Primary frozen section diagnosis by robotic microscopy and virtual slide telepathology: the University Health Network experience

Seminars in Diagnostic Pathology (2009) 26, 165-176

Primary frozen section diagnosis by robotic microscopy and virtual slide telepathology: the University Health Network experience Andrew J. Evans, MD, PhD, Runjan Chetty, MB, BCh, DPhil, Blaise A. Clarke, MB, ChB, Sidney Croul, MD, Danny M. Ghazarian, MB, PhD, Tim-Rasmus Kiehl, MD, Bayardo Perez Ordonez, MD, Suganthi Ilaalagan, BSc, MLT, Sylvia L. Asa, MD, PhD From the Department of Pathology Laboratory Medicine Program, University Health Network, Toronto, Ontario, Canada. KEYWORDS Telepathology; Frozen section; Robotic microscopy; Virtual slide telepathology

Although telepathology (TP) has not been widely implemented for primary frozen section diagnoses, interest in its use is growing as we move into an age of increasing sub-specialization and centralization of pathology services. University Health Network (UHN) is a 3-site academic institution in downtown Toronto. The pathology department is consolidated at its Toronto General Hospital (TGH) site. The Toronto Western Hospital (TWH), located 1 mile to west of TGH, has no on-site pathologist and generates 5-10 frozen section cases per week. Over 95% of these frozen sections are submitted by neurosurgeons, in most cases to confirm the presence of lesional tissue and establish a tissue diagnosis. In 2004, we implemented a robotic microscopy (RM) TP system to cover these frozen sections. In 2006, we changed to a virtual slide (VS) TP system. Between November 2004 and September 2006, 350 primary frozen section diagnoses were made by RM. An additional 633 have been reported by VS TP since October 2006, giving a total of 983 frozen sections from 790 patients. Eighty-eight percent of these cases have been single specimens with total turnaround times averaging 19.98 and 15.68 minutes per case by RM and VS TP, respectively (P ⬍ 0.0001). Pathologists required an average of 9.65 minutes to review a slide by RM. This decreased 4 fold to 2.25 minutes following the change to VS TP (P ⬍ 0.00001). Diagnostic accuracy has been 98% with both modalities and our overall deferral rate has been 7.7%. Mid-case technical failure has occurred in 3 cases (0.3%) resulting in a delay where a pathologist went to TWH to report the frozen section. Discrepant cases have typically involved minor interpretive errors related to tumor type. None of our discrepant TP diagnoses have had clinical impact to date. We have found TP to be reliable and accurate for frozen section diagnoses. In addition to its superior speed and image quality, the VS approach readily facilitates consultation with colleagues on difficult cases. As a result, there has been greater overall pathologist satisfaction with VS TP. © 2009 Elsevier Inc. All rights reserved.

1. Introduction Reprinted with permission from Evans AJ, Chetty R, Clarke BA, et al: Primary frozen section diagnosis by robotic microscopy and virtual slide telepathology: the University Health Network experience. Hum Pathol 40:1070-1091, 2009. Address reprint requests and correspondence: Andrew J. Evans, MD, PhD, Staff Pathologist and Associate Professor, University Health Network, Toronto General Hospital, 200 Elizabeth Street, Toronto, Ontario, M5G 2C4, Canada. E-mail address: [email protected].

0740-2570/$ -see front matter © 2009 Elsevier Inc. All rights reserved. doi:10.1053/j.semdp.2009.09.006

Telepathology (TP), a term introduced in 1986 by Dr. Ronald Weinstein,1,2 refers to the delivery of pathology services over a distance. A more contemporary definition of TP would include the use of the internet to link a pathologist to a glass slide. Some centers have used TP extensively for routine surgical pathology,3-6 whereas others have used it more selectively for pathologist-to-pathologist consultation

166

Seminars in Diagnostic Pathology, Vol 26, No 4, November 2009

Table 1

Telepathology frozen section validation studies

Year

Primary author

Accuracy (%)

Deferral (%)

Time (Minutes/slide)

2000 2001 2002 2003 2007

Winokur20 Demichelis24 Kaplan21 Moser22 Frierson23

97 95 100 84.1 95

3 11 NA 7.4 NA

NA 6.2 2.8 14.2 NA

with frozen sections.7,8 TP was first used for primary frozen section diagnosis in 1989 in Norway,9 where video telemicroscopy was used to provide intra-operative coverage with 100% accuracy to five remote hospitals. In spite of the success of the Norwegian program and others that have followed,7,10-18 the pathology community as a whole has been slow to adopt this technology. The major barriers preventing more widespread use of TP for primary diagnostic purposes, including frozen sections, were identified by Weinstein et al in 200119 and are still relevant today. Firstly, the cost and time required to implement and maintain a TP program can be prohibitive in the absence of a solid business case. Secondly, there has been a general perception that TP is unacceptably slow and would not allow for the timely reporting of frozen sections. Finally, pathologists have generally been of the opinion that the technique will be inaccurate and that suboptimal image quality will result in errors with adverse patient outcomes and medicolegal consequences. The latter concern persists in spite of the existence of several retrospective validation studies demonstrating an average accuracy rate of 95% (range 84.1-100%)20-24 (Table 1) for TP frozen section diagnoses. Further, there have been at least 11 reports published between 1991 and 2007 showing an average diagnostic accuracy of 96% (range 89-100%) when TP was used for primary frozen section diagnosis7,10-18 (Table 2). With the exception of a few outliers,10,11,22 the majority of these papers have reported diagnostic accuracy values that fall within the accuracy range for frozen sections reported by conventional light microscopy.25 The imaging technology used for TP has evolved significantly from the low resolution video telemicroscopy used by Nordrum et al,9 to include static microscopy involving the transmission of selected images23,26,27 and dynamic robotic microscopy (RM). Most of the TP systems summarized in Tables 1 and 2 have used dynamic or static/dynamic RM whereby the viewing pathologist assumes remote control of the microscope. While such systems provide image quality that is more than adequate for diagnostic purposes, RM tends to be much slower than light microscopy and patience on the part of the pathologist is required if an accurate diagnosis is to be rendered. The development of virtual slide (VS) technology represents a marked advance in terms of the remote assessment of histologic slides. VS systems (Aperio, Vista CA; DMetrix, Tucson AZ, Olympus, Center Valley, PA; BioImagene Inc, Cupertino, CA) em-

