Robotic Telepathology for Intraoperative Remote Diagnosis Using a Still-Imaging Based System

Transcription

1 Anatomic Pathology / TELEMICROSCOPY SYSTEM FOR REMOTE DIAGNOSIS Robotic Telepathology for Intraoperative Remote Diagnosis Using a Still-Imaging Based System Francesca Demichelis, MSc, 1 Mattia Barbareschi, MD, 2 Sebastiana Boi, MD, 2 Claudio Clemente, MD, 3 Paolo Dalla Palma, MD, 2 Claudio Eccher, MSc, 1 and Stefano Forti, MSc 1 Key Words: Telemedicine; Active telepathology; Static imaging; Remote frozen section service Abstract The aim of the present study was to assess whether a telemicroscopy system based on static imaging could provide a remote intraoperative frozen section service. Three pathologists evaluated 70 consecutive frozen section cases (for a total of 210 diagnoses) using a static telemicroscopy system (STeMiSy) and light microscopy (LM). STeMiSy uses a robotic microscope, enabling full remote control by consultant pathologists in a near real-time manner. Clinically important concordance between STeMiSy and LM was 98.6% (95.2% overall concordance), indicating very good agreement. The rates of deferred diagnoses given by STeMiSy and LM were comparable (11.0% and 9.5%, respectively). Compared with the consensus diagnosis, the diagnostic accuracy of STeMiSy and LM was 95.2% and 96.2%. The mean viewing time per slide was 3.6 minutes, and the overall time to make a diagnosis by STeMiSy was 6.2 minutes, conforming to intraoperative practice requirements. Our study demonstrates that a static imaging active telepathology system is comparable to dynamic telepathology systems and can provide a routine frozen section service. Telepathology is an attractive tool for pathologists as it may give the opportunity to obtain timely consultations on difficult cases or provide an intraoperative frozen section diagnosis service to small hospitals lacking an in-house pathology service. 1,2 Telepathology systems basically include 2 different modes of operation for remote diagnosis, 1,3 one in which the selection of diagnostic images is accomplished by the pathologist requiring passive telepathology, 4,5 and the other one, based on the use of a robotic microscope driven at a distance, in which the consultant pathologist has the opportunity to select the images on which to make the diagnosis, or active telepathology. 2,6-11 Active telepathology is receiving increasing attention as the problems caused by inadequate selection of tissue fields by the referring pathologist can be avoided. 1 This impediment of full selection probably will limit future, routine use of telepathology for primary diagnostic purposes, while it will be suitable for second opinion, education, and training. In the literature, the active systems have been mentioned as robotic telepathology systems 6,9,12 or telemicroscopy (TM) systems. 13,14 TM can be performed through 2 distinct modes: (1) a dynamic mode, in which the live images are transmitted from a remote site and viewed in real-time at the local site (dynamic TM), 6-9 and (2) a static mode, in which still images are transmitted to a remote site in a near real-time manner (static TM) Some hybrid systems between these technologic solutions also have been explored, in which high-resolution still images are transmitted on demand to confirm diagnoses and for documentation. 7,15 Real-time TM, using live images, has been associated in practice mainly with frozen section services, 2 although a 744 Am J Clin Pathol 2001;116: American Society of Clinical Pathologists

2 Anatomic Pathology / ORIGINAL ARTICLE real-time telepathology service for evaluating fixed sections also has been described. 6,15 Presently, most TM studies have focused major attention on dynamic, real-time systems, 2,6-9,16 while active systems based on static imaging have not been studied as extensively In this retrospective validation study, we examined the performances of an active telepathology system based on static imaging for a remote frozen section service. Our interest in this solution is driven mainly by the reduced bandwidth requirement for the static-imaging system in comparison with the videoconference requisite. Major features of the system that have been measured were concordance between TM and light microscopy (LM), diagnostic accuracy of TM with