Concordance Between Telepathologic Diagnosis and Conventional Histopathologic Diagnosis: A Multiobserver Store-and-Forward Study on 20 Skin Specimens

Objective  To study the validity and feasibility of transferring images of cutaneous biopsy specimens via e-mail to remote physicians active in dermatopathology for teleconsultation.

Design  Twenty skin specimens previously diagnosed at the Department of Dermatology, University of Graz, Austria, were subsequently sent for teleconsultation using the store-and-forward method. For each case, 3 or 4 images at different magnifications were sent by e-mail to 16 colleagues (11 dermatopathologists and 5 pathologists) in 15 centers in 6 different countries. Six weeks later each observer received the hematoxylin-eosin–stained specimens to render a conventional diagnosis.

Setting  Dermatopathology and pathology units within institutional and private settings.

Material  Twenty small skin biopsy specimens of cutaneous diseases were selected randomly from a study set of 80.

Main Outcome Measure  Concordance between telepathologic diagnoses and conventional histopathologic diagnoses of 20 skin specimens.

Results  On average, 78% of the telediagnoses were correct (range, 60%-95%), whereas 85% of the conventional diagnoses were correct (range, 60%-95%). A perfect diagnostic concordance was obtained in 7 (35%) of 20 cases, and a significant difference was identified in only 1 case.

Conclusions  Results suggest that telepathology performed by physicians active in dermatopathology may serve as a reliable technique for the diagnosis of cutaneous diseases when experts in dermatopathology are not available locally. Furthermore, teledermatopathology is attractive because it provides an opportunity to obtain timely consultation on difficult cases.

IN RECENT years, continuous progress in computer technology has led to the introduction of a revolutionary diagnostic tool known as telemedicine.^1-3Telepathology is defined as the practice of pathology at a distance, by visualizing an image on a monitor rather than viewing a specimen directly through a microscope.^4-14

Using relatively simple equipment, dermatologists can rapidly transmit microscopic still images photographed at one site to remote centers to obtain a diagnostic consultation. Today's technologies allow a transfer of histopathologic images in both dynamic (real-time) and static (store-and-forward) systems. The first method provides remote consultation via a robotic microscope, which can be controlled by the consulting pathologist. The equipment used for robotic systems is relatively expensive and requires high data transmission rates to achieve real-time image transfer. Using the store-and-forward method, each image is captured and transmitted as a single file. The fields to be examined are selected by the referring pathologist and then transmitted. This system is much less expensive and does not require high-speed data transmission capability. A limited number of studies have addressed the validity of both these systems, and results suggest that static-image telepathology may be less accurate.^15-20

A number of studies have demonstrated that teledermatology represents a useful diagnostic tool, especially in communities where dermatologic services are not available.^21-29 Our group has published 2 studies demonstrating the reliability and reproducibility of store-and-forward teledermoscopy for the diagnosis of pigmented skin lesions.^30,31 In contrast, only a few teledermatopathology studies have been done.^10,19

We report the results of a multicenter teledermatopathologic study on 20 cutaneous specimens. A total of 67 JPEG (joint photographic experts group) compressed representative static images were sent by e-mail from Graz, Austria, to 16 telepathologists around the world (Austria, Czech Republic, Germany, Italy, Japan, and the United States). Each participant rendered a telepathologic diagnosis for each case. After 6 weeks, all participants received representative hematoxylin-eosin–stained specimens of each case and rendered another diagnosis. Comparing these diagnoses allowed us to evaluate the concordance between the telepathologic and conventional diagnoses.

Our study included 20 histopathologic biopsy specimens of cutaneous diseases selected randomly from a study set of 80. We concentrated on small tissue specimens to avoid sampling error. Essential clinical information, the number of images sent for each case, and the final diagnoses are given in Table 1. The clinical information was included on the actual histopathology laboratory submission form. For example, for case 1, a trichofolliculoma, the clinical information read "59-year-old female with a lesion on the forehead clinically diagnosed as sebaceous hyperplasia." In most cases, the information was scant. All original histopathologic specimens were examined with a conventional microscope (BX50; Olympus, Tokyo, Japan) at the Department of Dermatology in Graz by 2 expert dermatopathologists (H. Kerl and H.P.S.). One of them selected the fields to be photographed from the original specimen and chose the magnification level. The final diagnosis for these cases was made by both of the expert dermatopathologists in Graz by assessing clinical features and histopathologic findings.

