Introduction Whole slide imaging is the process of digitizing glass slides and the creation of Whole Slide Images (WSI), which enable the examination of pathology samples on a computer screen in a manner comparable to light microscopy. WSI have been used for different applications in pathology but their use for primary diagnostics is still limited. Implementing WSI for primary diagnostics would be a turning point necessitating extensive validation to unravel pitfalls and difficulties that could be encountered within the routine workflow. This article is aimed to describe the gradual integration of WSI into routine pathology diagnostics in a medium-sized routine pathology laboratory.
Methods This project was started with optimizing the digital work environment including the setting up of validation studies, scanning preferences, storing WSI and the implemented adjustments to the workflow for the laboratory and the pathologist. Afterwards scanning glass slides was initiated in the department of pathology at the Atrium Medical Center, Heerlen, The Netherlands, for performing primary diagnostics of breast biopsies. Later this was extended to other specimen types including resections.
Results The validation studies yielded a high concordance rate between WSI and conventional diagnoses. Routine primary WSI based diagnosis was possible in 82.1% of cases. Failure of digital diagnosis was mainly related to poor image quality and logistic problems.
Conclusion The quality of the currently produced WSI is sufficient for primary diagnostics in 82.1% of the cases. Improving image quality, adequate retrieval and controlling scanning error will definitely encourage the wide adaptation in routine diagnostics.
- Digital Pathology
Statistics from Altmetric.com
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.
Looking at glass slides through the conventional microscope has been the standard way of working for diagnostic histopathology and cytopathology for a long time.1 Over the last decades, new methods for rendering diagnosis have emerged for pathology practice.2 Rendering diagnosis depending on static or dynamic images transferred through network connections to remote places for second opinion and teleconsultation, commonly called telepathology, is widely accepted nowadays.3 These systems have been followed by more sophisticated methods of image acquisition called whole slide imaging (or virtual microscopy) where the whole glass slide is converted into a digital form (whole slide images, WSI) allowing the examination of pathology specimen on a computer display with the aid of an image viewer.4 ,5 Scanners dedicated to scan glass slides and the creation of WSI became available more than a decade ago.6 ,7 WSI combine the features of both the previous telepathology systems (static and dynamic), providing high resolution images with unlimited access to the entire pathology specimen at different magnifications.8 ,9 Remote access of WSI by different examiners from different places at any time is unique for WSI, which supports their use for different applications in pathology, especially for teleconsultation,10 ,11 telerevision and education.12–14 Examination of multiple digital slides simultaneously is also a property supplied by the image viewer, which allows comparison of different stains and sections. The above-mentioned features support the use of WSI for teleconsultation, frozen section diagnosis, clinical conferences, research and image and daily routine diagnostics, although WSI have not widely been applied for this purpose until now.
A major bottleneck in hindering the use of WSI in diagnostics is the time needed for scanning and exploring WSI. Moreover, the integration of WSI into daily routine practice is also accompanied by many challenges regarding workflow in the lab and for the pathologist. The presence of scanners with a more acceptable scanning speed (2–4 min per slide on 20× magnification) improved image quality and the ongoing reduction of storage costs encouraged some medical institutes to perform WSI scanning on a daily basis, either for the entire routine work or for selective cases. In The Netherlands, scanning the complete daily production of histopathology specimens to build up a digital archive is being done by the University Medical Center Utrecht only.7 Scanned slides are used for clinico-pathological conferences, revision, consultation, teaching and research but not yet for daily routine diagnostics. Scanning of a fixed number of cases on daily bases aimed for performing primary diagnostics of histopathological specimens has been implemented in the Atrium Medical Center, Heerlen (AMCH). According to our knowledge the AMCH is the first hospital in The Netherlands that uses WSI for primary diagnostics in pathology. The aim of this article is to share the experience in setting up primary WSI based diagnosis in a medium-sized routine pathology lab.
Materials and methods
The department of pathology of the AMCH handles about 21 000 histology and 16 000 cytology specimens per year. In 2006 it was decided to start a project on digital pathology, with one of the goals to use WSI for primary diagnostics. At the time the project started there were scanners that would be able to handle a sufficient part of the workload of the AMCH with adequate quality to start a pilot to validate WSI for primary diagnostics.