respect to the truth diagnosis, and the time needed for a diagnostic session by our static system. Materials and Methods System Architecture The static telemicroscopy system (STeMiSy) used in the present study was designed and developed in our laboratory as a client-server based application. The remote workstation consists of a PC (Pentium MMX running Windows 98 [Microsoft, Redmond, WA]) and a fully motorized microscope (Leica DM RXA) with a motorized x/y stage (No. 14/1114 with step motor control Leica DM STC [Leica Microsystems S.P.A., Milan, Italy]) equipped with 5 objectives (2.5, 5, 10, 20, 40 ). The microscope is connected to the PC via a serial RS 232C interface. Two cameras are mounted on the microscope through a beam-splitter: an isc2050 Digital Video Camera is used with the Leica AFS autofocus system (automatic focusing), and the SONY DXC-950P 3CCD Color Video Camera (resolving 750 TV lines) captures the images to be transmitted to the Matrox Meteor PCI Frame Grabber (Matrox, Montreal, Canada) via RGB connector (the Matrox Meteor has a resolution of pixels). The workstation incorporates a Sony Multiscan 300sf 21-inch monitor. The server application is installed on the remote workstation. The local workstation on which the client application is installed consists of a Pentium 200 running Windows NT (Microsoft). A Sony Multiscan 300sf 21- inch monitor is incorporated. The user interfaces for both applications are designed for a video spatial resolution of 1, pixels with a color depth of 24 bits per pixel. The transmission protocol is TCP-IP. During this study, the 2 workstations were part of 2 local area networks connected through ISDN lines (transmission rate, 128 kilobits/s) managed by routers (CISCO 3620 and 3640). Specifically designed controls and modules are included in both server and client applications for command and image transmissions (through Microsoft Winsock Control 6.0), for acquisition board managing and image processing (through the Matrox Imaging Library MIL 4.02 device libraries), for audio conferencing, and for local and remote microscope control (through Leica DMR Logical DLL, API version 4.03 and Autofocus DLL, API version 4.03). Images were software-compressed JPEG with compression quality factor 90. All were transmitted as pixel digital resolution images. System Operation After a slide is placed under the microscope, the technician moves the microscope stage to locate the 4 extremes of the tissue, looking at the real-time image on the PC screen. Single images are automatically taken at 2.5 magnification to cover the selected tissue fields. By combining the 2.5 magnification digital images, a panoramic overview of the tissue is arranged. The maximal tissue area that can be visualized through the panoramic view measures 19,328 µm 14,432 µm. The telepathologist asks for the connection to the server application. An audio conference can be activated to facilitate the technician-pathologist interaction. Once the connection has been established, the technician can send the combined panoramic view and the 2.5 magnification single images to the remote pathologist. Afterward, the telepathology session is controlled totally by the remote pathologist. Through the client application interface, the pathologist controls every local and remote action using the mouse. As shown in Image 1, the zoom in panoramic view is always viewed on the screen, while single images can be viewed at high resolution. A colored rectangular frame on the panoramic view indicates the region of the tissue shown. A grid system facility, based on cells with sizes corresponding to single-objective tissue areas, enables the pathologist to select and visualize already acquired images at specific magnifications. This facility also permits the pathologist to acquire new images (single or multiple choices are available). The grids can be overlaid to each image. Moreover, this grid system ensures that each tissue area is taken only once with each objective. Digital Archive At the end of each session, all data concerning the TM diagnosis are stored in a database. Date, time, progressively numerated images with corresponding microscope setting information (ie, objective, stage coordinates, lamp intensity American Society of Clinical Pathologists Am J Clin Pathol 2001;116:

3 Demichelis et al / TELEMICROSCOPY SYSTEM FOR REMOTE DIAGNOSIS Image 1 Client application interface of the static telemicroscopy system. The image illustrates the situation on the pathologist s workstation viewing a high resolution 20 image with an overlaid grid corresponding to 40 magnification. On the bottom left, the panoramic overview of the tissue is overwritten by a grid corresponding to the high resolution 20 image. The green dots on the panoramic overview distinguish the already acquired areas with 20 magnification; the red bordered area corresponds to the high-resolution image viewed on the screen. The top left side (the gray zone) contains lines associated with already acquired images, divided by magnification (vertical columns); the red line corresponds to the high-resolution image viewed on the screen. value), diagnosis, patient data, and the server workstation log file (with microscope actions and communication information) are stored. Case Material Seventy sequential frozen sections of cases obtained intraoperatively during surgical procedures were obtained in the routine practice of one of us (P.D.P.), who did not participate in the study. A single representative slide from each case (summarized in Table 1 ) was selected for the study. The frozen section specimens were stained with H&E. The consensus truth diagnoses were made on the frozen section specimens by conventional LM by a review panel (reference diagnoses had been determined before the trial began). The truth diagnoses were distributed as follows: 36 benign cases, 30 malignant cases, and 4 deferred cases. Study Protocol Three senior pathologists (M.B., S.B., C.C.) participated in this retrospective study, two of whom had previous experience with diagnostic digital images (pathologists A and C) and one without such experience (pathologist B). In both LM and TM systems, each pathologist examined 70 cases, making a diagnosis of malignant vs benign lesion and commenting about a more specific histologic diagnosis in accordance with the pathologist s practice procedures. The LM diagnostic session occurred almost 1 month after the TM session. The pathologists had at their disposal the gross 746 Am J Clin Pathol 2001;116: American Society of Clinical Pathologists

4 Anatomic Pathology / ORIGINAL ARTICLE Table 1 Validation Study Cases Organ/System Number Gastrointestinal 8 Gallbladder 2 Lymph node 10 Parotid gland 1 Breast 8 Ovary 8 Thyroid 5 Liver 2 Lung 7 Mesocolon or peritoneum 3 Skin 2 Oropharynx 1 Uterus 4 Kidney 4 Testicle 2 Pancreas 1 Parathyroid 2 Total 70 description of the surgical specimens reported on the histopathologic report of the frozen sections, provided by the Institute of Pathology of S. Chiara City Hospital, Trento, Italy. For each case, pathologists compiled a preformed template with a check-box diagnosis of malignant, benign, or deferred (postponed); a free-text diagnosis; and a checklist of possible reasons for uncertainty (image quality, tissue quality, insufficient experience, other). Data Analysis Diagnoses were scored as follows: type a, completely concordant; type b, discordant, not clinically relevant; or type c, discordant and clinically relevant, ie, a discordance that would have made a difference in the surgical management of the patient. Concordance between TM and LM was measured, examining for each case the specific diagnosis made in each mode. Cases in which one of the two diagnoses was deferred when the other was a definitive diagnosis were considered to have at least type b discordance. Cases in which both TM and LM diagnoses were deferred were examined for the specific diagnosis and scored as concordant if the reason for the deferral was the same. Statistical Analysis The kappa statistic of Cohen was used as a measure of concordance between TM and LM diagnoses. The McNemar test was used to analyze differences in diagnostic accuracy and deferrals between TM and LM. A P value of.05 was considered the limit of significance. Results Concordance Individual performance of each pathologist making diagnoses by the modes are summarized in Table 2. Clinically relevant and overall concordance between TM and LM were 98.6% and 95.2%, respectively. Seven times the diagnoses were discordant but clinically irrelevant (type b); 3 times the