Images were captured with a digital Kodak DCS 460 camera (Eastman Kodak Co, Rochester, NY), which uses a Nikon N90 body (Tokyo, Japan) mounted on an ordinary Olympus BX50 microscope. Each image was obtained with the digital camera in PICT format file (Macintosh native graphics format). For each case, images at 3 or 4 different magnifications were acquired (Figure 1). The original size of these images was 17 megabytes, 2036 × 3060 pixels (71.83 × 107.96 cm; 72 pixels per inch [ppi]), in RGB color mode (36 bit). Successively, all images were subsampled to a size of 511 × 768 pixels (18.02 × 27.08 cm; 72 ppi) in RGB color mode (24 bit) and compressed by using JPEG compression at a fixed quality setting. The software used for subsampling and compression was Adobe Photoshop 4.0 (Adobe, San Jose, Calif); the JPEG quality setting was set at 6 (this value is proportional to compression ratio). Finally, compressed images had a mean size of 53 kilobytes (kb) (range, 48-67 kb). This compression was required to allow a faster e-mail transmission. All images were sent to the remote centers via e-mail (Eudora Light, San Diego, Calif) together with basic patient data (age, sex, site and duration of the lesion, and clinical diagnosis).

The χ² test and Fisher exact test were used for quantitative parameters. Results were considered statistically significant at P≤.05 (2-sided).³²

The clinical data, number of images, final diagnosis by both the expert dermatopathologists in Graz, and the number of correct telediagnoses and correct conventional diagnoses for the 20 cases are given in Table 1. A diagnostic concordance between the telediagnoses and conventional diagnoses of the 16 pathologists was obtained in 7 (35%) of 20 cases (cases 1, 4, 8, 11, 16, 18, and 19). A significant difference between telediagnoses and conventional diagnoses was identified in only 1 case (case 20). Telediagnoses were superior to conventional diagnoses in 4 (20%) of 20 cases (cases 2, 10, 13, and 14), whereas in 9 (45%) of 20 cases (cases 3, 5, 6, 7, 9, 12, 15, 17, and 20), telediagnoses were inferior to conventional diagnoses.

The performance of each participant is summarized in Table 2. Detailed results on telepathologic diagnosis, conventional histopathologic diagnosis, and concordance between these 2 methods for each of the 20 skin specimens is given for each independent observer. If the same incorrect diagnosis was made on both telediagnosis and conventional diagnosis, this is interpreted as "agreement." For example, observer 1 made 16 correct telediagnoses and 19 correct conventional diagnoses, but only 15 diagnoses were concordant. Thus he or she changed all 4 incorrect telediagnoses, but also changed 1 correct telediagnosis. On average, 78% of the telediagnoses were correct, with a range from 60% to 95%, whereas conventional diagnoses were correct 85% of the time (range, 60%-95%). Differences of 10% or less from perfect agreement were found among 11 (69%) of 16 observers (observers 2, 3, 4, 7, 8, 10, 11, 13, 14, 15, and 16). Differences greater than 10% to 15% occurred among 4 observers (25%; observers 1, 6, 9, and 12), whereas for only 1 observer (6%; observer 5) was the difference from perfect agreement more than 15%. No statistical differences were found comparing the percentage of perfect agreement between telediagnoses and conventional diagnoses for each observer (P>.05).

Table 3 summarizes detailed data on telepathologic diagnoses and conventional histopathologic diagnoses of all 20 skin specimens. It allows one to identify which diagnoses were proposed in addition to the final diagnosis.

In this study, we evaluated the influence of image selection on diagnostic performance to verify concordance between conventional diagnosis and telediagnosis. The results obtained showed that telepathology accuracy was slightly lower than glass slide accuracy (78% and 85%, respectively). Nevertheless, among 15 (94%) of 16 observers, no significant differences in accuracy were found between telepathology and conventional pathology, suggesting the usefulness of teledermatopathology.

