P28: Advanced image analysis of immunohistochemistry on tumour specimens

Immunohistochemistry is an integral part of preclinical and clinical research activities; however, interpretation of immunohistochemical staining is a time consuming, subjective process open to inherent operator variability. Classical scoring models use a categorical scoring method, which only allows for limited statistical analysis. The aim of our work was to develop a continuous, quantitative and automated scoring system for cytoplasmic and nuclear markers, and link this to clinical outcome in patients with neoplastic disorders. Digital images of immunohistochemically stained tissue specimens were captured using an Aperio ScanScope CS slide scanner. Commercial (Aperio) algorithms were used to analyse the digital slides for expression levels of several cytoplasmic and nuclear markers. The algorithms used were based on three parameters: hue, hue width and colour saturation threshold. Receiver operator curves were constructed and optimal levels for each parameter calculated. Using these parameters, we were able to distinguish between epithelial and stromal tumour elements. These automated analytical algorithms correlated with manual scoring ( r =0.8; p =0.03) and was fully validated by a histopathologist. Automated image capture and analysis technologies offer the potential to produce objective reproducible quantitative interpretation of immunohistochemistry.