Automatic Multi-Scale Visual Annotations of Histopathology Images Based on Density and Topology Analysis

Digital pathology has emerged as a promising tool for the diagnosis and management of cancer. However, the accurate and efficient analysis of histopathology images is still a challenging task. Recently, artificial intelligence (AI) has matured enough to provide models for large-scale detection and classification of cellular structures. In this thesis, we propose an automatic multi-scale visual annotation approach for histopathology images. We use AI to automatically detect and classify whole slide images (WSIs) of histopathology tissue samples. We then create annotations using AI, density, and topology analysis. The proposed approach provides these visual annotations for histopathologists to ease their workflow. We provide 3 types of visual annotations: microscale, mesoscale, and multiscale annotations. Microscale annotations involve drawing different colored bounding boxes around nuclei of a specific cancer type inWSIs and are generated automatically by AI. Mesoscale annotations use kernel density estimation to provide colored contours overlayed on top of the WSI. Macroscale annotations are the most general and use topology to provide annotations that summarize the WSI. We evaluate the performance of the proposed approach through qualitative assessment by interviewing a histopathologist working in the field. We find that our proposed approach has the potential to aid histopathologists in the accurate and efficient analysis of histopathology images, and could contribute to the development of computer-aided diagnosis systems for cancer.