Robotic telepathology: efficacy and usability in pulmonary pathology.

Abstract

Robotic telepathology is well established in the USA as a method of case referral, but is less frequently used in the UK. Using cases covering a broad spectrum of pulmonary pathology, this study assessed its application in primary diagnosis and its functionality in terms of accuracy of diagnosis and time per case, for both small biopsies and open lung biopsies/resections. Forty cases (20 bronchoscopic and 20 surgical lung biopsy/resection specimens) were reviewed in blinded fashion by a single pathologist using robotic telepathology. Connection between the John Radcliffe and Royal Brompton Hospitals was via 10 Mb/s LAN to the Internet (supported by the Joint Academic Network). The cases were then randomized and reviewed a second time with conventional light microscopy. Diagnosis, initial time to reach diagnosis, and overall time per case were recorded. In two bronchoscopic biopsy cases, there were clinically significant differences between telepathology and conventional light microscopy, one probably attributable to user inexperience and the other to either speed of image capture or digital image quality. In the surgical lung biopsies and resections, there was one variation of opinion: with telepathology a case was considered to be probably mesothelioma, whereas this was thought less likely on light microscopy. In both instances, immunohistochemistry was requested prior to clinical management. Telepathology was 14 times slower than conventional light microscopy when examining bronchoscopic biopsies. The average time spent per slide was 7 min 21 s, compared with 32 s per slide with conventional light microscopy. When assessing open lung biopsies and resections, telepathology was five times slower, at 6 min 13 s compared with 1 min 10 s with conventional light microscopy. This study showed that robotic telepathology is accurate for primary diagnosis in pulmonary histopathology, but modifications in both laboratory protocols and telepathology hardware are needed to decrease the time difference between telepathology and conventional light microscopy, for telepathology to be usable within the framework of a busy referral practice.

