Automated Evaluation of the Proliferation Index using Deep Learning Neural Networks

Ki67 is a commonly used measure for cancer proliferation. This work proposes the deep learning techniques to count breast cancer cells in immunohistochemistry (IHC) slides automatically. Conventionally, pathologists count manually cells of various colors to identify positive and negative cells—a task that is time-consuming and prone to human error. To overcome this, this paper explores the potential of utilizing automated methods to count Ki67-positive cells in IHC slides. It compares the performance of two pre-trained deep convolutional neural networks. While the RMSE of U-Net model was 38.25 whereas StarDist showed an improved performance with an RMSE of 24.94. The findings show that deep learning models are very effective for this purpose and imply that manual assessment of Ki67-positive cells is unnecessary or not recommended in determining luminal breast cancer subtypes.