Superficial skin biopsies are common in dermatopathology
specimens and can pose a diagnostic challenge especially
in cosmetic sensitive areas such as the face. When broadly
transected so that the base of the lesion is not visualized,
well-differentiated squamous cell carcinomas (SCC), hypertrophic
actinic keratoses (HAK), irritated seborrheic keratoses
(ISK), and verruca vulgaris (VV) can look quite similar
on light microscopic examination. When we encounter
such a biopsy in which atypia is not visualized in the upper
half and thus a benign lesion is favored but a premalignant
or malignant squamous neoplasms cannot be excluded, we
sign out these cases as squamous acanthomas (SA) transected
cannot rule out malignancy with a note recommending
clinical follow up and /or repeat deeper biopsy. This diagnosis
is obviously frustrating to all parties involved and we
seek ways to be more unequivocal with our recommendation.
In our study, we investigate the use of morphometric
quantitative image analysis (IA) as a tool to aide in the diagnosis
of transected squamous neoplasms. IA involves using
computer software to objectively measure histologic image
characteristics. The exact software and technical approaches
may vary from study to study, but in the end, objective measurements
are made. Studies have investigated prognostic implications
of IA in various carcinomas such as colon, renal
cell, bladder, ovarian, and breast among others [1-11], and
other studies have looked at the diagnostic applications of
IA [12-22]. In addition, correlation of IA measurements to
genetic molecular alterations has been explored [13,23-26].
We used IA to evaluate specific analytical variables in diagnostically
clear neoplasms including the mean and median
nuclear sizes (NS), standard deviation of nuclear sizes as a
correlate of nuclear pleomorphism, and cellularity. We then
used the information obtained from known neoplasms to
construct IA diagnostic ranges that can be used to categorize
histologically challenging transected neoplasms on superficial
skin biopsies as either being benign (ISK and VV) or
pre-malignant/malignant (HAK and SCC).
Materials and Methods
60 diagnosed cases of ISK, VV, HAK, and SCC (15 cases of
each neoplasm) were retrieved from our archives. Also, 10
cases of transected SA cannot rule out malignancy were retrieved.
For the 60 cases of ISK, VV, HAK, and SCC, a representative
area of the stratum spinosum comprising 210,000 pixels was
selected. The stratum basale was excluded as it would not be
present on transected SAs
When analyzing the transected SAs, the largest possible area
of the stratum spinosum was selected with areas ranging
from 148,000 to 383,000 pixels.
Mean and median nuclear sizes (NS), pleomorphism (measured
by the standard deviation), and cellularity were determined
on the known neoplasms. Standard deviation of the
NS can be performed in morphometric studies to measure
the degree of nuclear pleomorphism [10,12,23,24,27,28].
Cellularity measurements were based on the number of cells
per 210,000 pixels. For the transected SAs, the cellularity
measurements were adjusted mathematically to the predicted
number per 210,000 pixels.
Using IA attributes of the known neoplasms, diagnostic
ranges were created. The IA attributes of the transected neo-plasms were then placed in these ranges to categorize them as
being either benign (ISK and VV) or pre-malignant/malignant
(HAK and SCC). Chart review and clinical follow-up information
was obtained on the transected neoplasms to confirm our
diagnostic categorizations. Statistical analysis was carried out
using Graphpad (GraphPad Prism version 5.04).
There were statistically significant differences between the
benign (ISK and VV) and the pre-malignant/malignant neoplasms
(HAK and SCC) when analyzing NS and cellularity.
Moving from ISK, VV, HAK, and to SCC, there was a progressive
increase in the NS as well as in the pleomorphism (Table 1). With cellularity, ISK had the highest cellularity, VV had a
lower cellularity, and HAK/SCC had the lowest cellularities.
Diagnostic ranges were created using the IA measurements of
the known neoplasms (Table 2). Using this table, the unknown
transected neoplasms were classified as either benign or premalignant/
malignant [Table 3a-Table 3c]. Some of the transected
neoplasms could not be reliably classified due to the overlap of
some diagnostic ranges, and were classified as indeterminate
in these instances. These classifications were clinically correlated.
