Abstract

Context.—Whole-slide images (WSI) are a tool for remote interpretation, archiving, and teaching. Ovarian frozen sections (FS) are common and hence determination of the operating characteristics of the interpretation of these specimens using WSI is important.

Objectives.—To test the reproducibility and accuracy of ovarian FS interpretation using WSI, as compared with routine analog interpretation, to understand the technology limits and unique interpretive pitfalls.

Design.—A sequential series of ovarian FS slides, representative of routine practice, were converted to WSI. Whole-slide images were examined by 2 pathologists, masked to all prior results. Correlation characteristics among the WSI, the original, and the final interpretations were analyzed.

Results.—A total of 52 cases, consisting of 71 FS slides, were included; 34 cases (65%) were benign, and 18 cases (35%) were malignant, borderline, and of uncertain potential (9 [17%], 7 [13%], and 2 [4%] of 52 cases, respectively). The correlation between WSI and FS interpretations was 96% (50 of 52) for each pathologist for benign versus malignant, borderline, and uncertain entities. Each pathologist undercalled 2 borderline malignant cases (4%) as benign cysts on WSI. There were no overcalls of benign cases. Specific issues within the benign and malignant groups involved endometriosis versus hemorrhagic corpora lutea, and granulosa cell tumor versus carcinoma, respectively.

Conclusions.—The correlation between original FS and WSI interpretations was very high. The few discordant cases represent recognized differential diagnostic issues. Ability to examine gross pathology and real-time consultation with surgeons might be expected to improve performance. Ovarian FS diagnosis by WSI is accurate and reproducible, and thus, remote interpretation, teaching, and digital archiving of ovarian FS specimens by this method can be reliable.

Whole-slide imaging (WSI) is now technologically feasible. Using modern, digital scanning devices, pathologists are capable of rendering entire pathology glass slides into digital files at resolutions of high enough quality to recapitulate the appearance of analog images of tissue sections viewed in a standard microscope. As scan resolutions have increased, these digital images (virtual slides) can simulate the microscopic image viewed at any of the magnifications traditionally used to make clinical interpretations. The uses of such digital whole-slide images are many and include archiving of digital slides, potentially obviating the need to retain large glass slide files; maintaining records of consultation slides returned to original institutions; allowing standardized materials to be “broadcast” to, or archived for, large numbers of individuals simultaneously for Internet and other digital media-based continuing medical education and testing methods; and using digital files to make remote interpretations (telepathology) via the Internet or internal network connections.

One important use of WSI technology is the remote, rapid interpretation of frozen sections (FS). The ability to rapidly produce a virtual slide of a FS would allow consultation in near real time for underserved areas and for difficult cases where expert consultation may be necessary for an optimized patient outcome. To validate the use of this technology in clinical practice, performance of interpretation between conventional onsite, glass slide and virtual slide methods must be compared in a large group of cases. Because ovarian FS are common and have significant intraoperative import, the present comparative study of routine and WSI interpretation on a large, consecutive sample set of such cases was performed. The data obtained in this experiment were then compared against prior studies to determine whether the use of WSI provided similarly accurate and reproducible levels of FS interpretation when compared with traditional glass slide methods.

MATERIALS AND METHODS

The protocol was approved by the appropriate institutional human subjects review board. Glass slides from a series of consecutive ovarian FS were retrieved from the files of a large referral university hospital. Frozen section slides were collected from the recent archive files beginning 6 months before study initiation. The original, routine FS interpretation (clinical interpretation) and the final, permanent section interpretation (reference diagnosis) were recorded for each slide and case. Original FS and final permanent diagnoses were made by general surgical pathologists with consultation as appropriate with gynecologic pathology subspecialists. Each glass slide was converted to a WSI (virtual slide) using a Zeiss Mirax Desk scanning device (Zeiss, Inc, Göttingen, Germany). The virtual slides were reviewed by 2 board-certified general surgical pathologists, each having approximately 1 year of experience viewing WSI specimens, using standard computer desktop monitors and the Mirax Desk viewer software. The WSI reviewers were different from the pathologists making the original FS and the permanent final diagnoses. The Mirax Desk viewer allows the pathologist to review the digital image at any magnification ordinarily used in a standard microscope with similar resolution capability (up to ×400 standard microscope magnification with magnification at any point between those of standard microscope objectives). No clinical information was given to the WSI reviewers and interpretations were made solely on the basis of the morphologic examination. Each pathologist recorded a result for each of the cases viewed in the digital format (WSI interpretation). The results of clinical and WSI interpretations were then compared and correlations were tabulated. Two sets of results were obtained. The first compared the correlations in the broad category of benign versus malignant, borderline, and uncertain, and the second compared correlations for specific interpretation. In all cases, the final, permanent section diagnosis determined the reference diagnosis.