ploy slide scanners that can rapidly digitize histological sections on glass slides from low to high magnification to produce images of outstanding quality that can be viewed over the internet. Validation studies with VS technology on formalin-fixed, paraffin-embedded tissue sections have reported excellent results in terms of diagnostic accuracy, when compared to conventional light microscopy.28-30 Data on the performance of VS systems on frozen sections is limited to a single pilot study of 15 cases.31 University Health Network (UHN) is a three-site academic medical center located in downtown Toronto, comprising Toronto General Hospital (TGH), Princess Margaret Hospital (PMH) and Toronto Western Hospital (TWH). TGH and PMH are located directly across from each other, while TWH is located approximately 1 mile (1.6 km) to the west of TGH. TWH, home to the Krembil Neuroscience Center, has had no regular on-site anatomical pathologist for over 12 years and no regular on-site neuropathologist since our department consolidated at TGH in early 2006. The surgical services at TWH have historically generated up to 10 frozen section cases per week, most of these requested by neurosurgeons. Since the vast majority of our routine reporting, teaching and research activities take place at TGH or PMH, sending a single pathologist over to TWH to cover a small volume of frozen sections was inefficient from a workflow perspective. In addition, there was no easy method of consulting with a colleague on difficult cases. Sending tissue by ground transportation from TWH to TGH for frozen section assessment was slow and unreliable. TP was viewed as a viable solution and a solid business case could be made in support of its implementation. This report is not a retrospective validation study, but is a performance summary of a service we have been using for patient care for almost 4 years. We began with RM in October, 2004 and switched to a VS system in October,

Table 2

Primary frozen section diagnoses by telepathology

Year

Accuracy Primary author (%)

Deferral Time (%) (Minutes/slide)

1991 1995 1997 1999 2000 2003 2003 2005 2005 2006 2007

Nordrum9 Oberholzer10 Steffen11 Della Mea12 Dawson13 Huttarew17 Terpe15 Sukal16 Hitchcock18 Huttarew14 Horbinski7

0 6* 4* NA NA NA NA NA NA 0 12-20

100 90.3 89 100 97 99.4 98 NA** 95.3 97.9 95.5-96.9

15 20-40 NA 4.5 3 1-36 15 NA NA 10.7 NA

NA, not available. *These reports did not provide deferral rates, but rather the percentage of cases during which technical problems were encountered. **This report did not provide diagnostic accuracy data, but concluded that TP was a useful adjunct in Moh’s surgery.

Evans et al

Primary Frozen Section Diagnosis

2006, allowing us to compare the performance of these two TP modalities.

2. Methods 2.1. Timeline for implementation The RM system arrived at UHN in early 2003 and approximately 18 months were devoted to due diligence prior to going live for patient care. Archived frozen section slides, covering a wide range of tissue types, were reviewed (principally by AJE and RC) in order to develop an approach to reviewing slides by RM in an accurate and timeefficient manner. During this time, the pathologists were in the same room as the robotic microscope. After reviewing approximately 300 cases, we had optimized our comfort level with the system and were in a position to determine the time required to accurately review a frozen section. Designated information technology (IT) personnel from UHN were involved in the implementation process from its earliest stages. Information meetings were held with the TWH surgeons whose practice required intra-operative pathology consultations. The surgeons were shown the RM system and they contributed to the development of a standard operating procedure (SOP) for performing frozen sections in the absence of an on-site pathologist. The SOP includes descriptions of daily maintenance, system checks, workflow, roles of surgeons, histotechnologists (HT) and pathologists, as well as expectations for TAT and back-up plans in the event of system failure. The SOP was presented for approval to the UHN Medical Advisory Committee and our malpractice insurance providers at the Canadian Medicolegal Protective Association. Guidelines from Health Canada (Therapeutic Products Program) concerning the use of medical devices were reviewed to ensure that we would be in compliance with any existing standards. The RM was installed in the surgical pathology laboratory at TWH in July of 2004 and the system went live for patient care following a 5-month period of HT training that included the performance of numerous dry-run frozen sections.

2.2. Telepathology frozen section protocol At the beginning of each day, the pathologist covering TP reviews the operating room schedule to identify cases likely to require frozen section support. Relevant clinical history and radiology can be reviewed through the UHN hospital electronic patient record. The pathologist reviews the schedule with the HT and completes a system check to ensure that the system is fully operational before specimens arrive. Operating room staff will typically call the HT 15 minutes before tissue is to be sent. The HT contacts the pathologist by telephone and/or pager when the tissue arrives and relays identifying information, the dimensions and gross characteristics of the specimen along with any clinical

167 information written on the requisition form. The pathologist instructs the HT on how to process the specimen. Frozen sections are cut at a thickness of 7 ␮m and are prepared in a standard manner using optimal cutting temperature (OCT) media, a ⫺20° C cryostat and H&E staining. The HT places only one 7 ␮m section on a given slide in order to minimize distortion due to air-drying. Touch preps or smears are used only if requested by the pathologist and are prepared in a standard fashion with immediate alcohol fixation followed by H&E staining. The HT records the time at which the tissue was received and the time at which a diagnostic image was first available to the pathologist. The pathologist records the time at which they first began to review an image and the time at which they reported a diagnosis to the surgeon. Total turnaround time (TAT) is defined as the interval between the receipt of tissue in the surgical pathology lab and the time of communication of a diagnosis. Slide preparation time (SPT) refers to the interval between receiving tissue and the availability of images for review by the pathologist. Slide interpretation time (SIT) refers to the interval between the receipt of the first diagnostic image by the pathologist and the communication of a diagnosis. All times are recorded using the clocks on the UHN network computers. The pathologist reports the diagnosis directly to the surgeon by telephone and the records the diagnosis in our laboratory information system (Copath, Cerner DHT, Boston, MA). In the event of technical problem with the TP system, we follow a basic trouble-shooting procedure. If the problem cannot be rectified within 5 minutes, the surgeon is notified of the delay and a pathologist goes over to TWH to report the case by light microscopy.

