years, respectively
nostic value and for the relative amount LBD 9 thickness (n = 80), largest (difference T4–T1 5.6 years, p = 0.091)
of prognostic information. Differences cross-sectional area (n = 91) and (Fig. 4A). When only tumour thickness
with a p < 0.05 were considered signifi- tumour volume (n = 91). Median (n = 91) was taken into account (1: 0–
cant, all p-values being two-sided. All tumour thickness (mm), median largest 6.0 mm, 2: 6.1–9.0 mm, 3: 9.1–
statistical analyses were performed using cross-sectional area (mm2) and median 15.0 mm, and 4: >15.0 mm), only one
IBM SPSS statistics version 24 (Armonk, volume (mm3) were used as cut-offs for tumour was assigned category 4 (1% of
NY, USA). The Sankey plot was gener- large versus small tumours. Also, LBD the cohort) and again no significant
ated with the JsFiddle online editor, multiplied by thickness (mm2) was used separation of distinct survival groups
version 0.5a2. in this particular part of the survival could be obtained (p = 0.218) (Fig. 4B).
analysis instead of the AJCC T cate-
gory to obtain a continuous variable
Results for direct comparison with the other
Survival across area and volume groups
measurement methods. Previous studies have shown that prog-
Sample distribution
Patients with tumour thickness nosis becomes increasingly concen-
In the main UM cohort, the frequency of above median (>8 mm) did not have a trated to the mean when patients
cases across T categories 1–4 roughly significantly worse OS than patients accumulate in the intermediate T2–T3

513
Acta Ophthalmologica 2018

AJCC size groups (Yue et al. 2017). In

an attempt to improve separation
of prognosis, cohort 1 was redivided
into three groups with similar sam-
ple distribution: based on analysis of
receiver operating characteristic with
equal emphasis on sensitivity and
specificity for all-cause mortality within
5 and 15 years from diagnosis of UM,
the thresholds for these three new
volume size groups were set to 400
and 1200 mm3. As a result, larger
proportions of the cohort could be
included in the groups with either poor
or relatively good survival (large versus
small tumours). Furthermore, signifi-
cant separation of prognosis was
obtained. Mean survival was 5.8 years
(95% CI 3.2–8.6), 8.0 years (95% CI
5.6–10.4) and 14.4 years (95% CI 10.7–
18.2) for the large, medium and small
size groups, respectively (difference
large–small 8.6 years, p = 0.001)
(Fig. 4C).
Repeating the same approach for the
largest cross-sectional area, setting the
thresholds at 68 and 115 mm2
improved the results to a similar but
lesser extent: mean survival at 6.6 years
(95% CI 3.4–9.9), 8.1 years (95% CI
5.7–10.4) and 12.6 years (95% CI 9.6–
15.7) for the large, medium and small
cross-sectional area groups, respec-
tively (difference large–small 6.9 years,
p = 0.012). Repeating this approach
for the measurements underlying the
current T category size classification,
that is LBD multiplied by thickness,
setting the thresholds for LBD 9
thickness at 72 and 142 mm2, did
however not improve the results at
all: mean survival at 7.3 years (95% CI
1.7–3.9), 9.3 years (95% CI 6.4–12.2)
and 13.4 years (95% CI 9.2–17.6), for a
large, medium and small redefined
LBD and tumour thickness group,
respectively (difference large-small
6.1 years, p = 0.10).
This also holds true for a compar-
ison of Cox regression hazard ratios for
all-cause mortality (HR): stratification
of tumours into small, medium and
large volume and area groups yielded
gradually and significantly increased
HR, but not for the T categories or
thickness groups.
As the model was developed and
first tested on the same patient material
Fig. 3. Kaplan–Meier overall survival estimates for the uveal melanoma cohort 1, respective (cohort 1), it was quite possibly overfit.
tumour size group separated by the median, including (A) tumour thickness, (B) largest basal To validate the volume estimation
diameter (LBD) 9 thickness, (C) largest cross-sectional area and (D) estimated tumour volume method outside of the primary setting,
parameters. Pretreated tumours excluded. Diagnosis date set at time of enucleation. we therefore transferred it to cohorts 2

