Context.—Digital whole slide imaging is the anticipated future of anatomic pathology, where sign-out of glass slides will be replaced by scanned images. Whole slide imaging has been successfully used in surgical pathology, but its usefulness and clinical application have been limited in cytology for several reasons, including lack of availability of z-axis depth focusing and large file size. Recently, several systems have become available in the United States for whole slide imaging with z-axis technology.
Objective.—To determine the accuracy and efficiency of whole slide imaging, as compared with traditional glass slides, for use in cervicovaginal diagnostic cytology.
Design.—Eleven cervicovaginal cytology cases (ThinPrep and SurePath) scanned at ×20, ×40, and ×40 z-stack magnifications using the BioImagene iScan Coreo Au 3.0 scanner were evaluated by 4 cytotechnologists and 3 pathologists in a blinded study. Different magnification scans were recorded as separate cases and presented in a randomized sequence. Corresponding glass slides were also reviewed. For each case, the diagnoses and total time to reach each diagnosis were recorded.
Results.—Diagnostic accuracy was higher and average time per case was lower with glass slides as compared with all digital images. Among the digital images, the ×40 or ×40 z-stack had the highest diagnostic accuracy and lowest interpretation time.
Conclusions.—Whole slide imaging is a viable option for the purposes of teaching and consultations, and as a means of archiving cases. However, considering the large file size and total time to reach diagnosis on digital images, whole slide imaging is not yet ready for daily cervicovaginal diagnostic cytology screening use.
Digital imaging is the creation, storage, and transmission of an image file using a computer and is considered by many to be the future of anatomic pathology. Most pathologists currently use some form of digital imaging, such as static images obtained by microscope-mounted optical cameras. Within the field of cytology, digital images are routinely being used for automated computer-assisted screening of Papanicolaou test slides, as well as for training and education. Proficiency testing is currently performed on glass slides, though the future use of digital slides is anticipated. The development of greater image quality and resolution within digital pathology has promoted the use of telepathology, including telecytology, which has been shown to be acceptable for both adequacy checks and rapid cytology diagnoses.1–4 Other uses of digital imaging in pathology include image-enhanced reports, live imaging through robotic microscopy,5 remote surgical frozen section review,6,7 quality assurance testing,3 and image cytometry/analysis.
Whole slide imaging (WSI), a subset of digital imaging, is the process of scanning an entire glass slide and converting the data into a high-resolution digital image that is viewed and manipulated on a computer. Potential applications of WSI include case sign-out, archiving for consultation or legal cases, and education. Advantages of this technology include the ability of multiple viewers to access the same slide at once and the exponential increase in the ease of slide retrieval and transmission.
Since WSI was first developed in 1999, several studies have been performed that demonstrate its utility in surgical pathology specimens.8,9 However, its usefulness and clinical application in cytology have historically been limited for several reasons. Among these are (1) the lack of availability of technology to incorporate z-axis viewing; (2) the lack of availability of high-volume scanning technology, which is necessary for high-volume anatomic pathology service lines such as diagnostic cytology; and (3) the large size of the digitized files. More recently, at least 3 systems have become available in the United States for WSI that are capable of high-speed digitization of slides at multiple magnifications. These scanning systems also digitize multiple focal planes (x-, y-, and z-axes), creating a virtual 3-dimensional depth of vision called a “z-stack” (referring to the z-axis), which is a critical component of morphologic analysis when reviewing cytology slides.
In this pilot study, we evaluate the feasibility of using WSI in the screening and interpretation of liquid-based Papanicolaou tests on a routine basis, as compared with conventional methods of manual glass slide screening. Our study includes evaluation of both the accuracy of the final diagnosis and the time to reach a diagnosis.
MATERIALS AND METHODS
Eleven cervicovaginal cytology cases, including both ThinPrep (BD Diagnostics - TriPath, Burlington, North Carolina) (TP) and SurePath (Hologic Inc., Marlborough, Massachusetts) (SP) preparations, were evaluated under a protocol approved by the Institutional Review Board of our institution. Four cytotechnologists and 3 cytopathologists participated in the evaluation. Digitally scanned images were reviewed first, and different magnification scans were presented as separate cases in a random order. The participants were blinded and unaware that the same case was being repeated at different magnifications in a randomized sequence. The diagnoses as well as the time it took to reach a diagnosis were recorded. After completing all digital slides, the participants evaluated the corresponding glass slides. The diagnoses and time it took to screen the glass slides manually were recorded.
The 11 randomized cases included diagnoses of endometrial adenocarcinoma (SP); low-grade squamous intraepithelial lesion (TP and SP); high-grade squamous intraepithelial lesion (TP and SP); atypical squamous cells, cannot rule out high-grade lesion (SP); atrophic vaginitis (SP); Trichomonas vaginitis (TP); herpes (TP); adenocarcinoma in situ (SP); and negative for intraepithelial lesion and malignancy with bacterial shift (SP).