Full text

Page 1

Original Paper
Robotic telepathology: efficacy and usability in
pulmonary pathology
F. J. W.-M. Leong1, A. G. Nicholson2and J. O’D. McGee3*
1Nuffield Department of Clinical Laboratory Sciences, John Radcliffe Hospital, Oxford, UK
2Department of Histopathology, Royal Brompton Hospital, London, UK
3Nuffield Department of Medicine, John Radcliffe Hospital, Oxford, UK
*Correspondence to:
Professor J. O’D. McGee,
University of Oxford, Nuffield
Department of Medicine,
Level 5, John Radcliffe Hospital,
Headington, Oxford OX3 9DU,
UK.
E-mail:
james.mcgee@ndm.ox.ac.uk
Received: 14 March 2001
Revised: 2 October 2001
Accepted: 22 January 2002
Abstract
Robotic telepathology is well established in the USA as a method of case referral, but is less
frequently used in the UK. Using cases covering a broad spectrum of pulmonary pathology, this
study assessed its application in primary diagnosis and its functionality in terms of accuracy of
diagnosis and time per case, for both small biopsies and open lung biopsies/resections. Forty cases
(20 bronchoscopic and 20 surgical lung biopsy/resection specimens) were reviewed in blinded
fashion by a single pathologist using robotic telepathology. Connection between the John
Radcliffe and Royal Brompton Hospitals was via 10 Mb/s LAN to the Internet (supported by the
Joint Academic Network). The cases were then randomized and reviewed a second time with
conventional light microscopy. Diagnosis, initial time to reach diagnosis, and overall time per case
were recorded. In two bronchoscopic biopsy cases, there were clinically significant differences
between telepathology and conventional light microscopy, one probably attributable to user
inexperience and the other to either speed of image capture or digital image quality. In the
surgical lung biopsies and resections, there was one variation of opinion: with telepathology a case
was considered to be probably mesothelioma, whereas this was thought less likely on light
microscopy. In both instances, immunohistochemistry was requested prior to clinical manage-
ment. Telepathology was 14 times slower than conventional light microscopy when examining
bronchoscopic biopsies. The average time spent per slide was 7 min 21 s, compared with 32 s per
slide with conventional light microscopy. When assessing open lung biopsies and resections,
telepathology was five times slower, at 6 min 13 s compared with 1 min 10 s with conventional
light microscopy. This study showed that robotic telepathology is accurate for primary diagnosis
in pulmonary histopathology, but modifications in both laboratory protocols and telepathology
hardware are needed to decrease the time difference between telepathology and conventional light
microscopy, for telepathology to be usable within the framework of a busy referral practice.
Copyright # 2002 John Wiley & Sons, Ltd.
Keywords:telepathology; telemedicine; dynamic telepathology; lung
Introduction
Increasingly, histopathologists in the West are develop-
ing areas of subspecialization or interest. Given this,
and the rarity of certain types of specimens and path-
ologies, it is conceivable that some institutions will
become referral centres for the bulk of organ-specific
pathology. In the United States, major referral centres
such as the Armed Forces Institute of Pathology have
been accepting telepathology consultations for several
years [1,2]. However, telepathology in the United
Kingdom is still a relatively unsophisticated practice,
used by the minority of pathologists. Most of these
have experience of sending static images as an email
attachment (store-and-forward telepathology), viewing
selected images on a CD ROM, or viewing live images
via a system designed for video-conferencing (dynamic
telepathology).
Fully robotic systems represent the high-end of
telepathology, offering complete remote control over
all aspects of the distant microscope and a more
sophisticated, specialized software interface than that
seen with store-and-forward systems [3–5].
Robotic telepathology has been used extensively in
general histopathology practice [3,4,6–15]. Previous
studies have addressed its functionality in relation to
frozen section diagnosis and panel discussions of dif-
ficult cases of lung cancer [16–19], using robotic tele-
pathology or store-and-forward telepathology. Our
intention in this study was to assess robotic telepath-
ology further, particularly its usability and reliability in
the diagnosis of bronchoscopic biopsies, surgical lung
biopsies and pulmonary resections.
Materials and methods
Two sets of 20 cases were selected from the files of the
Royal Brompton Hospital and the John Radcliffe
Journal of Pathology
J Pathol 2002; 197: 211–217.
Published online 22 March 2002 in Wiley InterScience (www.interscience.wiley.com). DOI: 10.1002/path.1112
Copyright # 2002 John Wiley & Sons, Ltd.

Page 2

Hospital, comprising a total of 80 slides. The first set
comprised 20 bronchoscopic biopsies (cases 1–20), each
slide bearing three sections from three separate levels,
the levels taken at approximately 50 mm intervals in the
tissue. One case had tissue taken at two levels only.
The cases represented a sequential selection of bio-
psies taken at the Royal Brompton Hospital during
1999. The second set of cases comprised 20 randomly
212F. J. W.-M. Leong et al.
Copyright # 2002 John Wiley & Sons, Ltd.J Pathol 2002; 197: 211–217.