514
 Acta Ophthalmologica 2018

and 3: when applied to US images of

UM in vivo, the tumour volume estima-
tions still outperformed LBD 9 thick-
ness in HR for all-cause mortality
between tumours larger and smaller
than the median (HR 20.4 and 15.9 for
large versus small tumours as defined by
estimated volume and LBD 9 thick-
ness, respectively). When applied to
our small cohort of digitally scanned
breast cancer slides, the HR for
tumours larger versus smaller than the
median as defined by the classic largest
tumour diameter measurement was not
significant. As defined by the estimated
tumour volume method, however,
the HR for tumours larger than the
median was significantly higher than the
smaller tumours (HR 4.9, p = 0.047,
Table 2).
That the volume estimations indeed
added new information, and not merely
multiplied the information already
obtained by the measurements of LBD
and thickness is illustrated in Fig. 5, in
which reclassifications between small
volume groups and higher T categories
and vice versa area frequent event.

Relative survival rates

Overall cumulative RSR at 5 and
10 years (n = 93) were 60% and 47%,
respectively. Naturally, a larger pro-
portion of patients were classified into
the largest size group using the three-
tiered volume method compared to the
four-tiered T category, tumour thick-
ness and largest cross-sectional area
methods. Differences in RSR at 5 years
between intermediate groups 2–3 in the
latter methods were minimal at 2–3%,
spanning a narrow range of 54–62%.
However, 10-year survival ranged
widely within the largest size groups
from 0% in T category and thickness
to 31 and 33% in the volume and area
groups, respectively. Interestingly, by
merging area groups 2–3, survival in
corresponding area and volume groups
stood out as quite similar. Table 3
summarizes RSR.

Likelihood ratios
When the T categories and estimated
tumour volumes were tested for their
Fig. 4. Kaplan–Meier overall survival estimates. (A) T categories 1–4 as defined by the American Joint
Committee on Cancer (AJCC) staging manual, 8th Edition (Kivel€a et al. 2017), (B) tumour thickness
individual prognostic value on OS by
groups 1–4 (0–6.0 mm, 6.1–9.0 mm, 9.1–15.0 mm and >15.0 mm, respectively) and (C) tumour volume Cox regression LR v2 in our main UM
groups small, medium and large (0–400 mm3, 401–1200 mm3 and >1200 mm3, respectively). The latter cohort, estimated tumour volumes
yielded superior separation of survival of 8.6 years between the smallest and largest tumour volume (v2 = 11.01, p = 0.001) but not T cat-
groups (p = 0.001). Pretreated patients excluded. LBD, largest basal diameter. egories (v2 = 2.86, p = 0.091)

