TrainableImmunohistochemicalHER2/neu Image Analysis. A Multisite Performance Study Using 260 Breast Tissue Specimens

TrainableImmunohistochemicalHER2/neu Image Analysis A Multisite Performance Study Using 260 Breast Tissue Specimens Aziza Nassar, MD, MPH; Cynthia Cohen, MD; Sally S. Agersborg, MD, PhD; Weidong Zhou, MD; Kathleen A. Lynch, MD; Maher Albitar, MD; Edward A. Barker, MD; Burton L. Vanderbilt, MD; Jim Thompson, PhD; Eugene R. Heyman, PhD; Holger Lange, PhD; Allen Olson, PhD; Momin T. Siddiqui, MD N Context. Aperio Technologies, Inc (Vista, California) provides a new immunohistochemistry (IHC) HER2 Image Analysis (IA) system that allows tuning of the intensity thresholds of the HER2/neu scoring scheme to adapt to the staining characteristics of different reagents. Objective. To compare the trainable IHC HER2 IA system for different reagents to conventional manual microscopy (MM) in a multisite study. Design. Two hundred sixty formalin-fixed, paraffinembedded breast cancer specimens from 3 clinical sites were assayed: 180 specimens stained with Dako s HercepTest (Carpinteria, California), and 80 specimens stained with Ventana s PATHWAY HER-2/neu (Tucson, California). At each site, 3 pathologists performed a blinded reading of the glass slides with the use of a light microscope. The glass slides were then scanned and after a wash-out period and randomization, the same pathologists outlined a representative set of tumor regions to be analyzed by IHC HER2 IA. Each of the methods, MM and IA, was evaluated separately and comparatively by using k statistics of negative HER2/neu scores (0, 1+) versus equivocal HER2/neu scores (2+) versus positive HER2/ neu scores (3+) among the different pathologists. Results. k Values for IA and MM were obtained across all sites. MM: 0.565 0.864; IA: 0.895 0.947; MM versus IA: 0.683 0.892 for site 1; MM: 0.771 0.837; IA: 0.726 0.917; MM versus IA: 0.687 0.877 for site 2; MM: 0.463 0.674; IA: 0.864 0.918; MM versus IA: 0.497 0.626 for site 3. Conclusion. Aperio s trainable IHC HER2 IA system shows substantial equivalence to MM for Dako s HercepTest and Ventana s PATHWAY HER-2/neu at 3 clinical sites. Image analysis improved interpathologist agreement in the different clinical sites. (Arch Pathol Lab Med. 2011;135:896 902) HER2/neu is a proto-oncogene located on the long arm of human chromosome 17 (17q11.2 q12). 1 3 It is amplified in 15% to 20% of breast cancers, leading to Accepted for publication October 4, 2010. From the Department of Pathology and Laboratory Medicine, Emory University School of Medicine, Atlanta, Georgia (Drs Nassar and Cohen); the Hematology Oncology Department (Drs Agersborg, Zhou, and Lynch) and Hematology Department (Dr Albitar), Quest Diagnostics Nichols Institute, San Juan Capistrano, California; the Pathology Department, Medical Laboratory Associates, Seattle, Washington (Dr Barker); the Pathology Department, Vanderbilt Medical Services PC, Bellingham, Washington (Dr Vanderbilt); the Pathology Department, Pathogenesys LLC, San Juan Capistrano, California (Dr Thompson); the Department of Health Sciences, Biostatistics, University of Maryland, College Park (Dr Heyman); Aperio, Vista, California (Drs Lange and Olson); and the Department of Pathology and Laboratory Medicine, Emory University Hospital, Atlanta, Georgia (Dr Siddiqui). Dr Vanderbilt is now with the Department of Pathology, Bartlett Regional Hospital, Juneau, Alaska. Dr Nassar is now with the Department of Pathology, Mayo Clinic, Rochester, Minnesota. All pathologists (Drs Nassar, Cohen, Siddiqui, Agersborg, Zhou, Lynch, Barker, Vanderbilt, Albitar, and Thompson) who participated in the study received a monetary fund of $3000 and a laptop computer for implementing the research. All pathologists were paid by Aperio Inc. Dr Heyman, the biostatistician, was paid on an hourly basis as a biostatistics consultant. Drs Olson and Lange are employees of Aperio. Reprints: Aziza Nassar, MD, Department of Laboratory Medicine and Pathology, Mayo Clinic, 200 First Street SW, Rochester, MN 55905 (e-mail: nassar.aziza@mayo.edu). increased expression of the protein gene product. 4,5 In 90% to 95% of tumors, HER2/neu protein expression is the result of gene amplification detectable by fluorescence in situ hybridization (FISH). 