2.3. Robotic microscopy telepathology (November 2004-October 2006) Frozen section slides were initially reviewed by RM using a dynamic server-client system, the Leica TPS2 (Leica Microsystems Canada, Richmond Hill, Ontario) (Figure 1). The server at TWH consisted of a Dell Pentium computer with a 17⬙ RGB monitor and a Sony 3CCD camera mounted on a remote controlled microscope equipped with 6 objectives (2.5x, 5x, 10x, 20x, 40x and 63x). The Leica TPS2 also included a macroscopy station, allowing pathologists to review a gross specimen if required. The client station consisted of an identical Dell computer and monitor and was set up in a dedicated location at TGH. Review of a frozen section by RM could begin only after the HT generated a low-resolution overview scan of the slide and sent a consultation request to the pathologist at the client computer. The pathologist confirmed successful receipt of the case with the HT, who would remain near the microscope at TWH in the event of a malfunction. The pathologist used the mouse for multidirectional slide navigation and to control the magnification, focus (fine and coarse adjustments) and light intensity of the images being sent by the server (Figure 1). By RM, frozen section slides

168

Seminars in Diagnostic Pathology, Vol 26, No 4, November 2009

Figure 1 The Lieca TPS2 RM system was used between November 2004 and October 2006 to make 350 primary frozen section diagnoses. (A) The server comprised a fully robotic microscope with 6 objectives (2.5x-63x) and a CCD3 camera along with a macroscopy station (to the left of the microscope) for gross review of specimens. (B) The working screen of the client station consisted of a thumbnail image with a reticle (upper right) to indicate the location of the microscopic field on the slide. The pathologist controlled the microscope with mouse clicks using the control buttons located below the thumbnail image. Diagnostic images, sent by the server as 640⫻480 compressed JPEG’s at a standard network speed of 100 MB per second, are viewed on the left side of the working screen. Each image required 1-1.5 seconds to reach the client.

were reviewed as a series of 640 ⫻ 480 dpi compressed JPEG files transmitted over the UHN intranet at a standard network speed of 100 MB per second. Each manipulation of the microscope by the pathologist produced a new image that required 1-1.5 seconds to reach the client. A reticle was used in conjunction with the overview scan to ensure thorough review of the slide.

2.4. Virtual slide telepathology (October 2006 –present) In September 2006, an Aperio ScanScope CS (Aperio, Vista CA/Quorum Technologies, Guelph, ON, Canada) was installed at TWH on a trial basis. Over three weeks, the HT’s were trained on its use and frozen section slides were reviewed in parallel with RM. Cases were first reported by RM and then immediately reviewed by VS TP. It rapidly became clear that there were numerous advantages to the VS approach and a formal change away from RM was made after only 1 month. We selected 20x as the default-scan magnification for all frozen sections and cytology slides, however, slides can be scanned at 40x if required by the pathologist. When both frozen section and cytology slides are requested, the HT typically scans the frozen section first (requiring approximately 1 minute at 20x with a tile width of 240 ⫻ 240 pixels). The cytology slide is scanned while the pathologist is reviewing the frozen section (requiring 1-3 minutes at 20⫻ depending on the area of the slide that is scanned). The HT notifies the pathologist by telephone when a given slide has been scanned. The pathologist connects directly with the scanner’s computer across the UHN network and confirms with the HT that he/she is able to view the appropriate slide. The virtual slides are viewed as compressed JPEG files with a compression ratio of 30. With the system architecture shown in Figure 2, there is no need to move digital image files across the network and pathologists can review slides from

their office computer using a standard 21⬙ RGB flat panel monitor. Consultation with colleagues is also much faster than with RM, as several pathologists can simultaneously review a slide from their own offices without having to walk to a client station. The ImageScope software (Aperio, Vista CA/Quorum Technologies, Guelph, ON, Canada) facilitates smooth navigation of the slide to identify areas of interest at low magnification. It also allows for rapid movement to higher magnification to confirm diagnostic details. In this sense, slide review by VS TP more closely resembles conventional light microscopy than RM.

3. Results 3.1. Disclosure There are no biases concerning the data presented in this report as we have only client/vendor relationships with Leica Microsystems and Aperio.

3.2. Characteristics of the frozen sections examined by TP We have used TP to make diagnoses on 983 frozen sections from 790 patients, 350 by RM and 633 using the VS approach, between November 2004 and August 2008. The vast majority (97%) of these frozen sections have originated from neurosurgical procedures. Over 90% of the neurosurgical specimens were submitted to confirm the presence of lesional tissue and to establish a primary diagnosis for tumor type and histological grade. Less than 5% of all frozen sections have involved the evaluation of resection margins. Gross assessment has been required in ⬍1% of cases, those being partial gastrectomies where the surgeon

Evans et al

Primary Frozen Section Diagnosis

169

Figure 2 The Aperio ScanScope CS was permanently installed at TWH in October 2006 and has been used to make 633 primary frozen section diagnoses. The architecture of our VS TP system is shown in panel (A). Pathologists connect directly with the scanner’s computer via the internal local area network (LAN), preventing the need to move potentially large digital image files across a busy network. This system allows several different pathologists to view a case at the same time using their own office computers. (B) Full screen diagnostic images can easily be navigated at low to intermediate magnification by dragging with the mouse. Magnification can be adjusted by either the scroll wheel on the mouse or by dragging the bar in the upper left corner of the screen. Shown here is a frozen section from a metastatic carcinoma. A magnifier is placed over the interface between the tumor and adjacent normal brain.

was interested in the proximity of the tumor to the closest surgical margin. The neuropathology cases, summarized in Table 3, cover the spectrum of frozen sections that one

Table 3 section

Distribution of cases encountered at frozen

Neuropathology Reactive gliosis Infectious Glial Tumors Astrocytoma, low grade Astrocytoma, high grade Oligodendroglioma Ependymoma Meningioma Nerve-Sheath Tumors Schwannoma Neurofibroma Malignant peripheral nerve sheath tumor (MPNST) Pituitary/Craniopharygioma Central nervous system (CNS) Metastases Lymphoma/Plasmacytoma Soft tissue/sarcoma Miscellaneous Tumors Chordoma Myxopapillary ependymoma Medulloblastoma PNET Ganglioglioma Hemangioblastoma Dysgerminoma Langerhans Cell Histiocytosis General/Other Surgery Margins Miscellaneous Tissue Identification