RESULTS

There were 52 FS cases analyzed in the series, comprising a total of 71 slides. Thirty-five cases consisted of single FS slides, 14 cases consisted of 2 slides, and 3 cases consisted of 3 slides. Each slide took an average of 9 minutes to scan (range, 1 minute, 51 seconds to 20 minutes, 33 seconds). The reference diagnoses represented in the set included 34 benign cases (65% of total cases) and 18 malignant, borderline, and uncertain cases (35% of total cases; 9 [17%], 7 [13%], and 2 [4%] of 52 cases, respectively). The specific diagnoses represented in the total data set are presented in Table 1. The overall correlation rate between the clinical and WSI interpretations for the benign category versus the malignant, borderline, and uncertain category was 96% (50 of 52) for each of the 2 pathologists. Each pathologist underinterpreted 2 WSI cases as benign cysts; both of which were interpreted as borderline tumors on the original FS and confirmed as such on permanent final review (reference diagnosis). When compared with the specific permanent and clinical FS interpretations, WSI pathologist interpretations showed misclassifications of several malignant and benign processes (Table 2). The most common misinterpretations were benign ovarian parenchyma or leiomyoma in cases of ovarian fibroma, endometriosis in cases of hemorrhagic corpora lutea, and adenocarcinoma in cases of granulosa cell tumor (Figure 1, a and b). However, in several cases, the WSI interpretation was more precise or closer to the permanent interpretation than was the original FS interpretation. For example, a correct WSI interpretation of a borderline tumor was rendered in one case when the original FS interpretation was benign, and in another case, a correct WSI interpretation of adenocarcinoma was rendered when the FS interpretation was of a borderline tumor (Figures 2, a and b).

Figure 1. a and b, Illustrated in the Mirax viewer (Zeiss, Inc, Göttingen, Germany) is an example of a granulosa cell tumor of the ovary that was correctly interpreted as such by the original frozen section and erroneously interpreted as a carcinoma, possibly of neuroendocrine origin, in 1 of the 2 digital whole-slide interpretations. The second digital whole-slide interpretation was correct (hematoxylin-eosin, original magnifications ×0.29 [a] and ×20 [b]).

Figure 2. a and b, Illustrated in the Mirax viewer (Zeiss, Inc, Göttingen, Germany) is an example of a metastatic endocervical adenocarcinoma in the ovary, which was interpreted as a borderline tumor on the original frozen section and correctly interpreted as invasive adenocarcinoma (query endometrial origin) in 1 of the 2 digital whole-slide interpretations. The second digital whole-slide interpretation was also of a borderline tumor (hematoxylin-eosin, original magnifications ×0.38 [a] and ×10 [b]).

Figure 1. a and b, Illustrated in the Mirax viewer (Zeiss, Inc, Göttingen, Germany) is an example of a granulosa cell tumor of the ovary that was correctly interpreted as such by the original frozen section and erroneously interpreted as a carcinoma, possibly of neuroendocrine origin, in 1 of the 2 digital whole-slide interpretations. The second digital whole-slide interpretation was correct (hematoxylin-eosin, original magnifications ×0.29 [a] and ×20 [b]).