2,6 HER2/neu overexpression in the absence of amplification is rare (3%). 7 HER2/neu status in the primary breast carcinoma is usually maintained in metastatic lesions. 8 Overexpression of HER2/neu is associated with poor prognosis, response to trastuzumab treatment, resistance to several chemotherapeutic agents, early recurrence, and decreased disease-free and overall survival in lymph node positive patients. 2 In addition to its prognostic role, HER2/neu has now become more important as a predictive marker of treatment response to trastuzumab, a human murine monoclonal antibody to the HER2/neu protein. 9 In 1998, trastuzumab (marketed as Herceptin, Genentech Inc, South San Francisco, California) was approved for the targeted therapy of patients with HER2/neu-overexpressing metastatic breast cancer by the US Food and Drug Administration (FDA). It has also recently been shown to be very effective in the adjuvant setting. 2,9 The 2 methodologies in current clinical use to assess HER2/neu status in breast cancer are FISH to evaluate HER2/neu gene amplification and immunohistochemistry (IHC) to detect protein overexpression. 1 According to the College of American Pathologists (CAP) and American 896 Arch Pathol Lab Med Vol 135, July 2011 Aperio and HER2 Image Analysis Nassar et al

Society of Clinical Oncology (ASCO) guidelines, HER2/ neu IHC 2+ is regarded as equivocal and should be followed by testing for HER2/neu gene amplification by FISH. 4 A pathologist and/or technologist visually quantifies FISH results. The results are thus vulnerable to subjective interpretation, and the interpreter s experience plays a role in result reproducibility. Results of IHC and FISH can be influenced by duration and method of tissue fixation, decalcification, antigen retrieval method, antibody/probe type, or by differences in the interpretation of staining intensity by the observer. 4 Several different HER2/neu antibodies are available: the FDA-approved Dako (Dako, Glostrup, Denmark) Hercep Test 10 and the Ventana (Tucson, California) PATHWAY anti-her2/neu (4B5) rabbit monoclonal antibody are mostly used and provide excellent sensitivity, specificity, and interlaboratory reproducibility. 1 It has been suggested that the use of digital microscopy improves the accuracy and interobserver reproducibility of HER2/neu IHC analysis. Digital measurement of staining intensity is more accurate than measurement with the human eye because it is not influenced by factors such as the ambient light, pathologist fatigue, or interobserver and intraobserver variation. 3 Consistent objective and reproducible results of HER2/neu assessment can be generated by a number of available automated scoring systems such as the automated cellular imaging system (ACIS) (ChromaVision, Inc, San Juan Capistrano, California) 11,12 optimized for use with Dako HercepTest, Micrometastasis Detection System (MDS, Applied Imaging, San Jose, California), 13 Extended Slide Wizard (Tripath Imaging, Inc, Burlington, North Carolina), and others. 14 16 To be widely accepted for clinical use, digital pathology systems need to be cleared/approved by the FDA, and performance studies must demonstrate this new technology to the clinical community. This study evaluates the performance of Aperio Technologies (Vista, California) Digital Pathology Platform and compares it with that of manual microscopy (MM), while using 2 different FDAapproved antibodies, Dako HercepTest and Ventana PATHWAY anti-her2-neu (clone CB11) mouse monoclonal antibody, for evaluating IHC in 3 different clinical centers. MATERIALS AND METHODS Three Clinical Laboratory Improvement Amendments of 1988 qualified clinical sites participated in the study. Before their participation in the study, each clinical site obtained exemption status from an institutional review board. A multisite performance study was conducted at an academic center (site 1), a reference laboratory (site 2), and a private group practice (site 3). Aperio Technologies trainable IHC HER2 Image Analysis (IA) system was compared to conventional MM, with different reagents. Each institution performed its own study, with no exchange of slides between the different centers. Two hundred and sixty formalin-fixed, paraffin-embedded invasive breast carcinoma specimens from these 