For the scanner, criteria were formulated regarding scanning speed, loading capacity and price. After testing the available scanners in the market at that time (from Aperio, Zeiss, Hamamatsu and Olympus), the Mirax Scan (3DHistech, Budapest, Hungary) was selected as the most suitable scanner to meet the goals of this project. This scanner was able to load and automatically scan 150 slides in one run, divided over multiple cassettes. Scanning standard glass slides at 20× took between 2.25 and 6.45 min depending on the size of the specimen.15 After registration during several weeks of the objectives used with conventional microscopy, pathologists agreed that 20× would be adequate for diagnosing most cases. The slides were labelled with a 2D barcode, containing the specimen number, to allow easy image retrieval and a smooth workflow. WSI were stored on external hard disks with a total capacity of 750 gigabyte accessible over the network. The images belonging to one specimen were grouped in folders with the specimen number based on the barcode.
Preliminary experiments revealed that several steps had to be optimised to arrive at proper scans for digital diagnosis. This concerned the size of the section that had to fit underneath the coverglass. For this, dissection was modified and technicians were instructed to extra carefully position the sections centrally on the glass slides. Section thickness was standardised at 4 µm. The processing of small biopsies was also changed from putting three consecutive sections from the same level on three rows to putting three sections from different levels on one row when it was noticed that additional information came from additional levels rather than from additional sections at the same level.
During digital diagnosis making, pathologists were asked to note for each case if it was possible to make the diagnosis based on the WSI. When unable to make the diagnosis on the digital image, they noted the cause of diagnostic difficulty (image quality, technical problems or other causes).
Diagnostic validation studies
The first retrospective validation study concerned digital re-diagnosis of 100 breast needle core biopsies that had been routinely diagnosed conventionally. Five pathologists from three different countries (Greece, Anna Batistatou; Hungary, Janina Kulka; and The Netherlands, Marius Nap, Nathalie Van de Vijver and Paul Theunissen) participated in this study. The degree of agreement between the five pathologists was calculated using the κ statistic.
The second prospective validation study concerned the diagnosis of 85 cases from different body systems by WSI and conventional light microscopy. During a period of several months, about five cases were scanned and digitally diagnosed every week. Each time, the participating pathologists were asked to render the diagnosis on WSI only. Afterwards, another pathologist was asked to re-diagnose these cases by light microscopy blinded to the WSI based diagnosis. If the WSI diagnosis matched with the light microscopy based diagnosis, the report was signed out immediately. If not, the report was adjusted based on the information derived from light microscopy diagnosis. In this way, the pathologists could closely monitor to which extent WSI can be used for primary diagnostics. At the end of the study, the agreement between WSI based and conventional diagnoses was assessed.
Implementation of WSI in the routine pathology diagnostic workflow
Scanning and rendering diagnosis on WSI for all breast needle core biopsy specimens were initiated at the end of 2007. Afterwards, it was decided to dedicate one day per week for digital diagnosis of all tissue biopsies and resection specimens. Since four pathologists participated in this project, each pathologist performed digital diagnosis for all his routine work one day per month. Pathologists were supplied with the usual clinical information and were free to request additional histochemical- or immunostains. From January 2009 onwards, on average eight cases per day were diagnosed digitally by two pathologists who enjoyed doing diagnostics digitally. The number of scans per year from 2007 to 2010 is illustrated in table 1.
Retrospective WSI validation study
Comparing the light microscopy based diagnosis and WSI based diagnosis of the 100 breast needle core biopsies yielded a very high agreement between the five pathologists with a κ score of 0.97.
Prospective WSI validation study
The results were comparable to the retrospective study with almost perfect agreement between the light microscopic and WSI based diagnosis as published on the website of the department at the time (Year report 2009).
Unfortunately, at the time these experiments were done, we had no intention to use this for publication and the results of this validation were not stored for formal statistics. Repeating the experiment at a later stage would not be logical since the conditions of individual experience have changed too much to give a realistic impression.
WSI based diagnosis in routine workflow
The total number of scanned cases for primary diagnostics was 3923 in 2010, from which 3222 cases were completely digitally diagnosed and 701 cases were not digitally signed out. Table 2 shows the number of scanned cases per pathologist sorted by the diagnostic modality. There were four major causes for failing digital diagnosis (see table 3):
Scanning problems: unsharp images and incomplete scanning.
Logistic problems: scan could not be located, network problems and scans were not available in time to fit in the routine work schedule.
Technical problems: bad (H&E) staining, bad positioning and tissue folding.
Extra procedures required: additional sections, immunohistochemical stains or internal consultation. Table 4 summarises the exact causes of extra procedures.