discordance was classified as clinically relevant (type c). The kappa coefficient between the 2 methods of diagnosis was kappa = 0.97 [SE (kappa) = 0.06; z = kappa/se (kappa) = 17.0; P <.0001], indicating a very good strength of agreement. The overall case deferral rates were 11.0% (23/210 cases) and 9.5% (20/210 cases) by TM and LM, respectively. For each pathologist, no statistically significant difference between deferrals by the 2 reading methods was found (P >.05). Of 23 cases, 15 were deferred both by TM and by LM (Table 2). Intraobserver Discordance By examining cases in which TM and LM diagnoses were discordant Table 3, it is immediately apparent that most discrepancies were related to minor variations in the level of confidence for making some diagnoses or to minor interpretation differences (cases 20, 53, 64, 70), which were scored as b type discordance. A few considerations merit comment. Two pathologists made a diagnosis of malignancy or suggestive Table 2 Individual Performance of Telemicroscopic and Light Microscopic Diagnoses Pathologist Performance A B C Total Concordance (%) * 66/70 (94) 66/70 (94) 68/70 (97) 200/210 (95.2) Clinically relevant concordance(%) 69/70 (99) 69/70 (99) 69/70 (99) 207/210 (98.6) Cases deferred, telemicroscopy (%) 9/70 (13) 10/70 (14) 4/70 (6) 23/210 (11.0) Cases deferred, light microscopy (%) 9/70 (13) 8/70 (11) 3/70 (4) 20/210 (9.5) * Discordance includes types b (discordant but clinically irrelevant) and c (discordance was classified as clinically relevant). Discordance includes type c only. American Society of Clinical Pathologists Am J Clin Pathol 2001;116:

5 Demichelis et al / TELEMICROSCOPY SYSTEM FOR REMOTE DIAGNOSIS Table 3 Intraobserver Discordance: Telemicroscopic vs Light Microscopic Diagnoses * Diagnosis Case No. Specimen Pathologist Telemicroscopic Light Microscopic 20 Bronchial resection margin A Benign Deferred; suggestive of squamous atypia B Benign Deferred; suggestive of squamous atypia 53 Liver biopsy A Malignant; hepatocellular carcinoma Deferred; suggestive of hepatocellular carcinoma 54 Lymph node A Malignant; carcinoma Deferred; sinus histiocytosis (during prostatectomy) B Deferred; probably negative for neoplasia Benign; no evidence of neoplasia C Deferred; probably negative for neoplasia Benign; no evidence of neoplasia 57 Gastric wall biopsy B Deferred; no convincing evidence Malignant; focus of diffuse undifferentiated (in patient with ovarian of neoplasia carcinoma neoplasm) C Benign; no evidence of neoplasia Deferred; focus of suspicious cells 64 Thyroid A Deferred; possible thyroiditis Benign; thyroiditis 70 Thyroid nodule B Deferred; follicular neoplasm Benign; thyroid adenoma * Diagnoses in boldface represent clinically relevant discordance (intraobserver discordance). of malignancy in case 57 using LM, while making a diagnosis of benignity by TM. By reevaluation of the original slide and of the images acquired by the pathologists, it was apparent that the malignant cells were not scattered throughout the specimen but were focused in a small marginal area, which was found easily with LM but not at the first panoramic images ( 2.5) of the TM system Image 2. This panoramic image was indeed somehow blurred and did not permit the identification of a small focus of neoplastic cells. To obtain a satisfactory level of image quality in this particular case, it would have been necessary to scan the whole section with the 5 objective, a procedure that is too time consuming to be reasonably adopted. A similar problem was noticed by pathologists B and C concerning case 54: both pathologists were not confident in excluding the presence of small neoplastic foci by using the TM panoramic image, while this exclusion was done easily with the LM system. By interviewing the pathologists after the study had been completed, all reported that this loss of detail problem on panoramic images was of major concern for cases in which it was important to rule out even small foci of neoplastic cells (such as exclusion of metastatic cancer in otherwise macroscopically negative lymph nodes). Diagnostic Accuracy In Table 4, the performance of the pathologists making TM and LM diagnoses compared with truth diagnoses is shown. Clinically relevant concordance between TM and LM diagnoses and truth diagnoses were 95.2% A B Image 2 Gastric proximal resection margin in a case of antral adenocarcinoma. A, The panoramic view was negative with the exception of a tiny area of higher cellular density, present in only 1 of 16 frames. The arrow points to the cellular area ( 2.5). B, This focus, at higher magnification, proved to be positive for adenocarcinoma ( 20). 748 Am J Clin Pathol 2001;116: American Society of Clinical Pathologists