% of total cases 1 1 7 22 3 1 13 8 1 1 4 19 1 4 ⬍1 ⬍1 ⬍1 ⬍1 ⬍1 ⬍1 ⬍1 ⬍1 3 1

would expect at a specialized neurosurgery center. These frozen sections most commonly involve astrocytic tumors and metastases followed by meningiomas and nerve sheath tumors. Frozen sections on pituitary lesions are not routinely performed at UHN. Our non-neuropathology cases have included incisional breast biopsies, liver biopsies from potential donor organs, peritoneal nodules incidentally found during surgery for intra-abdominal cancers, thyroid nodules, parathyroid biopsies submitted to confirm their localization during thyroid surgery and the occasional gastrectomy specimen requiring surgical margin assessment. On one occasion we received uterine contents to confirm the presence of products of conception in order to rule out an ectopic pregnancy. Most of our frozen sections (88%) have involved single pieces of tissue, in the range of 0.5 to 1.0 cm in maximum dimension, which could be evaluated with a single block and slide. The cases are well spaced throughout the day and it is a rare occurrence for multiple frozen sections to arrive simultaneously from different operating rooms. Multiple block cases have ranged from 2 to 15 separate specimens from a single patient. These tend to arise in the following situations: quadrant sampling of clinically atypical nerve-sheath tumors in neurofibromatosis patients, the assessment of peripheral nerve biopsies during nerve reconstruction surgery and resection margins most commonly related to the excision of olfactory neuroblastoma. Our largest case to date involved the assessment of 15 separate surgeon-defined resection margins during a salvage resection of a sarcoma post-radiation therapy.

3.3. Slide preparation, interpretation and turnaround times The performance of our RM and VS TP systems in terms of SPT, SIT and TAT is summarized in Table 4. By RM, SPT averaged 10.33 minutes per single block case (range ⫽

170 Table 4

Seminars in Diagnostic Pathology, Vol 26, No 4, November 2009 Time requirements for single-block frozen sections

Telepathology modality Robotic (range) Virtual-Slide (range)

Slide preparation time (minutes) a

10.33 (9-42 ) 12.26* (8-20)

Slide interpretation time (minutes)

Total turnaround time (minutes)

9.65 (⬍1-25) 3.42**b (⬍1-10)

19.98 (11-45) 15.68* (9-35)

a In one case examined by RM TP, a total of 42 minutes elapsed between the receipt of tissue in the surgical pathology lab and the time that the case was available for review by the pathologist. The prolonged slide preparation time was due to a malfunction of the pager carried by the histotechnologist. b The pathologist interpretation time for cases examined by VS TP includes the examination of both frozen section slides and smears. Smears were performed in addition to frozen sections in 30% of cases. The use of smears was a function of the pathologist preference. The average SIT for VS TP cases with only frozen section slides has been 2.41 minutes per slide. *P ⬍ 0.0001 (2-tail, 2-sample Student’s t-test). **P ⬍ 0.000001 (2-tail, 2-sample Student’s t-test).

9 to 42 minutes). This was significantly faster than the 12.26 minutes (range ⫽ 8 to 20 minutes) we have experienced with VS TP. The difference is attributed to the extra steps required to generate a diagnostic image using the VS approach. Specifically, the slide is placed in the scanner for a pre-snap image in order to localize the tissue and assess the adequacy of automatically placed focus points. This process takes approximately 1 minute and is required before a high quality diagnostic scan can be produced. The final scan requires an additional 1-2 minutes at 20x. In contrast, generation of the overview scan and connection of the RM server with the client typically required less than 1 minute. The mean SIT for single block frozen sections using RM was 9.65 minutes per slide (range ⫽ ⬍1 to 25 minutes). This gave an average TAT for RM of 19.98 minutes per single block case (range ⫽ 11 to 45 minutes). We did not record the number of images required to make a diagnosis by RM. Cytology slides such as smears or touch preps were rarely used for the RM cases. Our mean overall SIT for single block cases evaluated by VS TP has been significantly faster at 3.42 minutes (range ⫽ ⬍1 to 10 minutes). Notably, this includes the review of additional smear/touch prep slides in 30% of cases. The mean SIT was 2.25 minutes per slide using VS TP when no cytology slide was requested. Following the introduction of VS technology, our TAT has fallen significantly to 15.68 minutes per single block case (range ⫽ 9 to 35 minutes), due mainly to a 4-fold reduction in SIT. With VS TP, 70% of single slide cases have been reported in ⬍2 minutes and 32% have required ⬍1 minute. In contrast, over 80% of the cases evaluated by RM were associated with a SIT ⬎5 minutes. Frozen sections at the upper ends of the SIT and TAT ranges for both RM and VS TP were typically challenging cases that required consultation with 1 or more colleagues. Since an appreciable period of time was required to review a single slide by RM, we tended to have a pathologist on-site at TWH to report scheduled multi-specimen frozen sections by light microscopy. The 4 fold reduction in SIT with VS TP, has allowed us to review all multiple-block frozen section cases remotely without the need to travel to TWH to preserve acceptable TAT. On occasion, using both RM and VS

TP, we have been able to review multiple surgeon-defined mucosal margins from a single patient by placing up to three carefully labeled frozen sections on a single slide. Given the extra steps required for TP, our TAT at TWH has been slower than what we typically encounter at our TGH and PMH sites where all frozen sections are reviewed by light microscopy. Review of data from a representative 2-month period shows average TAT values for single block frozen sections of 12.10 minutes (range 5-22 minutes) for our PMH site and 14.80 minutes (range 5-30) for our TGH site as compared to 19.98 minutes by RM and 15.68 minutes with VS TP. It is, however, difficult to make strict inter-site comparisons since the types of frozen section cases differ considerably between the three sites.

3.4. Diagnostic accuracy All slides examined by TP are reviewed by light microscopy in conjunction with the formalin-fixed paraffin-embedded sections before a final report is signed out. Review of the intra-operative slides by light microscopy serves as the “truth” diagnosis. Agreement between the TP and “truth” diagnoses constituted an accurate TP assessment. Discrepant TP frozen section diagnoses are reconciled by consultation between the frozen section pathologist and the pathologist issuing the final report. The quality assurance review has been carried out by a pathologist who was not involved in any way with the intra-operative consultation in approximately 60% of our cases. After subtracting deferred cases and frozen sections associated with non-representative sampling, our diagnostic accuracy was 97.5% with RM and has been 98% with VS TP. There has been no significant difference in diagnostic accuracy between these two modalities and the introduction of TP has not been associated with any change in accuracy rates in comparison to historical values at TWH. Discrepant cases were classified according to criteria from the Association of Directors of Anatomic and Surgical Pathology (Hum Path 2006) that have been adopted by the College of American Pathologists for lab accreditation pur-