Telepathology involves the use of the new telematic technologies to transmit images for the purposes of diagnosis, consultation, and education. In the past years, many articles on telepathology have been published.^4-20 Moreover, many Web sites teaching with telepathology can be found easily with the common search engines. The major problem using e-mail is that the security and confidentiality of the patient data may be compromised by hackers.¹¹

The main problems with static telepathology are image selection and image quality. Image selection is more of a problem for the store-and-forward system. It is important to emphasize that the referring pathologist must select images that adequately present the case without attempting to influence the remote pathologists. The main risk lies in a personal and incomplete photographic characterization of the case on the basis of a diagnostic hypothesis, inadvertently leading the remote pathologist toward a diagnostic misinterpretation. The referring pathologist and the consultant should understand the risks involved in embracing and using these technologies. The store-and-forward method should be considered in small dermatopathology biopsy specimens where the sampling bias is not a major problem. In a large solid tumor, for example, image selection by the local pathologist could represent a greater obstacle.

Problems such as these might be avoided by using a remote-control telepathology system, where the consultant pathologist controls the microscope, directly choosing the areas of viewing and the magnification. One disadvantage, however, of using the real-time systems is that they are more technologically complex and expensive. Furthermore, the consultant pathologist must be online together with the referring pathologist.

The quality of the histopathologic images depends on the equipment used and the capacity of the telecommunications link. For our study, all 67 histopathologic images were compressed using JPEG compression. As demonstrated by Okumura et al³³ in 1997 and by Kittler et al³⁴ in 1998, with this type of compression the lost information is most likely to be imperceptible to the individual doing the remote assessment. The compression of the images was necessary to allow a faster e-mail transmission; JPEG is an accepted standard image type for the multipurpose Internet mail extension (MIME). This technology is an extension to the e-mail message format facilitating the delivery of multimedia messages such as images, sounds, and movies.

Making telepathologic diagnosis on a high-resolution monitor is qualitatively different than directly viewing images through a conventional light microscope. Becker et al³⁵ demonstrated the importance of experience in using a telepathology system both for accurate reading of video images and for effective communication between sending and receiving pathologists. The telepathologic diagnostic accuracy of many of these systems is untested. Our study suggests that for small biopsy specimens, the sampling procedure is not really problematic and that the store-and-forward method works well.

In 5 cases (cases 4, 8, 11, 13, and 16), the telepathologic diagnosis was identical to the final diagnosis established by clinicopathologic correlation, whereas the conventional histopathologic diagnosis was the same as the final diagnosis in 6 cases (cases 4, 5, 8, 11, 12, and 16). In 2 cases (cases 1 and 2), there was only a slight difference between telediagnoses and conventional diagnoses compared with the final diagnosis. In case 1, three of 16 pathologists preferred the diagnosis of trichoepithelioma instead of trichofolliculoma not only when viewing transmitted images but also by conventional microscopy. In case 2, two remote pathologists and 4 pathologists performing conventional diagnosis made the diagnosis of xanthoma instead of xanthelasma. In cases 6, 7, and 9, only the combination of both clinical and histopathologic findings permitted one to obtain the final diagnoses (pityriasis lichenoides for the first case, Grover disease for the second, and Hailey-Hailey disease for the third). On the basis of only the histopathologic features, many participants diagnosed other bullous diseases such as pemphigus, Darier disease, herpesvirus infection, and acantholytic dermatitis for cases 7 and 9 (Table 3). This result underscores the importance of meticulous clinicopathologic correlation for the diagnosis of certain inflammatory skin diseases. Such a clinicopathologic correlation could be achieved by evaluating a relevant clinical image, something that could be integrated in a telepathology protocol.

This study is based on only 20 skin specimens. We have confirmed that the store-and-forward method is a useful tool within the field of dermatopathology at least when dealing with small biopsy specimens where sampling bias is not a major obstacle. Particularly in inflammatory skin diseases, a meticulous clinicopathologic correlation is still crucial for achieving competent diagnosis.

Accepted for publication May 1, 2001.

Corresponding author: H. Peter Soyer, MD, Department of Dermatology, University of Graz, Auenbruggerplatz 8, A-8036 Graz, Austria (e-mail: peter.soyer@unigraz.at).