Evaluation of HER-2 expression in breast cancer using image analysis incorporating automatic feature recognition

20007 The currently favoured method for evaluation of HER-2 in routine clinical practice and research studies is immunohistochemistry (IHC). As standardised scoring of protein expression, using a scale of 0 - 3+, generates a significant number of false positives, fluorescent in situ hybridisation (FISH) is used to confirm the presence of gene amplification. Both techniques are laborious, and in the case of by-eye scoring of IHC, semi-quantitative at best. We have developed a high-throughput platform for the quantitative analysis of immunostained slides, based on fast, high-resolution scanning followed by analysis of digitised images (IA) using proprietary software, and in this study compared results obtained using this platform with those obtained using conventional methods. Archival sections of primary breast cancers collected at Nottingham City Hospital in 2004 and 2005 were stained for HER-2 (Herceptest, Dako), and evaluated by eye. Equal numbers of slides from scoring categories 0–3+ were then selected for further image analysis. The digitised images were subjected to automatic delineation to define areas of tumour parenchyma, and these areas further analysed using colour segmentation. Between 100 and 2000 fields were quantified on each section. Staining was expressed as a product of field fraction of coloured pixels and optical density. The results of IA demonstrate a continuum of staining values over the four conventional by-eye categories, with a non-linear correlation to by-eye scores. Inter-sample variation was greatest in the 3+ category, although the mean was much higher than that of the 2+ samples. The 2+ samples showed some variation, with several values not rising above baseline. To further investigate the relationship between FISH scores and IA results in the 2+ category, we analysed an additional set of slides in this group, and found a correlation between FISH and automated IA scores. We conclude that automated image analysis is sensitive to small differences in protein expression, has a wide dynamic range, and provides data superior to conventional by-eye scoring. In undecided cases HER-2 protein expression correlates with FISH data; therefore ultimately IA of HER-2 protein expression alone may provide a basis for clinical decisions. [Table: see text]