3 clinical sites were assayed: 80 specimens with almost equal HER2/neu score distribution from clinical site 1; 100 routine specimens from clinical site 2 all stained with Dako HercepTest; and 80 specimens with almost equal HER2/neu score distribution from clinical site 3, stained with Ventana PATHWAY HER2-neu. The specimens at the first clinical site were selected retrospectively by their clinical scores on file to provide an equal distribution of slides for the different HER2/neu scores. The specimens at the second clinical site were prospective routine specimens taken from its clinical operation, representing the true target population of cases in a typical clinical setting. The specimens at the third clinical site were selected by their clinical scores on file to provide an equal distribution of slides for the different HER2/neu scores. All specimens at the first and second clinical site were immunohistochemically stained with Dako in vitro diagnostic, FDA-approved HerceptTest (A085). All specimens at the third clinical site were immunohistochemically stained with Ventana in vitro diagnostic, FDA-approved PATHWAY HER2-neu (Clone CB11). At each site, a 20-slide representative training set with scores from 3 pathologists was used to automatically train the IA algorithm. At each site, 3 pathologists, using their microscopes, performed a blinded reading of the glass slides and reported the HER2/neu score (0, 1+, 2+, or 3+) for each slide. The glass slides were then scanned with an 320 objective and randomized. After a wash-out period exceeding 1 week, the same 3 pathologists outlined a representative set of tumor regions to be analyzed by IA (Figure 1). The IA system reported the HER2/neu score for each of the 3 pathologists for each of the slides. The IA itself was run in batch mode and in blinded fashion to avoid influencing the pathologists in their choice of the tumor regions. The whole process was repeated, this time with the different clinical samples (80 specimens from each of clinical sites 1 and 3; and 100 specimens from clinical site 2). The same 3 pathologists at each clinical site used Aperio s ScanScope Systems remote editing capability to outline a representative set of tumor regions for each digital slide. The pathologists were blinded to each other s annotations of tumor region outlines. Image Analysis was run for each slide, resulting in a separate IA score for each of the 3 pathologists. The IA algorithm reported the HER2/neu score as 0, 1+, 2+, or3+ for each of the digital slides. For HER2/neu tissue scoring applications, the ScanScope Aperio digital microscope locates tissues on a slide and generates a scanned image of the entire tissue (Figure 2). At review, the pathologist views this image on a computer monitor and selects multiple tissue regions of appropriately stained invasive carcinoma for scoring (Figure 3). The system then generates a series of scores for these areas, including both individual region scores and the average score of selected regions. In HER2/neu IHC evaluation, the score is the average brown pixel intensity, determined by using a color threshold designed to measure the brown color associated with the 3, 39-diaminobenzidine tetrahydrochloride (DAB) stain. The specific algorithm selects those pixels within the selected tissue regions that meet the brown DAB color threshold. This color selection chooses pixels that are associated with extracellular domains of the plasma membrane. The ScanScope Aperio digital microscope calculates the HER2/ neu IHC intensity score on a scale of 0 to 255, and also provides a binned score that relates to the HercepTest scoring system (1+, 2+, and 3+). Owing to the more quantitative data of ScanScope Aperio digital microscopy, this score can be reported in tenths of integer units (eg, 1.3, 2.1). Cases with an average score of at least 2.2 are considered to have HER2/neu protein overexpression. A separate/smaller set of digital slides was used to establish the cut points. Then, the algorithm (with the determined cut points) was applied independently of the study data. Other details of the hardware and software for the ScanScope Aperio digital microscopy system are available online (http:// www.aperio.com; accessed August 8, 2010). Immunohistochemistry From each tissue block, 4-mm sections were cut, deparaffinized in xylene, and dehydrated through alcohol changes. For Dako HercepTest, processing was according to the manufacturers instructions with water-bath antigen retrieval as follows: slides were immersed in 10 mmol of citrate buffer per liter in a calibrated water bath (required temperature 95uC 99uC). They were then incubated for 40 (61) minutes at 95uC to99uc. After decanting the epitope retrieval solution, the sections were rinsed in the wash buffer, and later soaked in the buffer for 5 to Arch Pathol Lab Med Vol 135, July 2011 Aperio and HER2 Image Analysis Nassar et al 897