Evans et al

Primary Frozen Section Diagnosis

poses.32 The majority (85%) of the discrepant cases involved minor interpretive errors on the part of the pathologist(s) providing the frozen section diagnosis. We have had no discrepancies in non-neuropathology cases. The most common interpretive error has involved the misclassification of benign tumors such as the reporting of a schwannoma as a neurofibroma, or ependymoma. By RM, we had one false negative interpretation of a resection margin in an olfactory neuroblastoma case and one misinterpretation of a histiocytic infiltrate as a pleomorphic xanthoastrocytoma. In each case the true diagnosis was more obvious on sections cut from the frozen section tissue following formalin fixation and paraffin embedding. When diagnostic tissue was present on the frozen section slide, we have not had false positive or false negative diagnoses where tumors were misclassified with respect to being benign or malignant and tumor grading has not been a problem. We have had one false negative diagnosis with VS TP from a case with a clinical differential diagnosis of lymphoma or toxoplasmosis in an HIV-positive patient. A 40x scan was requested in this case, given the need to search for infectious organisms. The frozen section showed necrotic brain tissue with no evidence of lymphoma. While we did not identify toxoplasmosis on the scanned frozen section slide, the surgeon was told that we could not exclude that diagnosis. The clinical team made a decision to treat empirically for toxoplasmosis based on clinical and radiologic grounds while awaiting the final pathology report. Paraffin-embedded tissue that had not been examined by frozen section contained the bradyzoites of toxoplasmosis. On review of the frozen section slide by light microscopy, one bradyzoite was identified that was not recognized at the time of frozen section. Review of the 40x virtual slide also revealed the same bradyzoite that was not initially recognized (Figure 3). To date, none of our discrepant cases have had clinical impact and we cannot attribute their occurrence to poor image quality.

171

3.5. Deferral rates and non-representative sampling Frozen sections where a definitive diagnosis could not be established were considered to be deferred cases. These include all instances where a differential diagnosis was given to the surgeon or where terms such as “favor”, “deferred” or “await assessment of paraffin sections” appeared in the frozen section diagnosis. Our deferral rate was 7.8% using RM and has been 7.6% with VS TP. This gives an overall deferral rate of 7.7% with no significant difference between RM and VS TP. These deferrals have most commonly been associated with some degree of non-representative sampling and have not been attributable to overly cautious reporting by TP. A typical example would involve the grading of astrocytic tumors where the clinical and radiological impression was of a high-grade neoplasm and the frozen section tissue, assessed by both TP and light microscopy, showed increased cellularity and cytologic atypia, but no mitoses, endothelial proliferation or necrosis that would support a clear high-grade diagnosis. Frozen sections of mixed glial-neuronal tumors, such as ganglioglioma, where atypical ganglion cells were present without an obvious glial proliferation in the frozen section would constitute another example. By these criteria, non-representative sampling has been a factor in 7.0% of our RM cases and 5.5% of our VS cases. In these situations, the surgeons have either submitted more tissue for frozen section, or have decided to wait for a final diagnosis from tissue submitted directly in formalin.

3.6. Technical issues and episodes of failure: UHN intranet failure has prevented us from using TP on one occasion. Since the network failure occurred well in advance of a scheduled frozen section, a pathologist was

Figure 3 Toxoplasmosis in a brain biopsy from an HIV-positive patient. Panel (A) shows a 400x image of formalin-fixed, paraffinembedded brain tissue with a bradyzoite indicated by the arrow. Panel (B) shows an image taken from the 40x scan of a frozen section performed on a separate biopsy that was examined at the time of surgery. The bradyzoite indicated by the arrow was not recognized at the time of frozen section, but was found on retrospective review of the virtual slide.

172

Seminars in Diagnostic Pathology, Vol 26, No 4, November 2009

easily able to be on site at TWH when the tissue arrived. We did not experience mid-case technical failure with RM that required a pathologist to go to TWH. Excess mounting media would occasionally cause the slide to stick to the robotic stage, preventing the pathologist from navigating that slide. This was corrected by a phone call to the HT asking them to clean and replace the slide. The longest TAT encountered during our use of RM was 45 minutes for a single-block frozen section. This was due to a malfunction of the pager carried by the HT who was not in the laboratory when an unexpected frozen section arrived. On 4 occasions since changing to VS TP, we have had to send a pathologist over to TWH due to technical problems. The first instance was associated with relocating the scanner within the laboratory. The instrument was inadvertently assigned a dynamic IP address, as opposed to the required static IP. We were, thus, temporarily unable to connect with the scanner’s computer. The problem was discovered during a morning system check and a pathologist was able to go to TWH to report scheduled frozen sections without delay while the problem was corrected. The other 3 incidents involved mid-case failures giving a failure rate of 0.5% (3 out of 633 VS TP frozen sections). On two occasions, the biopsies consisted of small (1-2 mm) pieces of edematous brain tissue that gave rise to small, pale frozen sections that were not recognized by the scanner. As a result, we were unable to generate virtual slides and a pathologist went over to TWH to report the cases by light microscopy with a 10-15 minute delay in TAT. Following the consultation with the vendor, we were able to make adjustments to the scanner settings to prevent further failures of this nature. On one other occasion, excess mounting media on a slide fouled the scanner objective and the cover slip lifted off of the slide. The fouled objective required a thorough cleaning a pathologist went to TWH to report the case by light microscopy. This event underscored the need for meticulous attention to removing excess mounting media prior to placing slides in the scanner. With VS TP we have requested repeat scanning in a total of 12 cases (2% of 633), most of which were due to poor quality frozen sections associated with folding of tissue sections or prominent freezing artifact. Only 5 cases (0.8%) have had to be re-scanned because the initial slides were not well focused in the opinion of the pathologist. The delays associated with re-scanning have been minimal and have not resulted in a pathologist going to TWH to report the case by conventional light microscopy.

3.7. Evaluation of intra-operative cytology Touch preps or smears were not often requested during our use of RM due, largely, to pathologist preference. There was a general feeling that the evaluation of cytology slides by RM was too time consuming and provided little additional diagnostic value. Cytology slides have been used in 30% of the VS cases, due mainly to the addition of pathologists to the TWH roster who prefer the joint use of cytol-

ogy and histology for neuropathology cases. We have found that examining cytology is more efficient with the VS system, although we have not recorded the number of cases for which cytology was essential to make a diagnosis. The inability to adjust fine focus to account for depth of field differences has not been a problem with VS TP, as wellfocused areas of interest have always been identifiable somewhere on the cytology slides.