Figure 2. a and b, Illustrated in the Mirax viewer (Zeiss, Inc, Göttingen, Germany) is an example of a metastatic endocervical adenocarcinoma in the ovary, which was interpreted as a borderline tumor on the original frozen section and correctly interpreted as invasive adenocarcinoma (query endometrial origin) in 1 of the 2 digital whole-slide interpretations. The second digital whole-slide interpretation was also of a borderline tumor (hematoxylin-eosin, original magnifications ×0.38 [a] and ×10 [b]).

Table 1

Reference Diagnoses (Correct General Classification, %)a

Reference Diagnoses (Correct General Classification, %)a
Reference Diagnoses (Correct General Classification, %)a
Table 2

Specific Interpretation Comparisonsa

Specific Interpretation Comparisonsa
Specific Interpretation Comparisonsa

COMMENT

Whole-slide imaging interpretation of ovarian FS was found to be similar in accuracy to routine glass slide FS interpretation, for general classification of benign versus the category of malignant, uncertain, and borderline. Because this distinction is the most important result of a rapid interpretation, use of WSI to make this intraoperative decision appears to be safe and accurate. The WSI FS results herein are similar to previous studies of glass slide evaluations compared with permanent section final diagnosis. In a meta-analysis of 18 studies of FS versus permanent diagnosis, Geomini et al1 found that correlations ranged between sensitivities of 65% and 100%, with near-perfect specificities of greater than 97%. In another study of 243 ovarian FS interpretations compared with final permanent section diagnoses, Pinto et al2 showed correlations similar to the present study, with 98.5% agreement for malignant diagnoses, 94% agreement for benign diagnoses, and 78.6% correlation for borderline tumors. The borderline category is the most difficult and most prone to interobserver variability. Sensitivity for the detection of borderline tumors was reported to be 61%, and yet other authors show even lower borderline tumor sensitivities,3 whereas others have been higher at 71%.4 In addition, and again as was noted in the present study, Pinto et al2 suggested that granulosa cell tumors were problematic for FS interpretation with both overinterpretations as malignant and underinterpretations as benign being noted in their series. In one case in the present study, both WSI observers raised the differential diagnosis of a metastatic origin for a tumor that was of primary ovarian type on permanent section. Metastasis has been shown to enter the FS differential diagnosis in 14.5% of cases of primary ovarian malignancy and so is not an unexpected query when using WSI technology for primary assessment of FS.5 

Interestingly, in the present study, WSI interpretation matched the permanent final diagnosis better in some cases than did the original “clinical” FS diagnosis. This indicates that FS skill may be as important as the mode of presentation of the histologic section and again highlights the possibility that any observed variability may not be due to technology differences but to well-documented FS interobserver reproducibility issues.

The current study has limitations because the pathologist interpreting the WSI FS did not have access to clinical information, gross assessments, the exact site of the tissue sampled, or real-time patient information, all of which would be expected to improve performance, particularly with specific category evaluation. The key parameters of difference between microscopic and WSI evaluations relate to the method in which the tissue is viewed: via standard microscope viewing or using video screens with manipulation of images via a computer-based viewer. Although inherently different methods, the pathologists using the system were easily trained in its operation and were able to arrive at accurate interpretations. Both observers in this study reported spending about 3 to 5 minutes interpreting each case (although formal timing was not done and no information on the time to perform the original FS was captured for comparison purposes), and no technical issues related to interpretation were identified by either observer.

Other studies have shown that accurate interpretations can be rendered on permanent section WSI viewed by remote observers.6,7,8 A more recent study comparing robotic microscopy telepathology to WSI telepathology for FS diagnosis reported a 98% accuracy rate for both techniques on 350 and 633 FS, respectively, of all organ types in a real-time practice setting.9 These authors report that FS diagnosis by WSI decreased the turnaround time when compared with the robotic method by 4-fold. The present study reports the use of WSI technology capability in a subset of ovarian FS to document any specific issues in this organ, which is a common and sometime difficult FS category. Our results are similar to the prior study and highlight that interpretive issues related to the technology were not major issues for the reviewers.