Figure 1. Schematic diagram of the study design. Figure 2. Captured image of a HER2/neuimmunostained slide by Aperio s ImageScope (Vista, California) on a computer monitor. 898 Arch Pathol Lab Med Vol 135, July 2011 Aperio and HER2 Image Analysis Nassar et al

Figure 3. Captured image of a HER2/neuimmunostained slide on a computer monitor with annotations by the pathologist. 20 minutes before staining. The slides were loaded onto the Dako autostainer using the HercepTest. In the autostainer, the slides were rinsed, followed by 200 ml of peroxidase-blocking reagent for 5 minutes rinsing, then 200 ml of primary anti-her2/neu protein (or negative control reagent) for 30 minutes, rinsed twice, and finally immersed in 200 ml of substrate-chromogen solution (DAB) for 10 minutes. The slides were then removed from the autostainer, counterstained with hematoxylin, and finally coverslipped. Known positive, negative, and intermediate value controls were run with each patient set. For immunostaining with the Ventana Benchmark XT staining system, sections were transferred to Ventana wash solution. Endogenous peroxidase activity was blocked in 3% hydrogen peroxide. Slides were then incubated with Ventana PATHWAY anti-her2/neu (CB11) mouse monoclonal antibody at 37uC for 32 minutes and developed in DAB for 10 minutes. Finally, sections were counterstained with hematoxylin and mounted. Per ASCO/CAP guidelines, 4 HER2/neu IHC protein expression status was classified by applying the following criteria.. Negative for HER2/neu protein: IHC staining of 0 or 1+, with absence of or weak, incomplete membrane staining in any proportion of tumor cells.. Equivocal for HER2/neu protein: IHC staining of 2+, with complete membrane staining that is either nonuniform or weak in intensity but with obvious circumferential distribution in at least 10% of cells.. Positive for HER2/neu protein: IHC staining of 3+, with uniform intense chicken-wire circumferential membrane staining of more than 30% of invasive tumor cells.. The Aperio HER2 IA scoring system was as follows: negative, below 1.8; equivocal, from 1.8 to less than 2.2; positive, greater than or equal to 2.2. Statistical Analysis The statistical analyses for all 3 sites, for each of the methods, included analysis of MM, IA, and comparative analysis between the 2 methods (MM and IA). For each of the methods, MM and IA were evaluated separately and comparatively by using percentage agreement (PA), with an exact 95% confidence interval (CI), of a trichotomous categorization of the HER2/neu scores that combined 0 and 1+ as negative cases, 2+ as equivocal, and 3+ as positive cases. Simple k statistics with 95% CI were estimated for each agreement (interobserver and intraobserver) table, which provides the degree of agreement above that expected by chance alone. Statistical analyses were performed with SAS software version 9.2 (Chicago, Illinois). The interpretation of the k statistics is as follows: below 0, less than chance agreement; 0.01 to 0.20, slight agreement; 0.21 to 0.40, fair agreement; 0.41 to 0.60, moderate agreement; 0.61 to 0.80, substantial agreement; and 0.81 to 0.99, almost perfect agreement. RESULTS Comparable PA values were obtained for MM and IA for digital HER2/neu slides. Tables 1 through 6 show PA and k statistics along with an exact 95% CI for the clinically relevant trichotomous outcome of negative (0 and 1+) versus equivocal (2+) versus positive (3+) HER2/ neu scores. Each table presents the results for each of the methods MM, IA, and comparatively between MM and IA for the 3 different clinical sites, each with its 3 different pathologists. Percentage agreement values