Page 3

selected surgical lung biopsies or resection cases (cases
21–40) from the John Radcliffe Hospital. All sections
were cut from paraffin-embedded tissue and stained
with haematoxylin and eosin. In all 40 cases, the path-
ologist using the system (AGN) was given the clinical
details of the cases, but was unaware of the original
histopathology diagnosis.
Computer hardware
Both departments had 10 megabit per second (Mb/s)
local area networks (LAN) connected to the Internet.
The connection from the LAN to the Internet was also
10 Mb/s, shared among multiple users. In the United
Kingdom, Internet access is supported by the Joint
Academic Network (JANet), which connects all uni-
versity institutions and some hospitals. As this connec-
tion is influenced by the number of concurrent hospital
Internet users and general Internet congestion, it is
not possible to determine actual bandwidth accurately.
To obtain some estimate of the transmission speeds
achievable in this study, we found that it took on
average 171 s to download a 35-megabyte file via the
web from Oxford to London. This is approximately
1600 kilobits/s or equivalent to 26 single-channel ISDN
lines. We used a fully robotic telepathology system
(Nikon E1000, Zem software). The server software
runson a WindowsNT4
350 MHz, 128 MB RAM, 24 GB hard drive) and is
connected to the Internet via LAN. The client software
operates under Windows 95/98/NT4. Images were digi-
tized (Matrox video capture card) from the analogue
output of a digital camera (JVC model KY-F58). The
microscope was calibrated for Kohler illumination, the
camera set to auto white balance, and the computer
monitor colour calibrated with gamma correction. The
microscope performed focusing automatically with the
option for manual override. Each individually cap-
tured image was 768 576 in 24-bit colour. Colour depth
was maintained throughout the study. Lossy compres-
sion was applied to the images prior to transmission
(JPEG, medium quality). The remote pathologist had
full control over all aspects of microscope operation
(stage movement, objective magnification, lighting,
server(PentiumIII,
focus) and selected the best way of viewing the slide
from the multiple possible layouts (Figure 1).
Internet and security issues
The use of the Internet for telepathology has been
addressed in several papers [14,20–28] for the purpose
of both store-and-forward and robotic/dynamic tele-
pathology. The principal concerns have been accessi-
bility, bandwidth reliability, and security [29,30]. For
the purpose of our study, the Internet represented an
available and reasonably reliable means of commu-
nication. No patient data were transmitted and the
system used had in-built 60-bit encryption.
Slides were initially assessed for both diagnosis and
time to achieve diagnosis using the robotic telepathol-
ogy system described above, set up at the John Racliffe
Hospital. The images were analysed on a Windows
NT4 workstation (Dell, Optiplex GX1, Pentium II
processor, 128 MB RAM, 4 GB hard drive, 17-inch
flat-screen monitor set on 1280r1240 pixels at 16-bit
colour, 75 Hz refresh), at the Royal Brompton Hospi-
tal. Cases were timed firstly from the point of view of
coming to a definite diagnosis; with a second time
representing the period required to screen the remain-
ing tissue. This was done because when a pathologist’s
assessment of a case is broken down into segments,
there is usually an initial time interval for formulating
a diagnosis, followed by a further time interval when
the remaining tissue is screened to find or exclude
additional relevant pathology. It is important to note
that although the first time recorded is referred to as an
‘initial diagnosis’, it is in fact when a diagnosis was
known with certainty.
The telepathology system operated by initially creat-
ing a low-power overview image using a montage of
images captured with a 0.5r objective lens, which
allowed current position to be tracked (Figure 1a). The
100 s spent on this process were excluded from time
measurements. All images viewed were stored in a
cache file on the remote workstation, much like a web
browser. This increased efficiency by eliminating repeat
requests for images already seen.
Several weeks later, the slides were randomized and
sent for assessment by standard light microscopy,
Figure 1. Screenshots of the software interface. The software interface has multiple possible appearances, depending on con-
figuration and user preferences. These screenshots illustrate several of the possibilities. Consistent features are the icon toolbar at
the top, which provides access to commonly used functions; the overview image at the top right, which tracks movement and can
also show which parts of the slide have been viewed; and the fine-focus window on the right, which displays a ‘live’ image using
video-conferencing codec. Other windows are used to record annotations (textual and graphical) and activate arrow pointers that
can be seen by all users (b), display the number of simultaneous users (for group sessions), show an image gallery of captured images
(b, c, d) and show optional ancillary image inputs such as video-conferencing cameras. Some of these windows may be closed to
make more room for the main viewing area (a). The main viewing area occupies most of the left side of the screen and may comprise
a single (a, b, d, e) or multiple windows (c). The latter allows viewing of the slide at several different magnifications simultaneously.
Additionally, a single window may be expanded to fill the entire screen or across multiple monitors, thus increasing the field of view
(e). The purpose of this figure is to demonstrate the software interface, not the histology. The histological images shown are from
cases used in this study and known to demonstrate mild chronic bronchiolitis (a), metastatic mature teratoma (b, c), and metastatic
synovial sarcoma (d, e). The print appearances should not be taken to be representative of the image quality when viewed on screen
Efficacy of robotic telepathology in pulmonary pathology213
Copyright # 2002 John Wiley & Sons, Ltd.J Pathol 2002; 197: 211–217.