3.8. User satisfaction It is difficult to objectively assess user satisfaction. Retrospectively, the length of time required to thoroughly review a slide by RM was a source of frustration for all pathologists in comparison to VS TP. The significant reduction in SIT and improved image quality offered by the VS system were the main reasons we were able to transition away from RM after a validation period of only 1 month.

3.9. Virtual slide file sizes At 20x, the file sizes have averaged 35.7 MB/slide (range 3.7-180.0) and 132.2 MB/slide (range 23.0-293.6) for frozen sections and smear/touch prep slides respectively. It is worth noting that file sizes and scan times increase by a factor of 3-4 if 40x scans are used. It has been our experience that the screen re-fresh rate when moving between fields, or moving to higher magnification, is noticeably longer with 40x scans. For our purposes, the increased magnification offered by a 40x scan is generally not required to make frozen section diagnoses (Figure 4). We have concluded that the routine use of 40x scans would only increase our TAT and possibly create file storage issues over time.

4. Discussion 4.1. Why use telepathology at University Health Network? The decision to implement TP at UHN was made in order to improve the efficiency of pathologists who work in a consolidated department in a multi-site teaching hospital. TP has allowed us to provide an essential service while at the same time removing the disruptions in daily workflow that would come with going to the TWH site. Specifically, one could spend up to 2 hours on a single frozen section due to travel and a lack of being able predict when the surgeon would actually send the specimen. Our TP frozen section program is the first of its kind in Canada and, to the best of our knowledge, is the first to use VS technology for primary frozen section diagnosis. TP has been ideally suited to our needs. We have used it to cover a relatively small daily volume of frozen sections that are well spaced throughout

Evans et al

Primary Frozen Section Diagnosis

173

Figure 4 Images from a typical 20x scan VS TP frozen section (A) and its accompanying smear (B). A meningioma with no atypical features is shown in this example. It has been our experience that 20x scans are more than adequate for determining tumor type and histological grade in frozen sections.

the course of a given day. The specimens are small, single pieces of tissue that can be processed without gross review by a pathologist. We have dedicated HT’s who consistently provide high quality frozen section slides and a core group of pathologists who are committed to making the program a success. Finally, we are close enough to TWH that a pathologist can be on-site within 10-15 minutes in the rare event of a system failure. We acknowledge that TP may not be practical in situations where multiple frozen section specimens regularly arrive at the same time or where detailed gross examination by a pathologist is an absolute requirement. Concerns over image quality, diagnostic accuracy and TAT will likely continue to act as barriers to more widespread use of TP for frozen sections. These concerns are not trivial as frozen section interpretation can be one of the most stressful activities in surgical pathology. Aside from pressure to make rapid diagnoses, frozen section morphology can be suboptimal and there is the ever-present possibility that an incorrect diagnosis will lead to inappropriate intra-operative management that could ultimately be blamed on a pathologist. It has been our experience that TP, particularly VS TP, has been not been unacceptably slow or inaccurate. We have been able to maintain a diagnostic accuracy of 98% with TAT values that satisfy College of American Pathologists (CAP) laboratory accreditation criteria.33

4.2. Comparisons with other telepathology frozen section programs Comparisons of accuracy and TAT for frozen sections between centers must consider differences in the types of frozen sections performed in each center.34 The vast majority of our TP frozen sections are biopsies submitted by neurosurgeons to establish a diagnosis of an unknown pathological process. As such, it would not be appropriate to compare our performance with TP programs that have looked at resection margins, breast biopsies or lymph nodes. There are, however, two published reports on the use of TP for primary frozen section diagnosis in neuropathology that are very similar to our situation.7,14 Hutarew et al14 reported

the use of RM for 343 frozen sections between 2002 and 2005 in Salzburg, Austria. The hospitals were separated by 2 km and a laboratory technician managed the peripheral TP station. Their average TAT was 26.1 minutes with an average time for histological interpretation of 10.7 minutes per case. They reported a diagnostic accuracy of 97.9% and they were able to render a diagnosis in 100% of cases, despite minor technical problems. These results are virtually identical to ours for the 350 cases we reported by RM. Horbinski et al7 reported their experience with 402 TP frozen sections between 2002 and 2006. Eighteen city blocks in Pittsburgh, PA separate the hospitals in their situation. They compared 40 cases assessed by non-robotic, static/dynamic microscopy (where an on-site pathologist provided images to a consultant) to 362 cases done by RM (with no on-site pathologist) and 1227 cases reported by conventional light microscopy. They reported diagnostic discrepancy rates for TP that varied between 2.4% and 5.8% over the 4-year period and they noted no difference in diagnostic accuracy between their different TP modalities. As with our findings, most of their discrepant cases involved minor misclassifications related to tumor type and problems concerning tumor grading were not identified. Their average annual deferral rate was 10.6%. They concluded that with increasing experience the diagnostic accuracy and deferral rates for TP approached those of conventional microscopy. They specifically noted that they were not needlessly cautious when using TP and that the kinds of cases with deferred or discrepant diagnoses on TP were similar to those in their light microscopy series. While they did not report data for TAT and SIT, they were left with the impression that they spent less overall time on remote intra-operative consultations by using TP. The encouraging accuracy rates that we have observed, along with those reported by Hutarew et al14 and Horbinski et al,7 are essentially no different than the 2.7% discrepancy rate reported by Plesec and Prayson for 2156 neuropathology frozen sections evaluated by conventional light microscopy.35 Plesec and Prayson also found that misclassifications related to tumor type were the most common reason for discrepant frozen section diagnoses with light micros-

174

Seminars in Diagnostic Pathology, Vol 26, No 4, November 2009

copy. Since this observation mirrors our findings and those of Horbinski et al7 with TP, it is our contention that TP will not predispose pathologists to make interpretive errors that would have been avoided with conventional light microscopy.

4.3. Technical issues to consider Planning for technical failure is a critical issue that must be taken into account when implementing a TP frozen section service. The fact that we have experienced mid-case technical failure, albeit in only 3 of 983 frozen sections (0.3%), to underscore this point. It would also have been unrealistic for us to think that we would never encounter such problems. Being only 1 mile from TWH, however, has allowed us the option of going over to report these rare cases with a delay of only 10-15 minutes. It is clear that measures of redundancy, with their associated costs, would be required in situations where timely travel to the remote site is not an option.