between pathologists using MM (65.0% 91.3%) (Table 1) were comparable to PA values between MM and IA (65.0% 90.0%) (Table 5), as based on the tumor regions outlined by the pathologist. The k statistics show that there is moderate to almost perfect agreement (0.481 0.832) between the different pathologists using manual microscopy in the different clinical sites (Table 2). The interpathologist agreement for clinical site 2 is better than that for the other clinical sites (Table 2). The PA values for IA between pathologists, based on the tumor regions outlined by the pathologists (85.0% 94.0%) (Table 3), were more or less comparable to Arch Pathol Lab Med Vol 135, July 2011 Aperio and HER2 Image Analysis Nassar et al 899

Table 1. Manual Microscopy: Interpathologist Agreement Percentage Agreement (95% CI), % Clinical site 1 91.3 (82.8 96.4) 77.5 (66.8 86.1) 76.3 (65.4 85.1) Clinical site 2 84.0 (75.3 90.6) 82.0 (73.1 89.0) 90.0 (82.4 95.1) Clinical site 3 65.0 (53.5 75.3) 75.0 (64.1 84.0) 72.5 (61.4 81.9) Table 2. k Statistics for Interpathologist Agreement for Manual Microscopy k (95% CI) Clinical site 1 0.832 (0.716 0.949) 0.638 (0.496 0.781) 0.615 (0.471 0.759) Clinical site 2 0.723 (0.603 0.842) 0.672 (0.544 0.801) 0.814 (0.705 0.923) Clinical site 3 0.481 (0.343 0.618) 0.626 (0.488 0.765) 0.570 (0.428 0.713) Table 3. Image Analysis: Interpathologist Agreement Percentage Agreement (95% CI), % Clinical site 1 91.3 (82.8 96.4) 92.5 (84.4 97.2) 88.8 (79.7 94.7) Clinical site 2 85.0 (76.5 91.4) 94.0 (87.4 97.8) 87.0 (78.8 92.9) Clinical site 3 86.3 (76.7 92.9) 87.5 (78.2 93.8) 88.8 (79.7 94.7) Table 4. k Statistics for Interpathologist Agreement for Image Analysis k (95% CI) Clinical site 1 0.841 (0.736 0.947) 0.866 (0.764 0.968) 0.796 (0.676 0.915) Clinical site 2 0.725 (0.596 0.853) 0.892 (0.809 0.975) 0.761 (0.640 0.882) Clinical site 3 0.789 (0.675 0.902) 0.806 (0.695 0.917) 0.826 (0.720 0.931) Table 5. Manual Microscopy Versus Image Analysis: Individual Pathologist s Agreement Pathologist 1 Pathologist 2 Pathologist 3 Percentage Agreement (95% CI), % Clinical site 1 87.5 (78.2 93.8) 87.5 (78.2 93.8) 80.0 (69.6 88.1) Clinical site 2 90.0 (82.4 95.1) 79.0 (69.7 86.5) 88.0 (80.0 93.6) Clinical site 3 78.8 (68.2 87.1) 66.3 (54.8 76.4) 65.0 (53.5 75.3) Table 6. k Statistics for Individual Pathologist s Agreement for Manual Microscopy Versus Image Analysis Pathologist 1 Pathologist 2 Pathologist 3 k (95% CI) Clinical site 1 0.771 (0.639 0.903) 0.763 (0.637 0.890) 0.680 (0.545 0.815) Clinical site 2 0.825 (0.725 0.926) 0.619 (0.478 0.759) 0.773 (0.658 0.889) Clinical site 3 0.677 (0.545 0.809) 0.473 (0.328 0.619) 0.472 (0.324 0.620) the PA values between pathologists using MM (65.0% 91.3%) (Table 1). The k statistics for interpathologist agreement improved with the use of image analysis (Table 4). In fact, the k statistics show that there is substantial to almost perfect agreement (0.725 0.892) between the different pathologists in the 3 clinical sites (Table 4). The interpathologist agreement improved dramatically for clinical sites 1 and 3 (Table 4) with the use of image analysis. These data show that if pathologists rely on IA results instead of consulting another pathologist, IA would provide them with better agreement (agreement between 900 Arch Pathol Lab Med Vol 135, July 2011 Aperio and HER2 Image Analysis Nassar et al

MM and IA is in most cases higher than the agreement between pathologists) and a more reliable score (the agreement between IA, when used by different pathologists, is in most cases higher than the agreement between pathologists). The highest IA variations are introduced by interpathologist variations, which still yield an excellent PA, ranging from 85.0% to 94.0% in the comparison study, in terms of the clinically relevant negative (0 and 1+) versus equivocal (2+) versus positive (3+) HER2/neu scores (Table 3). The intrapathologist agreement was moderate to almost perfect (0.472 0.825) (Table 6). COMMENT Our results support the hypothesis that automation improves interobserver IHC quantitation results of HER2/neu when compared to MM. We demonstrate that reading digital HER2/neu slides on a computer monitor, using Aperio Technologies Digital Pathology Platform, is a substantial equivalence to conventional MM and