AI-centered visual analytics of histology
Histology images, microscopy scans of human tissue, is a key diagnostic domain requiring the analysis of large-scale datasets (ca. 100,000×100,000×100 pixels). The multiple scales and high density of the content mean severe visual analytics (VA) challenges. This project aims to merge AI techniques with visual exploration. The starting point is a compact data representation, a space retaining expressiveness for diagnostic features while allowing effective and dynamic training of AI models to be employed in interactive visualizations. Next, data augmentation and synthesis methods will be employed to tackle the challenge of generalizing the AI models. Interaction between core and applied visualization research are in focus. The Vis-MCP research themes are well represented, in particular interactive exploration, multi-scale analysis, and computational steering. Impact: A successful project will be a major step forward for VA in histology research and other domains with similar data characteristics. There is potential for impact on precision and productivity in healthcare.
External funding and synergies: The project builds on the previous SeRC efforts on visualization of 3D histology. There is a complementary connection to the Vinnova-supported center AIDA, providing a natural path to mature innovations and impact in healthcare. There are synergies also with WASP, through PhD students in the AI domain.