Cases were scanned using the BioImagene iScan Coreo Au 3.0 scanner (Ventana Medical Systems Inc/BioImagene, Sunnyvale, California) at ×20 (0.5 numeric aperture [NA] with 0.46 μm/pixel resolution), ×40 (0.75 NA with 0.23 μm/pixel resolution), and ×40 with a 7-layer z-stack (0.75 NA with 0.23 μm/pixel resolution). All cases were de-identified and each scan was considered one case. Images were viewed using a computer equipped with an Intel Pentium 3.44-GHz processor, 1 GB RAM, 64 MB VRAM, and a display monitor with 1600 × 1024 pixel resolution. The file storage formats included JPEG 2000, BioImagene imaging file, and TIFF file viewed in 24-bit true color, with file sizes varying from 325 MB up to 24 GB (due to the 7 levels of z-stack scanning). Conventional glass slide review was performed with Olympus BX40 CX microscopes (Olympus America Inc., Center Valley, Pennsylvania) at ×4 (0.1 NA), ×10 (0.3 NA), ×20 (0.5 NA), and ×40 (0.75 NA).
Difference in time of interpretation with the various scanned magnifications and glass slides was analyzed by 1-way analysis of variance. All pairwise multiple comparisons were analyzed by the Holm-Sidak method; overall significance level was .05. Disagreement with the reference diagnosis was considered in those cases with greater than 1 step difference. The accuracy at different magnifications for the 7 participants was then calculated.
RESULTS
Accuracy of interpretation was superior with glass slides when compared with all scanned magnifications (Table 1). Images scanned at ×20 had the lowest accuracy of detection of squamous and glandular lesions, as well as negative Papanicolaou tests. Accuracy of interpretation of the digital images was best achieved with ×40 magnification. However, z-stack digital images scanned at ×40 improved the accuracy of diagnosis in one case of adenocarcinoma of the endometrium (Figures 1 through 4, A through D).
The time required to interpret digital images scanned at all magnifications (×20, ×40, and ×40 z-stack) was significantly longer than the time required to achieve a diagnosis with glass slides (Table 2). Interpretation of images scanned at ×40 using the z-stack method was significantly faster than that of digital images scanned at ×20. There was a nonsignificant trend toward faster interpretation of z-stack images scanned at ×40 as compared with standard images scanned at ×40.
COMMENT
Several studies have recently been performed that suggest the accuracy of diagnosis with WSI is comparable to that with manual glass slides, though these studies have been limited to surgical specimens.8,9 Our study of liquid-based cytology specimens has also determined acceptable accuracy of imaging quality, particularly at higher scanned magnifications. However, digital imaging in cytology has yet to overcome many obstacles.
Disagreement with the reference diagnosis was greater in the interpretation of digital images than when reviewing glass slides. In part, this may be explained by lack of familiarity with the new technology. The diagnosis of cases with few abnormal cells or background distracters was more difficult on digital images than with glass slides. The inability to methodically screen the entire scanned slide added to the increased evaluation time and may explain some cases of false-negative diagnosis.
Thick cell groups in particular pose a problem for accurate cytologic diagnosis. Improved accuracy of interpretation of glandular lesions (as in the case of endometrial adenocarcinoma) was seen with the use of a ×40 z-stack image, as compared with other magnifications, suggesting that the use of z-stack scanning may ameliorate this problem. Performance of this magnification in the case of adenocarcinoma in situ could not be evaluated because this case was inadvertently not scanned at ×40 with a z-stack.
In cases of severe atrophic vaginitis and herpes infection, low diagnostic accuracy was most likely due to inability to adequately examine nuclear detail, especially in dark, overlapping cell groups. Previous studies have also concluded that hyperchromatic crowded groups are more difficult to diagnose on digital imaging,3 likely because of the small number of atypical cells present as well as insufficient resolution to permit adequate evaluation at ×40 without z-axis focusing ability. In some cases, ×100 magnification was necessary to accurately discern architecture and background.3
Among digital images, those scanned at ×20 magnification had the lowest accuracy. Inability to examine cellular detail and inability to change the image focal plane limited the ability to interpret these digital images. Images scanned at ×20 had very poor resolution at a higher virtual magnification, such as ×40. The participants were able to examine both nuclear and cytoplasmic details more completely with the scanning magnification of ×40 or ×40 with a z-stack, thereby increasing the diagnostic accuracy.
Perhaps the largest barrier to routine use of WSI is the enormous image file size, which leads to increased slide scanning time and slower slide review. Static images capture a limited area of the slide and require approximately 3 to 5 MB of memory, whereas comprehensive whole slide images necessarily require hundreds of megabytes per image. Current WSI technology enables a user to obtain depth of focus by completing multiple scans of a slide at slightly different focal planes, which are then combined to form the z-stack final image.5,10 Each image layer of the z-stack requires the same amount of scanning time and memory, leading to impractical loading and transmission times. Our study found WSI scanning times ranged from 3 minutes and 20 seconds to 95 minutes and 39 seconds, and file sizes ranged from 325 MB to 24 GB (Table 3).