therefore can be used as an alternative to the conventional microscope. This is even the case with multiple manufacturers HER2/neu IHC reagents, each expressing different staining characteristics, and with different pathologists. This study was done in 3 different institutions, and each institution had 3 different pathologists, with variable levels of experience for evaluating HER2/neu immunostain. These differences in experience have accounted for some of the variations that are seen in the percentage agreement, specifically in Table 1. The other factor that accounts for these variations is the different hot spot that each pathologist chose for evaluating the staining. The human eye is not perfect in detecting subtle differences in intensity, whereas the image analysis does have this capability. Digital microscopy improves the accuracy and reliability of HER2/neu IHC analysis, when compared to semiquantitative scoring by a pathologist. 3 As the same stained slides were used, this most likely reflects errors in manual interpretation and not IHC reagent limitations, as demonstrated by our study. We have shown that IA improved interobserver agreement between the different pathologists. Furthermore, the variability in intrapathologist agreement could be attributed to the experience level of the pathologist, familiarity with the use of digitalized images, and the choice of the different fields used for annotations. False-positive HercepTest results have been reported in 12% to 23% of cases. 13,17 HercepTest has been shown to give false-negative results in up to 28% of HER2/neu FISH-positive cases. 13 The CAP/ASCO HER2/neu guidelines recommend that laboratories show 95% concordance with another validated test, such as FISH, for HER2/neu IHC negative results and 3+ staining results. 4 Another automated IA system that is successful in HER2/neu testing is the automated cellular imaging system (ACIS) (Dako). ACIS is known to be more accurate than visual scoring and is reported to have 91% to 95% concordance with FISH results when evaluating HER2/ neu overexpression in whole tissue sections. 3,11,12,18 21 Using ACIS for HER2/neu expression helps pathologists establish consistency, especially among several pathologists, and provides clinicians with objective results for triaging patients for directed trastuzumab (Herceptin) therapy, specifically following the new CAP/ASCO guidelines for HER2/neu analysis. 4 Since the evaluation of staining intensity and percentage of cells with complete membrane positivity is subjective, the interobserver variability tends to be higher for scoring HER2/neu 2+ cases, and for discriminating 1+ and 2+, or 2+ and 3+ cases. 3 The percentage of disagreement in intraobserver reproducibility ranges from 0.9% to 3.7%. 3 The interobserver agreement is excellent for categorized variables (0, 1+ versus 2+ versus 3+) between 2 pathologists (k 5 0.929; 95% CI, 0.909 0.946). 3 Using the ACIS system, Bloom and Harrington 18 were able to eliminate interobserver variability in HER2/neu scoring by IHC. 18 They showed that 10 pathologists scoring the same IHC slides (129 cases of invasive breast carcinoma) with the assistance of digital microscopy improved concordance with FISH to 93% (k 5 0.86; P,.001), compared with 71% (k 5 0.51) for manual IHC analysis. 18 Similarly, using the ACIS system, Wang et al 20 found improved concordance of IHC with FISH, as did Tawfik et al 12 (94% concordance between IHC-ACIS [$2.2] and FISH [$2.0]) and Lüftner et al 22 (88.5% concordance, k 5 0.68, category good ). Lüftner et al 22 also showed very good concordance (95.1%, k 5 0.85) between manual interpretations and the automated IHC assay. The use of computer-aided microscopy has been suggested as a way to improve interobserver reproducibility in immunohistochemical evaluation. 23 26 Computeraided microscopy involves the digitization of stained tissue and the automated evaluation of immunohistochemistry with image analysis. This makes for a more objective assay than subjective manual microscopy, eliminating lack of reader experience. In the case of HER2/neu, the CAP/ASCO guidelines recognize image analysis as an effective tool for achieving consistent interpretation of immunohistochemical staining, provided that a pathologist confirms the result. 