4.4. Virtual slides-validation studies and benefits realized at University Health Network VS technology has the potential to revolutionize TP. While we have found it to be ideal for a specific purpose at UHN, it nonetheless requires some fine tuning and validation studies covering the complete spectrum of surgical pathology still need to be to be performed before pathologists can begin replacing their microscopes with computer screens and slide scanners. Validation studies performed to date with selected surgical pathology cases have certainly provided encouraging results. Tsuchihashi et al recently published the results of small pilot study involving the use of VS TP for 15 frozen sections in Kyoto Japan.31 Information concerning the types of cases was not provided, although the authors stated that all 15 cases were correctly diagnosed following review of only 10x scans with a SIT of around 5 minutes. The study of Gilbertson et al28 involved 3 pathologists making VS diagnoses for 25 cases (31 total parts) in dermatopathology and genitourinary pathology. Of 93 total diagnostic events, there were 12 discrepancies among the VS diagnoses. Three of these discrepancies were felt to be at least partially attributable to image quality. These investigators noted that there were poorly focused areas on every VS case and they cautioned readers against concluding that VS technology was as good as a microscope. It is worth noting, however, that even with these imperfections the consensus VS diagnoses showed 100% agreement with the light microscopy diagnoses.28 The ability to obtain rapid consultation on challenging cases is, in our view, a distinct advantage that our VS system holds over light microscopy. It is our opinion that this will contribute to improved patient care in the longterm. At UHN, TP has facilitated consultation on frozen sections that could not easily have taken place with a single

pathologist on-site at TWH. Two pathologists jointly reported most frozen sections when RM was in use. This was seen as a time saving measure in case a second pathologist had to walk to the client station to assist a colleague. Two or more colleagues have reviewed approximately 30% of VS TP cases. Our system allows several pathologists to review a frozen section slide simultaneously from their own offices and discuss them over the telephone with minimal delay in TAT. It has been our practice to be prudent and defer cases where definitive diagnoses cannot be made based on the sampled material, such as distinguishing between low-grade astrocytoma and reactive gliosis. Consultation with a colleague is virtually always sought before a deferred diagnosis is given to the surgeon.

4.5. Retention of frozen section virtual slides The glass slide will constitute the official record of a given frozen section for the foreseeable future. No guidelines currently exist concerning the retention of images obtained from TP frozen sections. With our RM system, the slide overviews were automatically archived on the server computer. We did not capture images from frozen section slides showing normal tissue. In cases where a pathological diagnosis was made, representative images to support that diagnosis were obtained and saved. We have retained all files from our VS TP cases in order to prepare a virtual slide library of neuropathology frozen sections for research as well as the training of residents and fellows. Our SOP calls for cases to be removed from the scanner’s computer and backed up on 2 external hard drives every 3 months. Figure 5 shows a scheme that we suggest as a practical approach to retaining VS files. We suggest there is no need to retain images from cases where there was complete agreement between TP diagnoses and the quality assurance by light microscopy, unless they are deemed to be of educational value. Virtual slides from all discrepant cases should be retained, and those from major discrepancies where there was an impact on patient care should be retained indefinitely. Cases with minor discrepancies could also be saved indefinitely for teaching purposes. The VS archive could be reviewed on an annual basis in order to determine which cases should continue to be stored. There is no question that formal guidelines on image retention must be developed as we move into the age of digital pathology.

5. Conclusion It has been our experience that TP with either RM or VS technology is safe, reliable and accurate for making primary frozen section diagnoses. Effective communication and planning is required at several levels in order for a TP frozen section program to function smoothly. While we cannot go so far as to conclude that VS TP is as good as a microscope, we have found it to be superior to RM in terms

Evans et al

Primary Frozen Section Diagnosis

Figure 5 This diagram summarizes possible guidelines that could be used for the retention of VS frozen section files. Cases for which there is complete agreement between the TP diagnosis and the quality assurance (QA) by light microscopy could reasonably be deleted from the archive 6 weeks after the final report has been signed out. Cases with discrepant TP diagnoses should be retained. Files from cases with major discrepancies associated with an impact on patient care should be retained indefinitely. Those from cases with minor discrepancies could be retained for 1 year after the final report has been signed out. The need for continued storage could be reviewed after that time. Alternately, all cases with perceived educational value could be archived if sufficient storage capacity exists. The VS archive could be reviewed annually in order to remove any unnecessary files. We have retained all of VS TP files for teaching purposes to this point in time.

of TAT and user satisfaction. Regional variation in the supply of pathologists and a desire to consolidate subspecialty pathologist services are major forces that will drive the implementation of TP in other Canadian centers. There is little doubt that VS technology will play a critical role in this process.

References 1. Weinstein RS, Bloom KJ, Rozek LS: Telepathology and the networking of pathology diagnostic services. Arch Pathol Lab Med 111:646652, 1987 2. Weinstein RS: Telepathology comes of age in Norway. Hum Pathol 22:511-513, 1991 3. Dunn BE, Choi H, Almagro UA, Recla DL: Combined robotic and nonrobotic telepathology as an integral service component of a geographically dispersed laboratory network. Hum Pathol 32:1300-1303, 2001 4. Dunn BE, Choi H, Almagro UA, Recla DL, Davis CW: Telepathology networking in VISN-12 of the Veterans Health Administration. Telemed J E Health 6:349-354, 2000 5. Dunn BE, Choi H, Almagro UA, Recla DL, Krupinski EA, Weinstein RS: Routine surgical telepathology in the Department of Veterans Affairs: experience-related improvements in pathologist performance in 2200 cases. Telemed J 5:323-337, 1999 6. Dunn BE, Almagro UA, Choi H, Recla DL, Weinstein RS: Use of telepathology for routine surgical pathology review in a test bed in the Department of Veterans Affairs. Telemed J 3:1-10, 1997

