MATERIALS AND METHODS

Slides of 50 consecutively received endoscopically resected colorectal polyps originally reported as hyperplastic polyps and 50 consecutively received colorectal polyps originally reported as tubular, villous, or tubulovillous adenomas were retrieved from the files of the department of histopathology, Royal Hallamshire Hospital from the beginning of April 1998. Serrated adenomas were not represented in this series. There was a single haematoxyllin and eosin stained slide for each case. By chance, exactly 50 of each type were received during the same time period.

Three observers were histopathology consultants and two were trainees working in the department of histopathology, Royal Hallamshire Hospital. The length of time that each had spent in histopathology was recorded. Each observer was given the 100 glass slides in the order in which they were received in the laboratory. The observer was asked to assign each polyp to the category of hyperplastic or adenomatous after light microscopic examination, but to make no other assessment (such as grade of dysplasia). The observers were asked to carry out the assignment as accurately as possible. The time taken for each observer to make each diagnosis was recorded.

Eighteen months after the glass slide observations the polyps were presented to the observers as offline telepathology images. The images were captured and presented using a telepathology workstation (PathSight, Fairfield Imaging Limited, Tunbridge Wells, Kent, UK). The images were stored in the Joint Photographic Experts Group (JPEG) format,¹⁷ with a compression ratio of 26 : 1, at a resolution of 1024 by 768 pixels, in 24 bit red green blue colour. The images were displayed on an 18 inch flat panel monitor in normal office ambient lighting conditions. The images were captured by a consultant histopathologist, who was not an observer in the study, using a set protocol. For each case, six images were produced—a macroscopic view of the slide label (to confirm the identity of the case), a low power scan of all the material on the slide, two images of the polyp epithelium using a ×2.5 objective lens, and two images of polyp epithelium using a ×10 objective lens. Each case was saved as a separate session. On loading a session into the telepathology system the six images appeared as thumbnail versions at the bottom of the screen and could be displayed at full size in the main viewing window by a drag and drop action with the computer's mouse (fig 1). A window on the right side of the screen displayed the position of each higher magnification image on an image of the whole scanned specimen. The order of presentation was randomised. None of the observers had any experience of telepathology. Each was given a single 15 minute training session using 10 additional cases. The diagnostic category assignment was identical to the glass slide observations and the time taken to process each case, from the start of loading a session to the written diagnosis, was recorded.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

The screen of the telepathology workstation with a case loaded. The six images are displayed as thumbnails at the bottom of the screen and can be dragged and dropped into the main window (here shown with a higher power view of the polyp epithelium). The locations of the higher power views are displayed dynamically on the low power image at the right side of the screen.

The agreement between all the observers on glass slides and offline telepathology images and the consensus diagnoses were assessed using κ statistics with 95% confidence intervals.¹⁸

RESULTS

The results are summarised in tables 1 to 4. The median time taken to produce the offline telepathology images of each case was 210 seconds (range, 112–478). The times taken to make the diagnoses on the offline telepathology images (table 1) were significantly shorter than the times taken to make the diagnoses on the glass slides (Mann-Whitney U test, p < 0.0005). The κ statistics for the interobserver agreement on the offline telepathology images were significantly higher than those for the interobserver agreement on the glass slides (Mann-Whitney U test, p = 0.001). For the glass slide diagnoses, 90 cases had a unanimous consensus diagnosis, six had a 4 : 1 majority consensus, and four had a 3 : 2 majority consensus. For the offline telepathology diagnoses, 96 cases had a unanimous consensus diagnosis and four had a 4 : 1 majority diagnosis.⁶

DISCUSSION

Our study shows that, in the restricted diagnostic domain examined, the skills required to use offline telepathology are learnt very quickly (in 15 minutes), that diagnoses from offline images can be made quickly (median time, five seconds), and are highly reproducible (range of κ statistics for interobserver agreement, 0.90–1.00). These results are for a restricted domain with only two discrete categories with no other considerations (which in colorectal polyps could include grade of dysplasia, completeness of excision, and absence/presence of stalk invasion). They are thus likely to represent the maximum level of performance, in terms of reproducibility and speed, that can be obtained using an offline telepathology system, but because the level of performance is so high, this suggests that there is considerable potential for offline telepathology as a diagnostic medium in histopathology.

