Optimization of Ki67 digital image analysis in breast cancer by automated tumor area identification

Accurate assessment of Ki67, a marker of cellular proliferation, is critical for breast cancer diagnostics and treatment decision-making. This study evaluates an automated Ki67-area identification approach, combined with digital image analysis (DIA) for exact Ki67 quantification. A total of 61 tissue samples from breast cancer patients from two clinical trials (GeparSixto and GeparSepto) were analyzed. The supervised DIA workflow employed automated Ki67-stained tumor area identification, followed by automated Ki67 scoring and was quality-controlled by trained pathologists. Comparisons with manual assessments were conducted to evaluate concordance and scoring precision. The DIA approach identified 232 tumor areas across 61 whole slide images (WSIs). The supervised system demonstrated a high correlation with manual scores (r = 0.78), and improved precision (standard deviation between evaluated tumor areas) was noted in therapy-naïve samples (p < 0.001). While manual assessments predominantly employed stepwise increments (5%), DIA provided finer, more continuous scoring, optimally reflecting the continuous nature of tumor proliferation. Overall, DIA scoring was more robust, with reduced inter-ROI variability and improved reproducibility compared with manual assessment. In conclusion, the study highlights the potential of supervised DIA to enhance Ki67 scoring accuracy and standardization. The methodology's effectiveness in addressing standardized Ki67 scoring suggests its utility in clinical diagnostic workflows. This approach contributes towards improved integration of computational pathology into routine practice for improved breast cancer management.