The pleomorphism range turned out to be the most diagnostically
useful (Table 3b). 5/5 categorized benign lesions were
clinically benign. 3/3 categorized pre-malignant/malignant
neoplasms were clinically pre-malignant/malignant. 2 lesions
could not be categorized and deemed indeterminate. One of
these was clinically benign and the other pre-malignant/malignant.
As a general rule, cases that had a pleomorphism < 89
could be correctly classified as benign, and cases that had pleomorphism
>110 could be correctly classified as pre-malignant/
When placing the IA attributes of the transected neoplasms
into the NS ranges and cellularity ranges, a significant number
of the cases were indeterminate and some were misclassified
(Table 3a,Table 3c). Using the NS range, 4 cases were correctly
classified, but one case was misclassified. Using the cellularity
range, 3 cases were correctly classified, but 2 cases were misclassified.
Using IA, we discovered that the most reliable range to distinguish
the benign neoplasms from the malignant ones was
the pleomorphism range. While the NS and cellularity of the
known benign and pre-malignant/malignant neoplasms were
significantly different from one another, the diagnostic ranges
created were not useful to reliably distinguish the transected
benign neoplasms from the pre-malignant/malignant ones.
The pleomorphism range is the also most useful range of the
three we investigated because it is independent of any specific
technical or methodological approach. We used certain software
(Image J and Photoshop) and took the image at a certain
resolution (3.7 megabytes) prior to analysis. If another study
used different software or took the image at a higher or lower
resolution, the NS and cellularity measurements could easily
be different than ours. This lack of standardization among IA
studies is a problem that needs to be addressed [17,29]. However,
as the pleomorphism range is resistant to methodological
variation, it has the best clinical utility and can be easily comparable
to other potential IA studies.
As technology advances, pathology will become increasingly
digitally based. It is foreseeable that one day the microscope
will be abandoned in favor of digital computer images. As this
happens, morphometric IA measurements will become easier
to perform [1,29,30] and will play a greater role as an aid in
diagnosis. While some may advance the idea that IA could potentially
replace the pathologist and make a diagnosis solely
based on morphometric measurements, this is highly unlikely.
As pathologists, we make numerous 'measurements' that are
not easily quantifiable and measured by IA. Also, much of
what we do is informed by our acquired medical knowledge
and clinico-pathological correlation [20,30]. Still, IA can play
an important role as a tool in diagnosis analogous to the role of
immunohistochemistry, especially in histologically challenging
cases  such as ours. IA helps to decrease subjectivity and
helps increase inter-observer agreement [1,3,4,6,12,17,23].
One of main drawbacks to IA analysis in our study is the time
spent to perform the analysis. [1,29,30]. For a given case, the
average time for analysis was approximately 30-45 minutes
which included the time needed to capture the image, manipulate
it to ready it for analysis, and then perform the analysis.
While we may have been hindered by our technical prowess,
faster methods to select cells and perform the analysis would
make IA more clinically applicable. As digital pathology advances,
the speed will surely increase and clinical studies such
as ours will help form the basis of the diagnostic ranges needed
for accurate diagnosis.
In summary, in our study we found statistically significant
differences between the IA attributes of benign versus premalignant
neoplasms. Using the table of diagnostic ranges
and excluding indeterminate cases, the unknown transected
neoplasms were correctly classified benign or pre-malignant/
malignant 80%, 60%, and 100% of the time respectively based
on NS, cellularity, and pleomorphism ranges. The pleomorphism
diagnostic range was the most useful and reliable. As a
general rule, cases that had a pleomorphism < 89 could be correctly
classified as benign, and cases that had pleomorphism
>110 could be correctly classified as pre-malignant/malignant.
The pleomorphism range accurately categorized ambiguous
transected squamous neoplasms as being either benign or
pre-malignant in 8/10 cases. As such, the SD range can help
pathologists diagnose otherwise ambiguous transected neoplasms
and assist the pathologist in creating a more decisive
treatment recommendation for the patient.