6 Anatomic Pathology / ORIGINAL ARTICLE Table 4 Individual Performance of Pathologists Making Telemicroscopic (TM) and Light Microscopic (LM) Diagnoses vs Truth Pathologist Performance A B C Total Concordance, TM vs truth (%) * 58/70 (83) 62/70 (89) 63/70 (90) 183/210 (87.1) Clinically relevant concordance, TM (%) 66/70 (94) 66/70 (94) 68/70 (97) 200/210 (95.2) Concordance, LM vs truth (%) * 62/70 (89) 66/70 (94) 63/70 (90) 191/210 (91.0) Clinically relevant concordance, LM (%) 67/70 (96) 68/70 (97) 67/70 (96) 202/210 (96.2) * Discordance includes types b (discordant but clinically irrelevant) and c (discordance was classified as clinically relevant). Discordance includes type c only. Table 5 Discordance for Each Pathologist With the Two Methods vs Truth Method Telemicroscopic Light Microscopic Error Error Pathologist Total FP FN DP DN PD ND Total FP FN DP DN PD ND A B C DN, pathologist deferred case that was negative for the truth ; DP, pathologist deferred case that was positive for the truth; FN, false negative; FP, false positive; ND, case diagnosed as negative by the pathologist was deferred by the panel; PD, case diagnosed as positive by the pathologist was deferred by the panel. by TM and 96.2% by LM. No statistically significant difference in the diagnostic accuracy of the 2 methods was found (P >.05). Discordant results are shown in Table 5. These discrepant diagnoses were as follows: (1) In 1 case of ovarian neoplasm, pathologist A made a diagnosis of benign lesion with both methods (TM and LM), while the panel diagnosis was ovarian adenosarcoma (the same case was deferred by both other pathologists with both methods; false negative). (2) In 1 gastric wall biopsy specimen, pathologist C missed a small focus of undifferentiated carcinoma with the TM method (false negative). (3) Pathologist A misdiagnosed a sinus histiocytosis of the lymph node as metastatic carcinoma with the TM method (false positive). (4) Pathologist C made a diagnosis of squamous carcinoma in situ on a bronchial Table 6 Intraobserver Correlation of Correct and Incorrect Diagnoses * Light Microscopic Telemicroscopic Correct Incorrect Total Correct Incorrect Total * Discordance includes type c (discordance was classified as clinically relevant). resection margin by LM, while the same case had been deferred by the panel. Intraobserver Correlation In Table 6, the concordance between correct and incorrect diagnoses using TM and LM for 3 pathologists is shown. Seven cases were misdiagnosed by both methods, suggesting a higher degree of difficulty in complexity of these cases. Time Studies The factors influencing the time needed for making a diagnosis are mechanical microscope operations (eg, stage movements, objective change, autofocus), image handling (capture, compression, transmission, storage), and the pathologist s diagnostic process. The overall time required to perform a TM session included the time required by the remote operator (technician or surgeon) to select the 4 extremes delimiting the tissue (technical time), the time needed by the system to acquire and transmit the images composing the panoramic view of the tissue (panoramic time), and the time taken by the pathologist to acquire and analyze the still images and make the diagnosis (viewing time). The mean overall time was 6.2 minutes (range, minutes) Table 7. American Society of Clinical Pathologists Am J Clin Pathol 2001;116:

