Research Laboratory for Digital Pathology and Human Factors in AI
Digital pathology is a thriving research domain that uses large volumes of imaging data accompanied by sensitive clinical data collected to heterogeneous data models. These models fundamentally improve how the histopathological and clinical diagnosis and oncological treatment of the patients is performed. We support the whole digital pathology workflow, starting with the digitization, data and metadata management and algorithm development. For the later, we operate a mass digitization infrastructure optimized for scanning of archive slides.
To ensure FAIR-compliant accessibility of the medical data and metadata we developed BIBBOX, a modular component-based toolkit, which integrates open source software tools with ID- and user-management to record provenance graphs for data quality and reproducibility. BIBBOX is today used in several international projects for data integration, data quality assessment and the publishing of reference and training data sets, used e.g. in medical research and machine learning.
In computational pathology we make machine decisions transparent, traceable and thus interpretable for a medical expert. We enable pathologists to understand the context and to question a machine decision for the "why". For this, we develop new methods to measure the quality of explanations and their "causality". Our research results enable the development of novel "Human-AI Interfaces", which support an efficient and ethically responsible use of artificial intelligence in pathology.