515
Acta Ophthalmologica 2018

Table 2. Cox regression hazard ratios for all-cause mortality in uveal melanoma and breast
cancer.
Discussion
In a rapidly advancing field, much has
Mean p for change
been learned about UM tumour
Cohort no. Tumour size Size OS, in HR from
and type n classification group n years HR previous step
growth, metastasis and treatment. Indi-
vidualized patient management tools
1. Stereo 168 T category T1 12 13.4 1 and targeted molecular therapies are
microscope T2 24 10.0 1.4 0.39* being evaluated in clinical trials (Field
photos of uveal T3 38 8.3 1.8 0.11* & Harbour 2014). In spite of this,
melanoma T4 6 2.8 1.9 0.28* genetic testing is costly and may not
glass slides Tumour 1 32 12.5 1
be universally available to all patients.
thickness only 2 28 12.0 1.2 0.52*
3 30 9.9 1.3 0.30*
Consequently, clinicopathological
4 1 4.6 4.6 0.14* characteristics such as tumour dimen-
Largest Small 27 12.6 1 sion parameters will continue to be
cross-sectional Medium 41 8.1 1.7 0.032 included for prognostication in the
area Large 23 6.6 2.1 0.011 foreseeable future. However, current
Estimated Small 27 14.4 1 standards of measuring LBD and
tumour volume Medium 41 8.0 2.2 0.004 thickness may also be seen as mere
Large 23 5.8 2.6 0.002
surrogate markers for true tumour size.
2. In vivo 22 Base 9 height ≤Median 11 4.9 1
ultrasound images >Median 10 2.4 15.9 0.009
Comparing previously published
of uveal melanoma Estimated ≤Median 11 5.2 1 methods on estimating tumour volume
tumour volume >Median 10 2.4 20.4 0.004 in UM showed a wide span of varia-
3. Digitally 20 Largest ≤Median 10 9.4 1 tion. Estimates based on a spheroidal
scanned breast tumour diameter >Median 10 7.3 3.2 0.103* tumour shape gave consistently larger
cancer glass Estimated ≤Median 10 11.7 1 estimates compared to the ones per-
slides tumour volume >Median 10 6.19 4.9 0.047 formed under the assumption of a
HR = hazard ratio; OS = overall survival. hemi-ellipsoidal shape. Moreover,
* p value for difference in HR not significant on a 0.05 level. when volume estimations were calcu-
lated with three-dimensional US, these
estimations were consistently smaller
contributed significant prognostic p = 0.001). Adding estimated tumour than the mathematical models (Singh
information. Volume estimations were volume to the T categories confirmed et al. 2007).
then added to the T categories for a that the tumour volume added signif- With these limitations in mind, this
comparison of their relative amount of icantly more prognostic information study suggests a novel way of measur-
prognostic information. The LR Dv2 (LR–Dv2 = 8.930, p = 0.003). For the ing tumour volume based on cross-
indicated that volume estimations did digitally scanned breast cancer slides sectional outlines and DIA. By doing
add significantly more prognostic (n = 20), LBD again failed to con- so, each tumour volume estimation is
information than the T categories tribute with significant prognostic specific to every tumour’s shape cir-
alone (LR –Dv2 = 7.06, p = 0.008). information (v2 = 2.314, p = 0.13), in cumventing the problem of presuming
When repeating this approach for contrast to the estimated tumour vol- a uniform shape of all tumours. Nat-
US images of UM in vivo (n = 21), T ume (v2 = 4.525, p = 0.033). When the urally, there are limitations even to this
categories again did not contribute latter was related to the former, how- method.
significant prognostic information ever, it did not add significantly more The first and perhaps most obvious
(v2 = 1.981, p = 0.16), while estimated information on OS in this very small objection concerns our assumption that
tumour volume did so (v2 = 11.614, cohort (LR–Dv2 = 1.683, p = 0.20). the largest cross-section is representa-
tive for the entire tumour’s shape. In
tumours with significant asymmetry
that is not noticeable in the largest
cross-section, volume estimations will
be inaccurate: the manual sectioning of
the tumour might not always reflect
true maximal trans-sectional area. A
subjective input that could affect repro-
ducibility is the selection of the slide
representing the largest tumour cross-
section. When choosing the midline
dividing the tumour into halves for
volume estimation, shifting the midline
axis of the tumour slightly to either side
Fig. 5. Sankey plot illustrating the flow of tumour size reclassifications between T categories (left)
and estimated tumour volume groups (right). Note that this includes pretreated tumours (n = 59)
may also affect the end-result. To
as the comparison retains the very same cases in both measurements. Volume group 1 = 0– reduce the latter effect, volumes of
400 mm3; Volume group 2 = 401–1200 mm3; Volume group 3 > 1200 mm3. each side of the tumour were estimated