4 The latter ensures that invasive cancer is assessed and not benign breast tissue or ductal carcinoma in situ. Computer-aided quantitative assessment of immunohistochemical staining has potentially several benefits. It can provide a true continuous and reproducible assessment of staining. The human eye has difficulty distinguishing subtle differences in staining intensity using a continuous scale. 26 Consequently, scoring systems tend to be nominal (ie, 0, 1+, 2+, and 3+). Studies for HER2/neu have shown that accurate distinction between nominal categories is difficult and often arbitrary, and this difficulty contributes to a significant lack of reproducibility. 27 Automated systems can consistently preselect stained areas and extract a score from them or point the same areas to different observers; the selection of different areas to be assessed by different observers has been identified as a source of interobserver variability. 28 Those hot spots are identified on the digital slides. Automated systems could be used for screening of strongly positive or strongly negative slides, such that the ever-increasing reading load of a pathologist will be relieved from obvious cases. In addition to increased reproducibility, automated systems have the potential to increase prognostic accuracy by revealing differences in biomarker expression that are not discernible to the pathologist owing to their inability to distinguish between fine levels of expression. 26 Digital image analysis can detect subtle changes in the intensity of biomarker expression that are indiscernible to the human eye. This accounts for the improvement that is seen in this study Arch Pathol Lab Med Vol 135, July 2011 Aperio and HER2 Image Analysis Nassar et al 901

using image analysis. Those hot spots that are identified by the IA system are always the same despite the different readers using the system. Aperio Technologies Inc provides trainable IHC HER2 IA, which allows adaptation to the staining characteristics of different reagents (Dako HercepTest, Ventana PATH- WAY) and correlation of the analysis results to a scoring standard, and is of substantial equivalence to conventional MM. Compared to manual microscopy, Image Analysis by Aperio ScanScope has improved the concordance among 3 pathologists in each of 3 different institutions. References 1. Powell WC, Hicks DG, Prescott N, et al. A new rabbit monoclonal antibody (4B5) for the immunohistochemical (IHC) determination of the HER2 status in breast cancer: comparison with CB11, fluorescence in situ hybridization (FISH) and interlaboratory reproducibility. Appl Immunohistochem Mol Morphol. 2007; 15(1):94 102. 2. Slamon DJ, Leyland-Jones B, Shak S, et al. Use of chemotherapy plus a monoclonal antibody against HER2 for metastatic breast cancer that overexpresses HER2. N Engl J Med. 2001;344(11):783 792. 3. Turashvili G, Leung S, Turbin D, et al. Inter-observer reproducibility of HER2 immunohistochemical assessment and concordance with fluorescent in situ hybridization (FISH): pathologist assessment compared to quantitative image analysis. BMC Cancer. 2009;9:165. 4. Wolff AC, Hammond ME, Schwartz JN, et al. American Society of Clinical Oncology/College of American Pathologists guideline recommendations for human epidermal growth factor receptor 2 testing in breast cancer. J Clin Oncol. 2007;25(1):118 145. 5. Carlson RW, Moench SJ, Hammond ME, et al. HER2 testing in breast cancer: NCCN Task Force report and recommendations [quiz in J Natl Compr Canc Netw. 2006;4(suppl 3):S23 S24]. J Natl Compr Canc Netw. 2006;4(suppl 3):S1 S22. 6. Pauletti G, Dandekar S, Rong H, et al. Assessment of methods for tissuebased detection of the HER-2/neu alteration in human breast cancer: a direct comparison of fluorescence in situ hybridization and immunohistochemistry. J Clin Oncol. 2000;18(21):3651 3664. 7. Persons DL, Bui MM, Lowery MC, et al. Fluorescence in situ hybridization (FISH) for detection of HER-2/neu amplification in breast cancer: a multicenter portability study. Ann Clin Lab Sci. 2000;30(1):41 48. 8. Park K, Han S, Kim HJ, Kim J, Shin E. HER2 status in pure ductal carcinoma in situ and in the intraductal and invasive components of invasive ductal carcinoma determined by fluorescence in situ hybridization and immunohistochemistry. Histopathology. 