7 Demichelis et al / TELEMICROSCOPY SYSTEM FOR REMOTE DIAGNOSIS Table 7 Pathologist Performance Making Telemicroscopic Diagnoses Pathologist A B C Total Mean ± SE time per slide (min) Previewing * 2.6 ± ± ± ± 1.1 Viewing 3.1 ± ± ± ± 2.3 Overall 5.7 ± ± ± ± 3.3 Mean ± SD No. of images per slide 15.9 ± ± ± ± 9.3 * Including technician operation and acquisition and transmission time for panoramic images. Interval between panoramic receipt and referral transmission. Including previewing and viewing time. Including panoramic images and diagnostic images. Both the number of diagnostic images acquired by the pathologist and the viewing times per slide did not decrease significantly during the study, suggesting immediate learning of system use. The slopes of linear regression of viewing time vs the number of cases were 1.1, +0.4, and +0.02, respectively for pathologists A, B, and C. The slopes of linear regression for the number of images versus case number were 0.01, +0.06, and 0.0, respectively, for pathologists A, B, and C. The average number of acquired images per slide, including panoramic images, was 17.0 (range, 4-33). The pure diagnostic image transmission time, included in our study in the viewing time, accounts on average for 10% of the viewing time. This means that with our system, the diagnostic process, regarding time, does not depend on the bandwidth; a narrower or larger bandwidth does not linearly influence the mean overall time of a TM session. Moreover, during image transmission, the pathologist can proceed with analysis. Discussion In this validation study, we demonstrated that our active TM system, based on still imaging, permits performance of accurate remote diagnoses on frozen sections in a short period. More findings suggest the compatibility of active TM systems for their use in the setting of intraoperative frozen section diagnosis in rural hospitals without an in-house pathology department. 2,7-9,11,12 In our study, the clinically important concordance between TM and LM was 98.6% (95.2% overall concordance), and the diagnostic accuracy of TM and LM was not statistically different. Moreover, the rate of deferred diagnoses given by TM and LM was comparable (11.0% and 9.5%, respectively), indicating that deferral rate was not ascribable to the reading method but to intrinsic diagnostic difficulty of the cases. Our results are in keeping with those of other studies that evaluated the feasibility and accuracy of the use of telepathology as a tool for remote frozen section diagnoses (reviewed in Nordrum 2 ) and showed a high reliability of the systems, which could allow their use in daily practice. Some of these studies investigated active systems based on still images or dynamic video signal, 7-9,14,16,17 and others studied the use of passive systems. 5,18-21 By using a dynamic active system, Shimosato et al 17 examined 144 final histologic specimens and 14 frozen sections, achieving a total diagnostic accuracy of 90.5% and a frozen section case accuracy of 86%. Nordrum, 2 using a hybrid dynamic/store-and-forward system, found that the correct diagnosis was made in 147 of 162 frozen sections (90.7% overall accuracy). Schwarzmann et al 8 performed several field tests using an active TM system for frozen section diagnosis: the concordance between diagnoses made by telepathology on frozen sections and by LM on embedded material was 82% for 109 cases of lung surgery, 97% for 139 cases of breast surgery, and 79% for 47 cases of consecutive surgery. Della Mea et al, 9 using a system providing static and dynamic features, found a diagnostic accuracy of 100% in 60 frozen section cases examined. Active TM systems based on still imaging were studied by others Oberholzer et al 10 reported a sensitivity of 85.7% and specificity of 100% for a diagnosis of malignancy on 53 frozen section cases. In the study by Tsuchihashi et al, 12 no indication of the system accuracy was provided. Winokur et al 11 conducted a study on frozen and paraffin sections on their static imaging telepathology system; they reported an overall diagnostic accuracy of 96.2% for the 192 diagnoses made by 3 pathologists compared with the reference truth. A principal concern in the use of telepathology in providing a routine remote frozen section service is the time needed to make the diagnoses. This time strongly depends on the experimental design of the systems (eg, bandwidth of telecommunications link, architectural software structure of 750 Am J Clin Pathol 2001;116: American Society of Clinical Pathologists

