Document resource
Background Eosinophils are a key component of the immune response within the tumor microenvironment (TME). However, in many cancer types, including breast cancer, the role of eosinophils in the TME and their resulting impact on cancer outcomes remains poorly understood - hindered, in part, by the reliance on their manual identification by trained pathologists.By combining whole-slide imaging with quantitative tissue analysis powered by machine learning, digital pathology provides the ability to transform the time-consuming manual scoring process into a scalable, high-throughput workflow capable of accelerating advancements and enabling deeper insights into TME-focused oncology research. Here, we present AI models for the detection of eosinophils and the classification of tumor/stroma regions in whole-slide images (WSIs) of H&E stained breast tissue. We further highlight spatial analyses capable of determining high-density (‘hot’) and low-density (‘cold’) eosinophil regions and the proximity of eosinophils to tumor that may provide additional insight into the potential functional roles of eosinophils within the TME.Methods WSIs of H&E stained breast tissue were acquired using an Aperio slide scanner at 40x magnification and imported into the Indica Halo software, which was GxP-validated by NeoGenomics. From these WSIs, field of views (FOVs) encompassing the TME were selected and a pathologist provided estimates of the tumor area and the eosinophils present. Image crops were then generated from these FOVs and sorted into training and validation sets. AI models for the detection of eosinophils and the classification of tumor/stroma were developed on the training set using Halo AI. The performance of the AI models was evaluated on the validation set by concordance analysis with the pathologists’ assessments and Spatial analysis of eosinophil and tumor proximity of eosinophils were performed using Halo’s spatial analysis tools.Results In total, one hundred breast cancer cases were evaluated in this study. The algorithmic detection of eosinophils and the classification of tumor/stroma regions strongly correlated with the pathologist’s evaluations.Conclusions The developed AI models enables accurate, high-throughput analysis of eosniphils in WSIs of H&E-stained breast tissue. The pipeline efficiently identifies eosinophil-rich samples, which can then be further interrogated by spatial analyses. This workflow offers a robust, scalable solution for discovery-driven research and clinical drug trials, while minimizes both inter- and intra-observer variation and enhances reproducibility and consistency.