Page 4

using an Olympus BX50, and the same data were
collected with regard to timing and diagnosis.
Results
The bronchoscopic biopsies (cases 1–20) comprised 11
inflammatory/infective conditions, one benign, two
dysplastic, one in situ and five invasive malignant
lesions (Figure 2). Seven hours and twenty minutes
were spent examining 60 slides with telepathology
(average 7 min 21 s per slide) compared with 31 min
using light microscopy (average 32 s per slide), the
latter being 14 times faster. This was also reflected in
the time taken to reach an initial diagnosis (Figure 3).
Within this group, there were two discrepancies bet-
ween telepathology and direct light microscopy. One case
(case 2) was incorrectly diagnosed as granulomatous
inflammation using telepathology, but was called only
non-specific chronic inflammation on direct microscopy.
The second (case 9) was incorrectly called invasive
non-small cell carcinoma with telepathology, but only
carcinoma in situ with direct microscopy. Both of these
errors occurred in the first 10 cases.
Thesecond setofcases
comprised nine inflammatory/infective conditions, two
benignlesions, andnine
(Figure 2). It took an average of 2 min 43 s per slide
using telepathology to derive an initial diagnosis,
compared with an average of 31 s per slide using
direct light microscopy (Figure 3). A total of 2 h 4 min
was spent fully assessing all 20 slides with telepath-
ology (average 6 min 13 s per slide) compared with
23 min 24 s (average 1 min 10 s per slide). In this
group, there was only one discrepancy. In case 23, meso-
thelioma or adenocarcinoma was diagnosed by tele-
pathology, but an ‘epithelioid proliferation favouring
reactive fibrous pleuritis and mesothelial hyperplasia’
was favoured using direct microscopy. However, a
differential diagnosis was suggested on both occasions
and there was suggested deferral to an immunohisto-
chemical panel. In addition, in one case (case 24), there
was concordance between telepathology and direct
microscopy with regard to diagnosis, but an incidental
tumourlet was noted using direct microscopy.
Overall, telepathology was approximately 14 times
slower per slide than direct microscopy when assessing
bronchoscopic biopsies, and five times slower for
surgical lung biopsies and pulmonary resections. In
relation to all 80 slides, telepathology was 10 times
slower.
(designated21–40)
invasivemalignancies
Figure 2. Breakdown of case type
Figure 3. Average time spent assessing a case: a comparison of telepathology (TP) with direct light microscopy (LM). Cases 1–20
were small biopsies, 21–40 were open lung biopsies or resection specimens
214 F. J. W.-M. Leong et al.
Copyright # 2002 John Wiley & Sons, Ltd. J Pathol 2002; 197: 211–217.