The results of our study also give an insight into the diagnostic process in this domain. The short time taken to make the diagnosis by offline telepathology (median, five seconds, including verification of case identity and written diagnosis) suggests that the diagnosis is made by pattern recognition rather than heuristic algorithms,¹⁹ which is to be expected because the system contains only two discrete diagnostic categories.²⁰ The finding that the κ statistics for interobserver agreement on the offline telepathology images were significantly higher than for the glass slide diagnoses suggests that the process of examining the glass slide has more potential for sampling error than offline telepathology images captured by an experienced histopathologist. When examining a glass slide, a histopathologist has to move the microscope stage to look at all the material on a slide because this cannot be seen in a single field of view for any sizeable specimen. Sampling error can be introduced if the histopathologist does not look at all the material, either by choice or by lack of feedback as to which areas have been viewed during the process. The pathologist then has to select areas within a specimen to examine at higher magnification to discern further information, such as the type of epithelium in colorectal polyps. The offline telepathology images in our study eliminated sampling error at the lowest magnification by scanning all the material with a ×2.5 objective lens and constructing a composite registered image. The sampling at higher magnification was made by an experienced consultant histopathologist and therefore might provide more reproducible sampling than that of less experienced trainees. The restrictive sampling of the offline images could give a higher degree of interobserver agreement than glass slide diagnosis but a lower degree of accuracy if misleading areas were selected for higher magnification examination.^{15, 21} However, our study included only hyperplastic and adenomatous polyps (no serrated aenomas were present), and because the epithelium in these diagnostic categories is distinctive, it is unlikely that misleading images could be selected.

“The process of examining the glass slide has more potential for sampling error than offline telepathology images captured by an experienced histopathologist”

The high degrees of reproducibility and speed of diagnosis in our study suggest that offline telepathology is a feasible medium for routine diagnostic histopathology, but further work is required to refine the protocols for working in such an environment. The time taken to capture the images for each case was not excessive (median, 210 seconds) but was carried out by a consultant histopathologist who could have made the diagnoses on glass slides in the same length of time as the observers in the study (median, 13 seconds), so there would be no justification for using this methodology where a consultant histopathologist was available to make the primary diagnosis. However, there are other situations where offline telepathology could be useful. In several countries at present, especially the UK, there is a severe shortage of trained histopathologists and there are hospitals with few or no histopathologists to provide a diagnostic service. This problem could be ameliorated by sending glass slides to services with adequate histopathologists, but this study suggests that offline telepathology could be used as a faster, and possibly more accurate, alternative. In such a system the images would be produced by a trained technician in the originating laboratory, rather than a consultant histopathologist as in our study, who would follow a well defined protocol for each type of specimen. There would not be any change in the quality of the low power scan of all the material on the slide with a non-histopathologist operator, but there could be a less directed sampling of images at higher magnification. This needs to be investigated in further studies. If sampling at higher magnifications by a technician, rather than a histopathologist, does produce less accuracy on a single set of offline images, then a further iteration could be added—histopathologists would annotate areas on the low power scan which they wished to view at higher power and return these to the originating laboratory for further image sampling. Another area where offline telepathology could play a role in diagnostic histopathology is the referral of cases to specialist centres for treatment where there is no uncertainty about the histopathology diagnosis, but where confirmation is required at the receiving centre, perhaps with discussion at a multidisciplinary team meeting, such as in the Calman-Hine model of cancer services that is used in the UK.²² In this context, a consultant histopathologist in the referring hospital could prepare offline telepathology images that could be viewed by pathologists in the receiving hospital and shown at meetings. With diagnostic errors in histopathology receiving more attention by regulatory bodies and in the media,²³ there is discussion of double reading of slides, such as occurs in some areas of radiological practice.²⁴ Such double reading would effectively double the work of histopathologists if carried out using glass slides, but this increase in work could be minimised by the use of offline telepathology images, which could be captured during the initial glass slide examination by the first pathologist.