175 7. Horbinski C, Fine JL, Medina-Flores R, Yagi Y, Wiley CA: Telepathology for intraoperative neuropathologic consultations at an academic medical center: a 5-year report. J Neuropathol Exp Neurol 66:750-759, 2007 8. Liang WY, Hsu CY, Lai CR, Ho DM, Chiang IJ: Low-cost telepathology system for intraoperative frozen-section consultation: our experience and review of the literature. Hum Pathol 39:56-62, 2008 9. Nordrum I, Engum B, Rinde E, Finseth A, Ericsson H, Kearney M, Stalsberg H, Eide TJ: Remote frozen section service: a telepathology project in northern Norway. Hum Pathol 22:514-518, 1991 10. Oberholzer M, Fischer HR, Christen H, Gerber S, Bruhlmann M, Mihatsch MJ, Gahm T, Famos M, Winkler C, Fehr P: Telepathology: frozen section diagnosis at a distance. Virchows Arch 426:3-9, 1995 11. Steffen B, Gianom D, Winkler C, Hosch HJ, Oberholzer M, Famos M: [Frozen section diagnosis using telepathology]. Swiss Surg 3:25-29, 1997 12. Della Mea V, Cataldi P, Pertoldi B, Beltrami CA: Combining dynamic and static robotic telepathology: a report on 184 consecutive cases of frozen sections, histology and cytology. Anal Cell Pathol 20:33-39, 2000 13. Dawson PJ, Johnson JG, Edgemon LJ, Brand CR, Hall E, Van Buskirk GF: Outpatient frozen sections by telepathology in a Veterans Administration medical center. Hum Pathol 31:786-788, 2000 14. Hutarew G, Schlicker HU, Idriceanu C, Strasser F, Dietze O: Four years experience with teleneuropathology. J Telemed Telecare 12:387391, 2006 15. Terpe HJ, Muller W, Liese A, Vogel CU, Broer KH: [Frozen section telepathology in the clinical routine of a breast cancer center]. Pathologe 24:150-153, 2003 16. Sukal SA, Busam KJ, Nehal KS: Clinical application of dynamic telepathology in Mohs surgery. Dermatol Surg 31:1700-1703, 2005 17. Hutarew G, Dandachi N, Strasser F, Prokop E, Dietze O: Two-year evaluation of telepathology. J Telemed Telecare 9:194-199, 2003 18. Hitchcock CL, Hitchcock LE: Three years of experience with routine use of telepathology in assessment of excisional and aspirate biopsies of breast lesions. Croat Med J 46:449-457, 2005 19. Weinstein RS, Descour MR, Liang C, Bhattacharyya AK, Graham AR, Davis JR, Scott KM, Richter L, Krupinski EA, Szymus J, Kayser K, Dunn BE: Telepathology overview: from concept to implementation. Hum Pathol 32:1283-1299, 2001 20. Winokur TS, McClellan S, Siegal GP, Redden D, Gore P, Lazenby A, Reddy V, Listinsky CM, Conner DA, Goldman J, Grimes G, Vaughn G, McDonald JM: A prospective trial of telepathology for intraoperative consultation (frozen sections). Hum Pathol 31:781-785, 2000 21. Kaplan KJ, Burgess JR, Sandberg GD, Myers CP, Bigott TR, Greenspan RB: Use of robotic telepathology for frozen-section diagnosis: a retrospective trial of a telepathology system for intraoperative consultation. Mod Pathol 15:1197-1204, 2002 22. Moser PL, Lorenz IH, Sogner P, Stadlmann S, Mikuz G, Kolbitsch C: The accuracy of telediagnosis of frozen sections is inferior to that of conventional diagnosis of frozen sections and paraffin-embedded sections. J Telemed Telecare 9:130-134, 2003 23. Frierson HF, Jr., Galgano MT: Frozen-section diagnosis by wireless telepathology and ultra portable computer: use in pathology resident/ faculty consultation. Hum Pathol 38:1330-1334, 2007 24. Demichelis F, Barbareschi M, Boi S, Clemente C, Dalla PP, Eccher C, Forti S: Robotic telepathology for intraoperative remote diagnosis using a still-imaging-based system. Am J Clin Pathol 116:744-752, 2001 25. Gephardt GN, Zarbo RJ: Interinstitutional comparison of frozen section consultations. A college of American Pathologists Q-Probes study of 90,538 cases in 461 institutions. Arch Pathol Lab Med 120:804-809, 1996 26. Crimmins D, Crooks D, Pickles A, Morris K: [Use of telepathology to provide rapid diagnosis of neurosurgical specimens]. Neurochirurgie 51:84-88, 2005 27. Brauchli K, Jagilly R, Oberli H, Kunze KD, Phillips G, Hurwitz N, Oberholzer M: Telepathology on the Solomon Islands–two years’

176

28.

29.

30.

31.

Seminars in Diagnostic Pathology, Vol 26, No 4, November 2009

experience with a hybrid Web- and email-based telepathology system. J Telemed Telecare 10 Suppl 1:14-17, 2004 Gilbertson JR, Ho J, Anthony L, Jukic DM, Yagi Y, Parwani AV: Primary histologic diagnosis using automated whole slide imaging: a validation study. BMC Clin Pathol 6:4, 2006 Fine JL, Grzybicki DM, Silowash R, Ho J, Gilbertson JR, Anthony L, Wilson R, Parwani AV, Bastacky SI, Epstein JI, Jukic DM: Evaluation of whole slide image immunohistochemistry interpretation in challenging prostate needle biopsies. Hum Pathol 39:564572, 2008 Ho J, Parwani AV, Jukic DM, Yagi Y, Anthony L, Gilbertson JR: Use of whole slide imaging in surgical pathology quality assurance: design and pilot validation studies. Hum Pathol 37:322-331, 2006 Tsuchihashi Y, Takamatsu T, Hashimoto Y, Takashima T, Nakano K, Fujita S: Use of virtual slide system for quick frozen intra-operative

32.

33.

34. 35.

telepathology diagnosis in Kyoto, Japan. Diagn Pathol 3 Suppl 1:S6, 2008 Association Of Directors Of Anatomic And Surgical Pathology, Nakhleh R, Coffin C, Cooper K: Recommendations for quality assurance and improvement in surgical and autopsy pathology. Hum Pathol 37:985-988, 2006 Novis DA, Zarbo RJ: Interinstitutional comparison of frozen section turnaround time. A College of American Pathologists Q-Probes study of 32868 frozen sections in 700 hospitals. Arch Pathol Lab Med 121:559-567, 1997 Sawady J, Berner JJ, Siegler EE: Accuracy of and reasons for frozen sections: a correlative, retrospective study. Hum Pathol 19:1019-1023, 1988 Plesec TP, Prayson RA: Frozen section discrepancy in the evaluation of central nervous system tumors. Arch Pathol Lab Med 131:15321540, 2007