The major cause of failing digital diagnosis (table 3) was related to image quality (N=209), with completely or partly unsharp WSI or incompletely scanned tissue, followed by logistic problems (N=61), extra procedures required (N=56) and (laboratory) technical problems (N=49); for the other cases (N=326), the cause remained unknown where the pathologist did not specify the causes of case deferral and non-digital diagnosis.
Thus, primary WSI based diagnosis was possible in 82.1% of the scanned cases where the estimated 99% CI is 0.805 to 0.836.
The aim of this article is to share the experience of using WSI for primary diagnostics in a medium routine pathology practice. The integration of WSI in the routine workflow was performed in a stepwise manner starting with minor adjustments of specimen handling, followed by two validation studies, finally resulting in implementing primary WSI based diagnostics for part of the routine work.
Studying the validity of WSI was a crucial point. The results of the two validation studies were very promising and encouraged the pathologists to start with primary WSI based diagnostics as part of the routine work. This was applied at first for breast needle core biopsies and later extended to other body systems. The gradual introduction of WSI based diagnosis in the daily routine was a very important step that revealed difficulties and problems associated with WSI based diagnostics, and allowed for timely finding solutions.
This trial showed that about 82.1% of scanned cases could be digitally signed out successfully. The main reasons for not being able to sign out the remaining cases were image quality, followed by logistical and technical problems. Image quality is expected to improve when focusing algorithms become faster and better, allowing for focusing on may be every pixel instead of a few fields scattered over the section. Another aspect that we had not expected to influence the process was the variation of thickness of the object glass. Unpredictable solitary or groups of slides appeared to have a thickness just under or above the tolerance of the automatic focusing range of the scanner. Although this variation could not be managed by the glass slide manufacturer, in newer versions of the hardware this problem has been solved. Some adjustment to the laboratory work, especially central placing of the tissue on the glass slides, will help to avoid incomplete scans. Also in tissues with low optical density the parameters for scanning can be adjusted, resulting in a higher sensitivity and complete images. Guaranteed network access and speed will probably require available Information technology (IT) support, which may not be easy to arrange for smaller laboratories. For a better flow of the routine work, WSI should be stored in an appropriate way so that the pathologist will be able to access them without delay preferably linked to the pathology report and stored in a sustainable digital archive for later referral if needed. The current work approach is that the pathologist accesses over the network the WSI stored on an external hard disk on which digital slides from the same case are stored in one a folder carrying the case number. For the time being, the existing storage system and accessibility are adequate for the department work. Nevertheless, there were about 61 cases with logistic problems where the pathologist could not locate the WSI or there were network problems which were responsible for case deferral. Performing scanning of all the tissue specimens on a daily basis would require a larger storage environment and easier access (eg, by using a storage area network). To this end, linking WSI to the pathology report would be of great advantage. It will not be necessary for the pathologist to open different files (the current work approach) to access WSI but can smoothly open the same pathology database where the patient history, macroscopic images and WSI are stored together. Linking of WSI with the patient history and pathology report has been successfully implemented in the University Medical Center Utrecht, which facilitates image retrieval for revision, clinico-pathological conferences and research.7 The same or very similar approach will be adopted in AMCH in the near future to overcome logistic problems.
Some technical laboratory errors and the need of extra stain were responsible for case deferral in 105 cases. The same could be encountered if the diagnosis was performed by glass slides and conventional microscope. Rescanning of glass slides with technical errors would be an option to enrich and complete the department's digital archive. However, this would not serve the primary goal of having WSI available in a timely and complete manner to perform WSI based routine diagnostic analysis in a medium-sized routine pathology lab. Most of the cases which required extra procedure were from skin lesions where the pathologists need extra information to complete the diagnosis by asking for extra stain, deeper sections, internal consult or even asking for the glass slide when the digital image information does not satisfy the diagnostic needs. In this study, we have only four cases where the presence of glass slide was necessary to complete the diagnosis; two of them to check the double refraction in gingival and skin infection as this feature cannot be supported in WSI. Perhaps the easiest category to start primary digital diagnosis with may be Gastro-intestinal (GI) and breast biopsies.16 ,17 Table 4 details these cases.
There was a remarkable variation (27%–90%) between the four pathologists in successfully diagnosing cases digitally. Although information as to the exact reasons for deferring cases was incomplete, this shows that adaptation of digital diagnosis may differ between pathologists when starting routine digital diagnosis.