To validate a new WSI-based system for telepathology applications in FS interpretation and teaching, performance of a large series of cases in each organ system must be compared with conventional microscope-based interpretations to ensure accuracy and patient safety. This study is the first such series, to our knowledge, of consecutive ovarian FS representative of a routine case-mix in general practice. Accuracy in this study could have been further improved by real-time information not provided to the WSI pathologists, and hence, based on the results, the use of WSI for ovarian FS interpretation is a reliable alternative for remote interpretation.

References

References
1
Geomini
,
P.
,
G.
Bremer
,
R.
Kruitwagen
, and
B. W. J.
Mol
.
Diagnostic accuracy of frozen section diagnosis of the adnexal mass: a metaanalysis.
Gynecol Oncol
2005
.
96
(
1
):
1
9
.
2
Pinto
,
P. B.
,
L. A. L. A.
Andrade
, and
S. F.
Derchain
.
Accuracy of intraoperative frozen section diagnosis of ovarian tumors.
Gynecol Oncol
2001
.
81
(
2
):
230
232
.
3
Rose
,
P. G.
,
R. B.
Rubin
,
B. E.
Nelson
,
R. E.
Hunter
, and
F. R.
Reale
.
Accuracy of frozen-section (intraoperative consultation) diagnosis of ovarian tumours.
Am J Obstet Gynecol
1994
.
171
(
3
):
823
826
.
4
Tempfer
,
C. B.
,
S.
Polterauer
,
E-K.
Bentz
,
A.
Reinthaller
, and
L. A.
Hefler
.
Accuracy of intraoperative frozen section analysis in borderline tumors of the ovary: a retrospective analysis of 96 cases and review of the literature.
Gynecol Oncol
2007
.
107
(
2
):
248
252
.
5
Stewart
,
C. J.
,
B. A.
Brennan
,
I. G.
Hammond
,
Y. C.
Leung
, and
A. J.
McCartney
.
Accuracy of frozen section in distinguishing primary ovarian neoplasia from tumors metastatic to the ovary.
Int J Gynecol Pathol
2005
.
24
(
4
):
356
362
.
6
Fine
,
J. L.
,
D. M.
Grzybicki
,
R.
Silowash
, et al
.
Evaluation of whole slide image immunohistochemistry interpretation in challenging prostate needle biopsies.
Hum Pathol
2008
.
39
(
4
):
564
572
.
7
Gilbertson
,
J. R.
,
J.
Ho
,
L.
Anthony
,
D. M.
Jukic
,
Y.
Yagi
, and
A. V.
Parwani
.
Primary histologic diagnosis using automated whole slide imaging: a validation study.
BMC Clin Pathol
2006
.
6
:
4
.
8
Li
,
X.
,
J. L.
Liu
,
H.
Xu
, et al
.
A feasibility study of virtual slides in surgical pathology in China.
Hum Pathol
2007
.
38
(
12
):
1842
1848
.
9
Evans
,
A. J.
,
R.
Chetty
,
B. A.
Clarke
, et al
.
Primary frozen section diagnosis by robotic microscopy and virtual slide telepathology: the University Health Network experience.
Hum Pathol
2009
.
40
(
8
):
1070
1081
.

Author notes

From the Department of Pathology, University of Massachusetts, Worcester (Drs Fallon and Prasad); the James Homer Wright Pathology Laboratories, Massachusetts General Hospital, Boston (Dr Wilbur); and the Department of Pathology, Harvard Medical School, Boston (Dr Wilbur). Dr Fallon is now with the Department of Pathology, North Shore Medical Center, Salem, Massachusetts. Dr Prasad is now with the Department of Pathology, Yale University School of Medicine, New Haven, Connecticut.

Financial Disclosure: Zeiss, Inc, provided support for whole-slide scanning and image review. The authors have no other relevant financial interest in the products or companies described in this article.

Presented at the annual meeting of the United States and Canadian Academy of Pathology, Denver, Colorado, March 1–7, 2008.