Effects of tissue decalcification on the quantification of breast cancer biomarkers by digital image analysis

Background: Recent technical advances in digital image capture and analysis greatly improve the measurement of protein expression in tissues. Breast cancer biomarkers provide a unique opportunity to utilize digital image analysis to evaluate sources of variability that are caused by the tissue preparation, in particular the decalcification treatment associated with the analysis of bone metastatic breast cancer, and to develop methods for comparison of digital data and categorical scores rendered by pathologists. Methods: Tissues were prospectively decalcified for up to 24 hours and stained by immunohistochemistry (IHC) for ER, PR, Ki-67 and p53. HER2 positive breast cancer sections were retrieved from the pathology archives, and annotated with the categorical HER2 expression scores from the pathology reports. Digital images were captured with Leica and Aperio slide scanners. The conversion of the digital to categorical scores was accomplished with a Gaussian mixture model and tested for accuracy by comparison to clinical scores. Results: We observe significant effects of the decalcification treatment on common breast cancer biomarkers that are used in the clinic. ER, PR and p53 staining intensities decreased 15 - 20%, whereas Ki-67 decreased > 90% during the first 6 hrs of treatment and stabilized thereafter. In comparison with the Aperio images, pixel intensities generated by the Leica system are lower. A novel statistical model for conversion of digital to categorical scores provides a systematic approach for conversion of nuclear and membrane stains and demonstrated a high concordance with clinical scores. Conclusion: Digital image analysis greatly improves the quantification of protein expression in human tissues. Decalcification affects the accuracy of immunohistochemical staining results and cannot be reversed by image analysis. Measurement data obtained on a continuous scoring scale can be converted to categorical scores for comparison with categorical dataset that are generated by pathologists.