8 Anatomic Pathology / ORIGINAL ARTICLE the system, type of preparation, type of study). It is difficult, therefore, to compare the results from different studies. Most report a time in the range of 3 to 12 minutes for the active dynamic mode of TM 6,7,9,13-17 and in the range of 15 to 40 minutes for the active mode based on static imaging. 10,12 In the present study, the average viewing time per slide was 3.6 minutes, and the overall time to make a diagnosis by TM was 6.2 minutes, both clearly within the limits of its routine use in real life practice. Of interest, the time we measured is similar to that found for dynamic TM. The most significant drawback of dynamic systems is the strict requirements of high bandwidth communications to perform a remote diagnosis session. 3 Videoconference technology needs at least a bandwidth of 384 kilobits per second (kbps) to ensure a sufficient transmission rate of frames per second and relatively good image quality. Conversely, a system like ours, based on static imaging, does not require specific telecommunication lines, and the bandwidth affects only the transmission times. In our system, this transmission time is only 26% of the previewing time and 10% of the viewing time. This means that, for example, a reduction by half of the bandwidth (ie, in our system from 128 kbps to 64 kbps) would produce a change in the overall diagnostic time of 15%, increasing the duration of the diagnostic procedure by 60 seconds. One of the main problems in making a TM-based diagnosis in the setting of intraoperative frozen sections is the type of case in which the pathologist needs to exclude the presence of even tiny foci of neoplastic cells, eg, frozen section evaluation of macroscopically normal lymph nodes during intraoperative staging procedures for cancer. In these cases, the quality of the initial panoramic view of the specimen is of utmost relevance: it is this image that guides the efficient selection of the areas to be examined at higher magnification. One of the pathologists involved in the present study suggested that in these selected cases it would be necessary to scan the whole slide through static and discrete medium power ( 5 or 10) images. A panoramic overview at 5 magnification will increment the number of images to be acquired (from an average of 12 to an average of 48). This option will be implemented in our system, and the overall diagnostic procedure will last 4 or 5 minutes more. A further aspect of our system is that all information regarding a telediagnostic session can be stored in a database (eg, images, clinical and technical information), so that the diagnostic session can be exactly reconstructed. This feature could become important for medicolegal reasons, as the system will be used in routine practice. The software design allows our system to be used also in a stand-alone configuration. This feature could permit storage of selected images of interesting histologic and cytologic cases for second opinion diagnoses and documentation. Moreover, such stored images would be of interest for training purposes, especially when an expert pathologist could select the most relevant images of paradigmatic or educational cases through the most appropriate diagnostic sequence. At present, the system is being tested in 2 regional hospitals (located in Trento and Rovereto) with pathology departments on site. Technological and management issues need to be monitored to provide system refinements and to gather organizational hints before introducing the service in a rural hospital for intraoperative routine diagnoses. In the rural site, an additional feature will be introduced in the system to allow for gross examination of the tissue. A digital camera will be connected to the server workstation, and still images will be sent to the consultant pathologist. By audioconferencing and simple graphic tools to assign the areas of interest, the pathologist will be able to guide the technician in the sampling of the surgical specimens. Our study supports the concept that an active telepathology system based on static imaging has the potential to provide a remote routine frozen section service in a hospital that does not have an on-site pathology department. The performance of our system is comparable to that of dynamic telepathology systems, and the level of concordance between the TM and LM is high. The diagnostic accuracy by TM shows good agreement with LM accuracy. Furthermore, the time needed to make a diagnosis by TM is comparable