Search

Background: Scoring proliferation through Ki67-immunohistochemistry is an important component in predicting therapy response to chemotherapy in breast cancer patients. Therefore, an accurate and standardized Ki67-scoring is pivotal both in routine diagnostics and larger multi-center studies aiming at improving present or establishing new cut-off values for existing or novel therapy regimens. However, recent studies have cast some doubt on the reliability of "visual" Ki67 scoring by pathologists, especially within the lower - yet clinically important - proliferation range. Here, we present and apply a novel automated image analysis approach for Ki67-quantification in breast cancer tissue. Methods: We perform automated Ki67-scoring in 1219 breast cancer patients from the GeparTrio study cohort using a novel image analysis approach that avoids detection biases due to morphological variability by using a generic minimum-model approach. The method is capable of tumor-stroma-separation and may be used to process large data sets fully unsupervised in batch mode while allowing for efficient visual checks of the results. We compare these results with a different in-house-developed subtiling-based automated quantification method and moreover, gauge our approach with manual scoring performed by pathologists. Results: The results show deviations of 10% (automated method 1 vs. manual), 9% (automated method 2 vs. manual) and 3% (automated method 1 vs. automated method 2) on average. The Ki67 scores show Pearson correlations between automated and manual scoring of r>0.8 (p < 0.001) for both automated methods and r>0.95 (p < 0.001) between the two tested automated methods. Conclusion: Because of the methodological differences of the presented techniques our results suggest a high robustness of the automated methods that at the same time show a good agreement with manual Ki67 scoring. Our approach therefore offers an automated and standardized means of Ki67 quantification applicable in routine diagnostics as well as larger clinical study settings, such as in the GeparTrio cohort shown here. [ABSTRACT FROM AUTHOR]