2006;48(6):702 707. 9. Laudadio J, Quigley DI, Tubbs R, Wolff DJ. HER2 testing: a review of detection methodologies and their clinical performance. Expert Rev Mol Diagn. 2007;7(1):53 64. 10. Rhodes A, Jasani B, Anderson E, Dodson AR, Balaton AJ. Evaluation of Her-2/neu immunohistochemical assay sensitivity and scoring on formalin-fixed and paraffin-processed cell lines and breast tumors: a comparative study involving results from laboratories in 21 countries. Am J Clin Pathol. 2002; 118(93):408 417. 11. Ciampa A, Xu B, Ayata G, et al. HER-2 status in breast cancer: correlation of gene amplification by FISH with immunohistochemistry expression using advanced cellular imaging system. Appl Immunohistochem Mol Morphol. 2006; 14(2):132 137. 12. Tawfik OW, Kimler BF, Davis M, et al. Comparison of immunohistochemistry by automated cellular imaging system (ACIS) versus fluorescence insitu hybridization in the evaluation of HER2 expression in primary breast carcinoma. Histopathology. 2006;48 (3):258 267. 13. Ellis CM, Dyson MJ, Stephenson TJ, Maltby EL. HER2 amplification status in breast cancer: a comparison between immunohistochemical staining and fluorescence in situ hybridisation using manual and automated quantitative image analysis scoring techniques. J Clin Pathol. 2005;58(7):710 714. 14. Hatanaka Y, Hashizume K, Nitta K, Kato T, Itoh I, Tani Y. Cytometrical image analysis for immunohistochemical hormone receptor status in breast carcinomas. Pathol Int. 2003;53(10):693 699. 15. Joshi AS, Sharangpani GM, Porter K, et al. Semi-automated imaging system to quantitate Her-2/neu membrane receptor immunoreactivity in human breast cancer. Cytometry A. 2007;71(5):273 285. 16. Skaland I, Øvestad I, Janssen EA, et al. Comparing subjective and digital image analysis HER2/neu expression scores with conventional and modified FISH scores in breast cancer. J Clin Pathol. 2008;61(1):68 71. 17. Lebeau A, Deimling D, Kaltz C, et al. Her-2/neu analysis in archival tissue samples of human breast cancer: comparison of immunohistochemistry and fluorescence in situ hybridization. J Clin Oncol. 2001;19(2):354 363. 18. Bloom K, Harrington D. Enhanced accuracy and reliability of HER-2/neu immunohistochemical scoring using digital microscopy. Am J Clin Pathol. 2004; 121(5):620 630. 19. Anderson JM, Ariga R, Govil H, et al. Assessment of Her-2/Neu status by immunohistochemistry and fluorescence in situ hybridization in mammary Paget disease and underlying carcinoma. Appl Immunohistochem Mol Morphol. 2003; 11(2):120 124. 20. Wang S, Hossein Saboorian M, Frenkel EP, et al. Aneusomy 17 in breast cancer: its role in HER-2/neu protein expression and implication for clinical assessment of HER-2/neu status. Mod Pathol. 2002;15(2):137 145. 21. Minot DM, Kipp BR, Root RM, et al. Automated cellular imaging system III for assessing HER2 status in breast cancer specimens development of a standardized scoring method that correlates with FISH. Am J Clin Pathol. 2009; 132(1):133 138. 22. Lüftner D, Henschke P, Kafka A, et al. Discordant results obtained for different methods of HER-2/neu testing in breast cancer a question of standardization, automation and timing. Int J Biol Markers. 2004;19(1):1 13. 23. Seidal T, Balaton AJ, Battifora H. Interpretation and quantification of immunostains. Am J Surg Pathol. 2001;25(9):1208 1210. 24. Braunschweig T, Chung JY, Hewitt SM. Perspectives in tissue microarrays. Combinatorial Chem High Throughput Screening. 2004;7(6):575 585. 25. Braunschweig T, Chung JY, Hewitt SM. Tissue microarrays: bridging the gap between research and the clinic. Expert Rev Proteomics. 2005;2(3):325 326. 26. Camp RL, Chung GG, Rimm DL. Automated subcellular localization and quantification of protein expression in tissue microarray. Nature Med. 2002; 8(11):1323 1327. 27. Paik S, Bryant J, Tan-Chiu E, et al. Real-world performance of Her2 testing National Surgical Adjuvant Breast and Bowel Project experience. J Nat Cancer Inst. 2002;94(11):852 854. 28. van Diest PJ, van Dam P, Henzen-Logmans SC, et al. A scoring system for immunohistochemical staining: consensus report of the task force for basic research of the EORTC-GCCG. Clin Pathol. 1997;50(10):801 804. 902 Arch Pathol Lab Med Vol 135, July 2011 Aperio and HER2 Image Analysis Nassar et al