Decentralized Computer-Assisted Image Analysis for ER, PR and HER-2/neu Immunohistochemical Analysis
Liron Pantanowitz ; Baystate Medical Center/Tufts University School of Medicine;
Content:
Accurate determination of estrogen receptor (ER), progesterone receptor (PR), and HER-2/neu status in breast cancer is required to guide appropriate therapy. Computer-assisted image analysis (CAIA) has been shown to provide an accurate and reproducible method to score immunohistochemical staining. Several guidelines support the use of CAIA for breast marker quantification, with appropriate pathologist supervision. The aim of this study was to implement CAIA for this purpose in a surgical pathology department spanning distant medical centers to mimic daily practice.
Technology:
Networked multimedia workstations (Dell Optiplex 745 personal computers), Spot Insight digital microscope cameras (Diagnostic Instruments), Pathiam web-based application (BioImagene), server (Oracle application and image database).
Design:
Digital camera settings were standardized (calibrated) and 3-5 microscopic field of views (29 cases) stored in jpeg format, using 20x eyepiece magnification, were acquired for ER, PR and HER-2/neu immunohistochemical stained breast (invasive and in situ) carcinomas. Nuclear immunostaining for ER and PR was analyzed using the Allred scoring system (proportion + intensity = total score) and membranous HER-2/neu staining per ASCO/CAP recommendations (scored 0, 1+, 2+, 3+). CAIA was carried out with defined control parameter sets established for each run. Fluorescence in situ hybridization (FISH) for HER-2/neu was obtained in a subset of cases. Score result and analysis time (minutes) between manual (pathologist) and CAIA were compared.
Results:
There was excellent concordance for ER, PR and HER-2/neu scores between pathologists and CAIA. For HER-2/neu results there was good correlation between manual scoring, CAIA and FISH. Rare discordant results (3 cases) for ER and PR were attributed to non-specific cytoplasmic staining of tumor cells. CAIA, including image acquisition time, was considerably longer to perform than manual scoring.
Conclusion:
Decentralized CAIA for immunohistochemistry designed to mimic daily surgical pathology workflow in practice is feasible. However, decentralized image acquisition from individual workstations requires vigilant standardization, entails a cumbersome workflow and is time consuming. Many of these issues could be overcome by using a central system with whole slide imaging.