516

Table 3. Five and 10-year relative survival dividing patients into three tumour
rates (RSR) in non-pretreated uveal melanomas volume groups showed an increasingly
enucleated in Sweden between 1984 and 1993. even distribution of patients between
10- tumour size groups, in contrast to the
5-year year T and tumour thickness categories
RSR RSR where most patients were concentrated
Size category (%) (%) within the intermediate categories.
Thus, more patients in the intermediate
Overall RSR, n = 93 60 47 groups could be assigned either a
T category 1–4 (LBD and thickness)
better or worse prognosis, possibly
T category 1, n = 11 82 50
T category 2, n = 25 60 45 providing a more individualized prog-
T category 3, n = 43 62 56 nosis. Second, by merging the interme-
T category 4, n = 5 20 0 diate area groups 2–3, RSR in the area
N/A, n = 9 category became strikingly similar to
Tumour thickness groups 1–4 the volume groups. This is not surpris-
1 (0–6.0 mm), n = 30 74 41 ing as area measurements partly under-
2 (6.1–9.0 mm), n = 29 54 42 lay volume estimations. However,
3 (9.1–15.0 mm), n = 26 57 60
more importantly, volume estimation
4 (>15.0 mm), n = 8 33 0
Area groups 1–4 as presented in this study did compare
Area subgroup 1, n = 15 80 58 favourably to categorization of
Area subgroup 2, n = 34 59 46 tumours by area, as well as by current
Area subgroup 3, n = 32 57 45 T criteria. Third, our findings suggest
Area subgroup 4, n = 12 44 33 that tumour thickness alone cannot be
Volume groups 1–3 relied upon when prognosticating UM.
Volume subgroup 1, n = 25 84 62 Fourth, tumour size does seem to lose
Volume subgroup 2, n = 45 53 44
prognostic value once the tumour has
Volume subgroup 3, n = 23 45 31
been pretreated.
In the original AJCC data set, 24%
of tumours were classified as T1, 64%
as T2–T3 and 12% as T4 (Kujala et al.
separately then added to represent the 2013). In the validation study, 35%
whole tumour. were classified as T1, 60% as T2–T3
A second concern is that the prog- and 6% as T4 (The AJCC Ophthalmic
nostic value of the volume estimations Oncology Task Force 2015). Both
presented might be less favourable if studies showed significant survival dif-
applied to other cohorts or the general ferences between T categories. In our
population as conclusions about its study, a slightly higher proportion of
performance were mainly drawn from tumours were assigned intermediate
the same material used for its develop- T2–T3 size categories (68%), fewer
ment. However, the repeated prognos- patients in T1 (12%). Our patient
tic superiority of the volume estimation population contained cases diagnosed
method in the small cohorts of digitally with UM between 1984 and 1993,
scanned breast cancer slides as well as whereas the mentioned studies
in UM in vivo gave us some reassurance included patients diagnosed later than
of our results’ reproducibility. 2000. Consequently, factors such as
Third, the volume estimation proce- increased population diabetic screen-
dure is relatively time-consuming in its ing, number of cataract surgeries per-
present form. Digital images will have formed as well as the elderly
to be generated either through a micro- population engaging in more visually
scope-mounted digital camera, a digital demanding activities thereby seeking
glass slide scanner or US images in medical advice earlier for decreasing
digital format. The DIA process takes vision in later years might have con-
another 5–10 min in its current appli- tributed to the difference in distribution
cation. Therefore, we strongly encour- (Bergman et al. 2003). It is not impos-
age the development of specific sible that the different patient distribu-
applications for volume estimations tion among tumour size categories in
based on the second theorem of Pap- this study might have influenced the
pus, either as plug-ins to be used within results.
the IMAGEJ environment, or as indepen- In this study, T1 and T3 10-year OS
dent easy-to-use software. was 50 and 29% (Fig. 4A), consider-
Four important conclusions might ably lower than the 10-year melanoma-
be drawn from this study. First, specific survival of 89 and 58% for the
Acta Ophthalmologica 2018
