LymphoidStructureMiner: AI-based exploration of the immunological contexture of lymphoid structures in translational research

Lymphoid structures (LS) are highly organized multicellular units of immunity that are able to recognize, respond to and eliminate a great variety of pathogens as well as cancer cells. Functional activity of LS is linked to their germinal centers (GC). Intriguingly, ectopic/tertiary LS were discovered in more than ten types of solid cancer and are associated with favorable patient outcomes. Among the pioneering discoveries is our research on LS as strong prognostic factor in metastatic colorectal cancer. Microscopy-based tissue image cytometry is one of the principal technologies that can be used to analyze LS. With the advent of advanced AI-based approaches with various applications in medical and non-medical computer vision tasks, there is a growing trend to utilize them in computational pathology. However, due to the complex architectural features of LS, multiple not-yet-solved challenges exist to apply these models for LS analysis. The Project will focus on developing novel deep learning-based methods for accurate detection and segmentation of LS and GC in whole-slide images and further cell-based analysis of their immunological contexture based on nuclei instance segmentation and classification of various immune cells with focus on B cells. Through the vision of the Project Team, the LymphoidStructureMiner will represent an invaluable tool for a plethora of biomedical research areas including but not limited to immuno-oncology. Our scientific questions, innovative AI-powered analytical approaches, which will be developed in the current study, and the knowledge obtained in the scope of the Project will be likely of great interest to a wide audience of scientists and clinical specialists in the light of biomarker assessment, patient stratification and directions in treatment options.