Page 5

Discussion
Most studies have addressed the use of telepathology
for general pathology. While there have been several
that have focused on lung pathology, direct compari-
son is difficult due to the differences in the nature of
the telepathology methodology or the types of cases
and endpoints used. Kayser et al. [16–18] examined the
use of store-and-forward telepathology for intraopera-
tive lung frozen sections and classification of paraffin-
embedded poorly differentiated bronchial carcinoma
and concluded that it was a viable tool for frozen
section consultation (primary opinion) and panel
discussion of difficultbronchial carcinoma cases
(second opinion). Wellnitz et al. [19] used robotic
telepathology to examine 109 frozen section specimens
from lung surgery patients. In this study, which used a
single ISDN line (64 kilobits/s), the average time per
case was 47.1 min (range 37.2–66.2 min). This was
attributed to the narrow bandwidth. There were 10
clinically significant errors and 11 cases were deferred,
compared with only three significant errors and one
deferred case using direct light microscopy.
The intention of our study was to extend earlier
work in pulmonary pathology, by assessing a broad
range of lung pathology cases with higher diagnostic
requirements than frozen sections, and also to under-
take this with more favourable bandwidth.
In our study, there were two cases where there was
significant disagreement between telepathology and
light microscopy. The first of these (case 2) was the
second case to be assessed and the error on telepathol-
ogy could possibly be attributed to user inexperience.
Another possibility is the resolution of the viewing
monitor used in this study, although there were six
other cases of non-specific chronic inflammation that
were correctly diagnosed. In case 9 (carcinoma in situ),
image quality may have been an issue, as we have
noticed that despite colour calibration, nuclei in a
digital image can appear slightly darker than normal.
Notably, there were no complaints regarding image
quality during the study and there was clear instruc-
tion that diagnostic opinions were not to be given in
the absence of sufficient confidence. We believe that
confidence stems from the basic skill level of the
pathologist and improves with familiarity of the digital
medium.
Robotic telepathology cannot match the speed of an
experienced consultant personally manipulating a slide
on a microscope stage and it could be argued that this
relative slowness is distracting and contributes to the
risk of errors such as in case 9. However, this problem
was not evident with case 11, a similar case that was
diagnosed correctly. A third case (23) was noted as a
significant disagreement, although in practical terms
final diagnosis would have only been made follow-
ing an immunohistochemical panel. It is therefore
debatable whether this represents a clinically significant
error. Differentiation between a benign versus a
malignant diagnosis in mesothelial proliferations is a
known problem area in diagnostic pathology [31], and
again, the slightly darker appearance on nuclei on
screen may account for the discrepancy. It is noticeable
that all three disagreements were due to an ‘over-
calling’ of the diagnosis on telepathology. These results
are in contrast to the study by Wellnitz et al. [19],
which found that 9 of the 10 errors were false-negative
diagnoses. The number of errors in our study was too
small to draw any conclusions but we could specu-
late that these were related to slight differences in the
digital image compared with the eyepiece view, with
which the assessor was unfamiliar. It seems likely that
greater usage of the system would lessen these errors,
as seen in previous studies [32,33] which found that
accuracy when assessing images in a digital medium
was related to, and improved with experience.
Telepathology, at least in this study, was slower than
direct microscopy. It is noticeable that when assessing
small biopsies, telepathology was 14 times slower on
the same slide but when assessing open lung biopsy
and resection tissue, this was reduced to a factor of
five. Past studies have not directly compared times to
diagnosis, focusing instead on the more significant
overall turnaround time.
We can attribute this difference in speed to hardware
and human issues, the limiting factor being mechanics
rather than transmission bandwidth. Current motor-
ized stages are unable to match the speed of a human
manipulating a regular stage and this is unlikely to
change in the near future. A robotic microscope also
moves the stage in smaller, more discrete intervals and
has to check focus each time. Accurate focusing was
time consuming and more focusing was required when
moving from one tissue piece to another. Focus was
achieved through a mixture of hardware autofocus and
manual override from the remote end. Different manu-
facturers have their own approach to this problem,
which is to a certain extent tissue dependent: the
thicker the tissue, the more difficult to find quickly a
focal plane. The feedback from this study has been
used to modify the autofocusing system, to provide a
quicker response.
Mechanics are not the only area that can be im-
proved. The reduction in times with experience is
one measure of the learning curve associated with
telepathology. The pathologist involved (AN) had
approximately 1 h to practise with the system before
undertaking the study, but rapidly adapted and was
assisted by one of the other authors (FJL), who was
more familiar with the software. We know from
previous studies [6,9] that the average time taken per
case decreases with experience. Dunn et al. [6] have
shown that over the course of 2000 telepathology cases
the time spent per case was reduced from 10.26 min to
3.58 min and viewing times per slide were reduced from
3.44 min to 1.13 min. Although they did not measure
the time spent viewing the slides with direct light
microscopy, they noted that their case turnaround time
decreased from 2.46 days to 1.5 days. Their concord-
ance rates with light microscopy were between 99%
Efficacy of robotic telepathology in pulmonary pathology 215
Copyright # 2002 John Wiley & Sons, Ltd.J Pathol 2002; 197: 211–217.