Digital pathology is rapidly becoming reality in routine diagnostics, enabling also the development of artificial intelligence (AI) based computational pathology. Expert-driven diagnostics process will face a paradigm shift as the AI-based decision support for diagnostics matches or outperforms human experts. Single studies with limited sample pools, however, lead to positively biased view of the true applicability of AI in routine clinical diagnostics. AI-based decision support tools will need to become more scalable and to generalize better from limited data to data from other laboratories and from other measurement scanner devices outside the original domain. Here, we study the generalization performance of AI-based methods in diagnostics applications and develop computational pathology tools for tasks beyond capabilities of human vision, such as for prediction the gene expression and mutational status directly from histopathological images, and for virtual staining based analytics.

2021

2025