corresponding categories in the AJCC
building data set (Kujala et al. 2013).
Of importance, influence from compet-
ing risks affects the results when apply-
ing OS in contrast to disease-specific
survival. On the other hand, it reduces
the potential source of error in cause of
death classifications. Still, one should
be cautious to generalize these results
to other cohorts and the general pop-
ulation before further validation of the
methods presented here has been con-
ducted including application of mela-
noma-related survival in contrast to
OS.
Relative survival rates has been
advocated as a popular surrogate for
disease-specific survival (Bergman
et al. 2003; Virgili et al. 2008). Both
studies included a larger study popula-
tion of >2900 patients collected during
a longer time span of up to 38 years,
with a RSR of 68.9-70.1% at 5 years
and 59.4% at 10 years. In this smaller
study, overall cumulative RSR was
slightly lower although to a similar
extent at both 5 and 10 years (Bergman
et al. 2003; Virgili et al. 2008).
As has been previously noted, three-
dimensional US volume estimation
might, in the future, present a more
accurate method of estimating tumour
volume (Singh et al. 2007). Therefore,
volume estimations based on US
images were also included but with a
significantly smaller cohort and
shorter follow-up period. Although
the results must be interpreted with
caution, there was a significant sur-
vival correlation. With these promising
results, a suggested next step would be
to apply the same method on a larger
cohort with a longer follow-up period
specifically addressing melanoma-
related survival.
Taken together, the findings in this
study showed that UMs grouped into
small, medium and large tumour vol-
ume groups, as estimated with the
method presented here, increased sep-
aration of distinct OS, RSR and HR
groups compared not only to conven-
tional T category parameters but also
to tumour thickness and cross-sectional
area. This is an important indication of
this method’s capabilities. Although
cohorts were of limited size, the long
follow-up of the patients in our main
cohort of up to 33 years, as well as the
repeated performance in our two vali-
dation cohorts, gives us a degree of
confidence in the findings.
517
Acta Ophthalmologica 2018
References
Abramoff MD, Magalh~aes PJ & Ram SJ
(2004): Image processing with ImageJ. Bio-
photonics Int 11: 36–42.
Bergman L, Seregard S, Nilsson B, Lundell G,
Ringborg U & Ragnarsson-Olding B (2003):
Uveal melanoma survival in Sweden from
1960 to 1998. Invest Ophthalmol Vis Sci 44:
3282–3287.
Char DH, Kroll S & Phillips TL (1997): Uveal
melanoma. Growth rate and prognosis.
Arch Ophthalmol 115: 1014–1018.
Cianfrocca M & Goldstein LJ (2004): Prog-
nostic and predictive factors in early-stage
breast cancer. Oncologist 9: 606–616.
Field MG & Harbour JW (2014): Recent
developments in prognostic and predictive
testing in uveal melanoma. Curr Opin Oph-
thalmol 25: 234–239.
Folberg R, Gamel JW, Greenberg RA,
Donoso LA & Naids RM (1985): Compar-
ison of direct and microslide pathology
measurements of uveal melanomas. Invest
Ophthalmol Vis Sci 26: 1788–1791.
Gass JD (1985): Comparison of uveal mela-
noma growth rates with mitotic index and
mortality. Arch Ophthalmol 103: 924–931.
Gill HS & Char DH (2012): Uveal melanoma
prognostication: from lesion size and cell
type to molecular class. Can J Ophthalmol
47: 246–253.
Guthoff R (1980): Modellmessungen zur Vol-
umenbestimmung des malignen Aderhaut-
melanoms. Graefe Arch Klin Ophthalmol
214: 139–146.
Hortobagyi GN, Connolly JL, D’Orsi CJ et al.
(2017): Breast. In: Amin MB, Edge SB,
Greene FL et al. (eds.) AJCC cancer staging
manual, 8th edn. Chicago: Springer 589–628.
Jovanovic P, Mihajlovic M, Djordjevic-Jocic J,
Vlajkovic S, Cekic S & Stefanovic V (2013):
Ocular melanoma: an overview of the cur-
rent status. Int J Clin Exp Pathol 6: 1230–
1244.
Kivel€a T, Simpson ER, Grossniklaus HE et al.
(2017): Uveal melanoma. In: Amin M et al.
(eds). AJCC cancer staging manual, 8th edn.
Chicago: Springer 805–817.
Kujala E & Toivonen P (2006): Calculated
tumour volume as a prognostic parameter
for survival in choroidal melanomas. Eye 20:
123–124; author reply 24–5.