Current commercially available whole slide scanners are unable to scan the edges of a glass slide. Therefore, specimens that extend to the edges of the slide, such as is often the case with TP slides, are not fully scanned (Figure 5, A and B). The potential loss of valuable diagnostic material in these areas is troublesome. This problem was not encountered with SP slide preparation because the slide surface area occupied by cellular material is not directly adjacent to the slide edges (Figure 6, A and B).
We found that even though the examiners could change the virtual magnification to a lower power for screening, such as ×20 or ×10, the system frequently “froze,” causing delay and frustration to the participants. Screening of the entire digitized slide was therefore cumbersome, and the participants felt they could not methodically screen the entire slide. This may explain the lower diagnostic accuracy in cases with few abnormal cells or in cases of high-grade dysplasia in a background of human papillomavirus effect or presence of other distracters. It is important to note that the loading speed of whole slide images is directly dependent on the speed of the network and computer, as well as file size. Thus, these issues may be alleviated in the future with upgraded network and computer capabilities, or new advances in file storage that decrease file size and access time. Lower accuracy of interpretation and longer evaluation times of digital slides, regardless of the scanning magnifications, were in large part due to the lack of familiarity of the participants with the new technology. We anticipate that increased frequency of use will increase the diagnostic accuracy and diagnostic speed of the user over time.
As a direct result of large file size, one of the biggest drawbacks to converting a laboratory to digital whole slide imaging is the sheer amount of storage space needed for archiving files. One suggested solution to conserving storage space11 is scanning smaller numbers of selected cases at ×40, supplementing or documenting select cases with still images at ×40 or above, and storing immunohistochemical and special stain slides as still images. Scanning slides at ×20 magnification would significantly decrease storage space and would be a good option for surgical pathology, but it is not a good digital storage solution for cytology specimens because nuclear detail is lost at this magnification. SurePath slides have a smaller occupied surface area relative to TP slides, which translates to faster slide scanning time and smaller file size by comparison. Another alternative may be the storage of only certain types of cases, particularly those that need to be shared with other people, such as cases presented at tumor boards, outside consult cases, and teaching cases.
Despite the many drawbacks, WSI may be useful in some cytopathology specimens, such as thyroid carcinoma and lung, breast, or colon adenocarcinomas, as the demand for molecular studies for prognostic and treatment purposes of these specimen types is increasing. Many studies now show that cytology material used for molecular testing is highly accurate, adequate, and comparable to molecular studies performed on formalin-fixed paraffin-embedded resection specimens.12 Molecular testing of cytology specimens is a viable and particularly attractive option in cases with scant formalin-fixed, paraffin-embedded tissue available, such as small biopsies. Whole slide imaging would preserve and store the cytology image for future specimen comparison, tumor board collaboration, and archival purposes, and the cellular tissue itself could be used for ancillary molecular studies directly from the cytology slide and/or cell block.
The use of digitized WSI has some advantages over traditional glass slides, including the ability of multiple users to view a file at the same time and expedited case retrieval. In addition, file sharing and consultation via email or other secure electronic methods eliminates the need for mailing of fragile slide and block material and potentially reduces case turnaround time.
Industry and government regulation in the United States for the use and storage of digitized pathology images has been slow to develop. The US Food and Drug Administration held its inaugural meeting to discuss the standardization of WSI for use in diagnostic surgical pathology in 2009. At that time, there was no consensus on WSI regulation.13–15 In late 2011, the Food and Drug Administration took part in a panel discussion on WSI systems at the Digital Pathology Association meeting, where it announced that WSI systems would be classified as Class III medical devices (highest risk) and regulated accordingly. The Food and Drug Administration anticipates another public meeting or publishing further guidance on the topic within the next year.16
CONCLUSION
Whole slide imaging has many potential applications in cytopathology, such as education, image analysis, diagnosis, report documentation, remote consultation, telecytology, tumor board collaborations, immunohistochemical control monitoring, quality control monitoring, outside consult archiving, and standard case archiving. However, considering the file size and the time undertaken to arrive at the final diagnosis on digital images, we believe whole slide imaging for cervicovaginal specimens is not yet ready for daily diagnostic cytology/screening use.
In our study, increasing the image resolution increased diagnostic accuracy and speed. Among the digitally imaged slides, the ×40 or ×40 z-stack images had the highest diagnostic accuracy. However, the best accuracy and lowest average time per case were still obtained with glass slides.
We gratefully acknowledge the important contributions of Michael Thrall, MD, for his expertise and training assistance in using the whole slide digital imager, and Kathryn Stockbauer, PhD, for their assistance with manuscript preparation.
References
Author notes
The authors have no relevant financial interest in the products or companies described in this article.
Competing Interests
This study was conducted with approval of the Institutional Review Board of The Methodist Hospital Research Institute (IRB0000-1224). A consent waiver was granted by the Institutional Review Board for this study.
Presented as a poster at the 58th Annual Scientific Meeting of the American Society of Cytopathology; November 13-15, 2010; Boston, Massachusetts.