An Advanced Automated Image Analysis Model for Scoring of ER, PR, HER-2 and Ki-67 in Breast Carcinoma

Abstract

Immunohistochemistry (IHC) plays an important role in evaluating the status of ER, PR, Ki-67 and human epidermal growth factor receptor 2 (HER-2) during diagnosis of breast cancer. Although some existing automated approaches can solve the high time-consumption and inter-/intra-observer variability drawbacks to a certain extent, most of them are can’t analyze both nuclear staining and cell membrane staining using the same method. This is attributed to the difference in localization of the positive signal of immunohistochemical staining in different biological markers. The present study proposes a novel automated image analysis model for scoring and grading of ER, PR, Ki-67 and HER-2 immunohistochemical images based on whole tissue sections in breast cancer. The scoring results of the trained model and manual interpretation of ER, PR, Ki-67 and HER-2 were then finally analyzed and compared. Experimental results show that the F1-measure was 0.8450, 0.8533 and 0.7962 for nuclear recognition of Ki-67, ER/PR and HER-2 respectively. For stain grading of Ki-67, ER/PR and HER-2, the F1-measure was 0.9776, 0.8306 and 0.9573 respectively. The scoring consistency of ER/PR, Ki-67 and HER-2 between our model and expert interpretation was 0.9279, 0.9712 and 0.8046 respectively. Our results demonstrate that artificial intelligence technology is a feasible and accurate method for accurate quantitative immunohistochemical analysis that can solve the drawbacks of low repeatability and time consumption brought by manual counting. The main contribution of our proposed model is that it can recognize both nuclear staining and cell membrane staining and grade the staining intensity as a sequential learning task.

Keywords

• Automatic analysis
• breast carcinoma
• deep learning
• Er
• Her-2
• immunohistochemistry