Kujala E, M€ akitie T & Kivel€a T (2003): Very
long-term prognosis of patients with malig-
nant uveal melanoma. Invest Ophthalmol
Vis Sci 44: 4651–4659.
Kujala E, Damato B, Coupland SE, Des-
jardins L, Bechrakis NE, Grange JD &
Kivel€a T (2013): Staging of ciliary body and
518
choroidal melanomas based on anatomic
extent. J Clin Oncol 31: 2825–2831.
Li W, Gragoudas ES & Egan KM (2003):
Tumor basal area and metastatic death
after proton beam irradiation for choroidal
melanoma. Arch Ophthalmol 121: 68–72.
Onken MD, Worley LA, Ehlers JP & Harbour
JW (2004): Gene expression profiling in
uveal melanoma reveals two molecular
classes and predicts metastatic death. Cancer
Res 64: 7205–7209.
Onken MD, Worley LA, Char DH et al.
(2012): Collaborative Ocular Oncology
Group report number 1: prospective valida-
tion of a multi-gene prognostic assay in uveal
melanoma. Ophthalmology 119: 1596–1603.
Richtig E, Langmann G, M€ ullner K, Richtig G
& Smolle J (2004): Calculated tumour volume
as a prognostic parameter for survival in
choroidal melanomas. Eye 18: 619–623.
Shields CL, Kaliki S, Furuta M, Fulco E,
Alarcon C & Shields JA (2013): American
Joint Committee on Cancer Classification of
Posterior Uveal Melanoma (Tumor Size
Category) Predicts Prognosis in 7731
patients. Ophthalmology 120: 2066–2071.
Singh AD, Terman S & Sculley L (2007):
Estimating choroidal melanoma volume:
comparison of methods. Ophthalmology
114: 1212–1214.
Singh AD, Turell ME & Topham AK (2011):
Uveal melanoma: trends in incidence, treat-
ment, and survival. Ophthalmology 118:
1881–1885.
Staby KM, Gravdal K, Mørk SJ, Heegaard S,
Vintermyr OK & Krohn J (2017): Prognos-
tic impact of chromosomal aberrations and
GNAQ, GNA11 and BAP1 mutations in
uveal melanoma. Acta Ophthalmol [Epub
ahead of print].
Statistiska Centralbyr an (The National Swed-
ish Central Bureau of Statistics) (1992):
Livsl€angden i Sverige: Livsl€angdstabeller
f€
or riket och l€anen 1981–1990 (Swedish Life
€
Tables). Orebro: SCB tryck.
Statistiska Centralbyr an (The National Swed-
ish Central Bureau of Statistics) (2001):
Livsl€angden i Sverige 1991–2000:
Livsl€angdstabeller f€or riket och l€anen (Swed-
€
ish Life Tables). Orebro: SCB tryck.
The AJCC Ophthalmic Oncology Task Force
(2015): International Validation of the
American Joint Committee on Cancer’s 7th
Edition Classification of Uveal Melanoma.
JAMA Ophthalmol 133: 376–383.
Virgili G, Gatta G, Ciccolallo L, Capocaccia
R, Biggeri A, Crocetti E, Lutz JM & Paci E
(2008): Survival in patients with uveal
melanoma in Europe. Arch Ophthalmol
126: 1413–1418.
Weisstein EW (2017): Pappus’s Centroid The-
orem. Available at: http://mathworld.wolfra
m.com/PappussCentroidTheorem.html
(Accessed on 7 March 2017).
Yue H, Qian J, Yuan Y, Zhang R, Bi Y, Meng
F & Xuan Y (2017): Clinicopathological
characteristics and prognosis for survival
after enucleation of uveal melanoma in
Chinese patients: long-term follow-up. Curr
Eye Res 42: 759–765.
Received on March 11th, 2017.
Accepted on November 13th, 2017.
Correspondence:
Thorsteinn Snaebj€ornsson Arnljots, MD
St. Erik Eye Hospital
Polhemsgatan 50
112 82 Stockholm
Sweden
Tel: +46 8 672 30 00
Fax: +46 8 672 33 75
Email: thorsteinn.snaebjornsson-
arnljots@sll.se
Financial support for this study was provided from
€
the St. Erik Research Foundation (S:t Eriks Ogon-
forskningsstiftelse).
Supporting Information
Additional Supporting Information
may be found in the online version of
this article:
Figure S1. Kaplan–Meier overall survi-
val estimates for pre-treated patients.
LBD, largest basal diameter.
Figure S2. Sample distribution of all
uveal melanoma tumours in the main
cohort categorized by different size
categories. (A–D) all patients, (E–H)
pre-treated patients excluded. T cate-
gory = tumour size classification of the
American Joint Committee on Cancer
(AJCC) staging manual 8th Edition
(Kivel€a et al. 2017); Tumour thickness
groups 1 4: 1: 0 6.0 mm, 2: 6.1 9.0
mm, 3: 9.1 15.0 mm and 4: <15.0 mm.
Figure S3. Kaplan–Meier overall survi-
val estimates based on different tumour
size characteristics for the in vivo
ultrasound uveal melanoma cohort
(A–B) and breast cancer cohort of
histopathological cross-sections (C–D).