An automatic, rapid and accurate method for the annotation of tumor components on whole slide images

Annotated pathology datasets are the cornerstone for developing computational pathology. However, the intrinsic complexity of pathology data often results in a scarcity of large, manually annotated datasets. To address this challenge, a rapid, accurate, and automatic method for annotating tumor cells is essential for advancing the field of computational pathology. Here, we introduce a novel annotation technology. In our approach, Hematoxylin-eosin (H&E) slides were first digitized and preserved. These H&E slides were then faded and re-stained using multiple biological technologies (MBT) to identify specific biomarkers. The re-stained MBT slides were subsequently scanned again to create new digital images. The original and re-stained digital images underwent a registration and segmentation process to ensure that all minute structures in both images aligned perfectly. The staining results from the MBT slides, referred to as2 label maps, were extracted and transferred back onto the H&E slides, and the data were merged. This method allows for precise and efficient annotation of all tumor cells, including those expressing specific biomarkers, as well as the components of the tumor microenvironment. By rapidly generating high-quality, high-volume datasets, this innovative method significantly enhances the ability of AI approaches to analyze and interpret pathology data. Consequently, it supports the development of highly accurate diagnostic, prognostic, and predictive decision-making systems in the field of computational pathology.