HER2 Immunohistochemistry: Workflow Experience with Image Analysis Based Interpretation of CB11 and 4B5 Clones.
David J Dabbs MD; University of Pittsburgh; Urvashi Surti PhD; University of Pittsburgh; Jeffrey L Fine MD; University of Pittsburgh; Rohit Bhargava MD; University of Pittsburgh;
Content:
HER2/neu immunohistochemistry (IHC) testing is under increased scrutiny in terms of laboratory accreditation, to ensure accurate and reproducible assessment of breast cancer patients for potential Trastuzumab therapy. Image analysis (IA) could improve HER2/neu testing by improving scoring reliability and by increasing its automation. This report details our IA experience during a recently completed validation trial, and how this will affect our imminent implementation of IA into production workflow.
Technology:
IHC was performed using an automated platform (Benchmark XT, Ventana Medical Systems, Tucson, AZ). The IA system (VIAS, Ventana) assigned each stain a value (0 to 3.50), which was then automatically rounded to the nearest whole number score (0, 1+, 2+, or 3+), which is intended to be directly analogous to pathologists scores: 0/1+ are negative; 2+ is weakly-positive (FISH required); and 3+ is positive.
Design:
CB11 and 4B5 Her2/neu antibody clones (Ventana) were validated against Fluorescent in-situ hybridization (FISH) (Vysis, Downers Grove, IL) using IA. A pathologist manually entered case data then performed IA. Afterward, system-generated PDF-formatted reports were printed and kept with other case paperwork. When in production, IA-derived scores will be dictated into a detailed HER2/neu results quick-text comment within the pathology report, as is currently done with manual interpretations.
Results:
118 cases were analyzed; a subset of these had corresponding FISH data. CB11 was 100% concordant with FISH (n=52). 4B5 was approximately 95% concordant (n=56 with 3 over-calls of weakly-positive stains); manually raising the 3+ cut-off increased concordance to 100%. Within a selected field IA could automatically differentiate tumor from non-tumor cells; it could not differentiate infiltrating tumor from in-situ tumor, nor could it find a field automatically.
Conclusion:
This is an exciting, validated application that will shortly be moved into production. As our practice transitions from CB11 to 4B5 for HER2/neu testing, IA should minimize scoring variability. IHC system bar-code functionality (supported but not yet implemented) can increase automation of IA; but a bi-directional LIS/IA interface could permit automation of: results reporting, workflow routing, and IA billing. Future systems should be more independent from pathologists as are other laboratory devicesas such, IA has the potential to greatly augment the scarcest resource in surgical pathology practice (the pathologist).