Scanning of the complete daily production of pathology routine work will become possible shortly with the introduction of scanners which are able to scan standard glass slides at 40× in less than 1 min. Primary diagnostics on WSI will facilitate pathology routine workflow through easy image sharing and retrieval and it will not be time consuming anymore to ask for a second opinion for difficult cases. Storing microscopic information in a digital from has also many advantages over storing physical slides since WSI can be stored permanently with constant quality. These images can be used for different applications such as teaching, research and revision. Performing quality assurance (QA) and teleconsultation based on WSI is less time consuming obviating the time needed to send glass slides to far places. This will be more efficient if a national storage or image exchange facility would be present where pathologists could look up digital slides from different institutes for teleconsultation and telerevision. Furthermore, WSI can be used for image analysis which will likely improve diagnostic accuracy and productivity.9 ,18
Primary diagnostics based solely on WSI was applied also in Kalmar County Hospital, Sweden. In this hospital, digitisation of the whole daily production of histopathological specimens was initiated around the same time as in AMCH and has resulted in a situation where 75% of the diagnostics are performed digitally in addition to teleconsultations for frozen sections from other hospitals without local pathologists. Assessing the feasibility of primary WSI based diagnostics is ongoing currently in the University of Pittsburgh Medical Center, USA,5 University Medical Center Utrecht, The Netherlands16 ,17 ,19 and University Medical Center Nijmegen in collaboration with AMCH (manuscript in preparation). In addition, the Digital Pathology Association has presented two papers at the Pathology Vision conference 2011 discussing the high level validation approach of WSI for primary diagnostics. The continuous efforts aimed for validating WSI for primary diagnostics will help eventually in wide use and acceptance of WSI for this purpose. A recent review (via Medscape) has discussed the legal aspects of primary diagnostics using WSI. In this review, it was stated that according to the available documentation it would be sufficient to have local descriptions of the workflow process and to monitor the performance of the individual pathologists, an approach very similar to that for monitoring the quality of conventional microscopy.
QA and quality control have not been worked out for primary digital pathology diagnosis, but are clearly an important issue. Guidelines are, for example, needed as to the desired resolution, completeness of the scans, colour depth, compression ratios, quality of focus, and duration and quality of storage. When starting primary digital diagnosis, it seems wise to have an initial period of full QA of about 50 cases in major diagnostic areas for every pathologist, and then to do random QA in about every 50th case.
In conclusion, the quality of the currently produced WSI is sufficient for primary diagnostics in histopathology. Large scale scanning of the whole daily routine production in histopathology should be accompanied by some modifications regarding adequate and timely image retrieval and controlling scanning errors. Solving these issues is necessary before complete replacement of glass slides within pathology work could be achieved. Still, with the currently available technology, it seems that at least 82.1% of routine pathology can be signed out digitally.
The quality of the currently produced whole slide images (WSI) is sufficient for primary diagnostics in routine histopathology for about 82.1% of cases.
Primary diagnostics on WSI will facilitate pathology routine workflow through easy image sharing and retrieval and it will cost less time to ask for a second opinion for difficult cases.
Linking of WSI into the pathology report will facilitate image retrieval for revision, clinico-pathological conferences, quality assurance and research.
Improving image quality and the presence of scanners which are able to scan glass slides at higher magnification in less time will definitely encourage the wider use of WSI for primary diagnostics.
We would like to thank Rob Teunissen from the Atrium Medical Center, Heerlen (AMCH) for helping us in understanding the set-up of digital pathology at the AMCH and supplying us with the necessary data. We also thank Anna Batistatou, from Greece, Janina Kulka from Hungary and Nathalie Van de vijver and Paul Theunissen from The Netherlands for participating in the first validation study.
Contributors The two validation studies described in the manuscript were mainly formulated and performed by MN and RC, and they registered the data. R Teunissen was involved in developing the software to register these data and to interpret data based on the set-up of digital pathology at the AMCH. He made exports of relevant data for further analysis. Data aggregation and analysis and also the main formulation of the manuscript were performed in the UMCU by SA-J, AH and PJvD. Afterwards, this article has been reviewed and agreed on by all the participating authors. All the authors have substantially contributed in writing, reading and approving the final manuscript.
Competing interests None.
Ethics approval This study was performed on tissue biopsies and resections from human beings. No institutional approval was needed.
Provenance and peer review Not commissioned; externally peer reviewed.