to the time required in an intraoperative service. Furthermore, the high degree of user-friendliness of the systems and the low degree of dependence of the diagnostic time on the bandwidth of telecommunications may contribute to the introduction of this system as a routine diagnostic service in the rural hospitals of our region. From the 1 Istituto Trentino di Cultura, Istituto per la Ricerca Scientifica e Tecnologica (ITC-irst), Trento, Italy; 2 Ospedale S. Chiara, Trento, Italy; and 3 Casa di Cura S. Pio X, Milan, Italy. Supported by a grant from the Public Health Ministry of the Italian Government (ex art. 12, comma 2, lett. b, D.L. 502/92), Rome, Italy. Address reprint requests to Mr Forti: ITC-irst, Via Sommarive 18, Povo Trento, Italy. Acknowledgments: We thank Daniela Aldovini, MD, and Goran Dvornik, MD, who participated in this study as consultant pathologists. We are grateful to Giancarlo Migliore (Leica Microsystems, Italy) for the provision of Leica equipment. We thank Michele Galvagni and all the staff of the Medical Informatics and Telemedicine Laboratory of ITC-irst for excellent technical support. References 1. Weinstein RS, Bhattacharyya AK, Graham AR, et al. Telepathology: a ten-year progress report. Hum Pathol. 1997;28:1-7. American Society of Clinical Pathologists Am J Clin Pathol 2001;116:

9 Demichelis et al / TELEMICROSCOPY SYSTEM FOR REMOTE DIAGNOSIS 2. Nordrum I. Real-time diagnoses in telepathology. Adv Clin Path. 1998;2: Black-Schaffer S, Flotte TJ. Current issues in telepathology. Telemed J. 1995;1: Weinberg DS, Allaert FA, Dusserre P, et al. Telepathology diagnosis by means of digital still images: an international validation study. Hum Pathol. 1996;27: Weinstein LJ, Epstein JI, Edlow D, et al. Static image analysis of skin specimens: the application of telepathology to frozen section evaluation. Hum Pathol. 1997;28: Dunn BE, Almagro UA, Choi H, et al. Dynamic-robotic telepathology: Department of Veterans Affairs feasibility study. Hum Pathol. 1997;28: Nordrum I, Eide TJ. Remote frozen section service in Norway. Arch Anat Cytol Pathol. 1995;43: Schwarzmann P, Binder B, Klose R, et al. Histkom: evaluation of active telepathology in field tests. Adv Clin Path. 1998;2: Della Mea V, Cataldi P, Pertoldi B, et al. Dynamic robotic telepathology: a preliminary evaluation on frozen sections, histology and cytology. J Telemed Telecare. 1999;5(suppl 1): S55-S Oberholzer M, Fischer HR, Christen H, et al. Telepathology: frozen section diagnosis at a distance. Virchows Arch. 1995;426: Winokur TS, McClellan S, Siegal GP, et al. An initial trial of a prototype telepathology system featuring static imaging with discrete control of a remote microscope. Am J Clin Pathol. 1998;110: Tsuchihashi Y, Mazaki T, Nakasato K, et al. The basic diagnostic approaches used in robotic still-image telepathology. J Telemed Telecare. 1999;5(suppl 1):S115-S Schwarzmann P, Schmid J, Schnorr C, et al. Telemicroscopy stations for telepathology based on broadband and ISDN connections. Arch Anat Cytol Pathol. 1995;43: Cataldi P, Pertoldi B, Della Mea V, et al. Validation of realtime telemicroscopy: a preliminary report on 184 cases. Adv Clin Path. 1998;2: Dunn BE, Almagro UA, Choi H, et al. Use of telepathology for routine surgical pathology: review in a test bed in the Department of Veterans Affairs. Telemed J. 1997;3: Nordrum I, Engum B, Rinde E, et al. Remote frozen section service: a telepathology project in northern Norway. Hum Pathol. 1991;22: Shimosato Y, Yagi Y, Yamagishi K, et al. Experience and present status of telepathology in the National Cancer Center Hospital, Tokyo. Zentralbl Pathol. 1992;138: Becker RL, Specht CS, Jones R, et al. Use of remote video microscopy (telepathology) as an adjunct to neurosurgical frozen section consultation. Hum Pathol. 1993;24: Della Mea V, Cataldi P, Boi S, et al. Image sampling in static telepathology for frozen section diagnosis, J Clin Pathol. 1999;52: Fujita M, Suzuki Y, Takahashi M, et al. The validity of intraoperative frozen section diagnosis based on videomicroscopy (telepathology). Gen Diagn Pathol. 1995;141: Kayser K, Drlicek M. Visual telecommunication for expert consultation of intraoperative sections. Zentralbl Pathol. 1992;138: Am J Clin Pathol 2001;116: American Society of Clinical Pathologists