Digitization of the imaging diagnostics in pathology has the potential to increase both efficiency and quality of care. In order to realize this potential, cross-disciplinary research efforts are needed that can combine clinical expertise with knowledge in artificial intelligence and human-computer interaction.
Ground Truth for AI Models
The CMIV research agenda in digital pathology tackles several image analysis challenges. An important groundwork is to create systematically annotated collections of imaging data to be used as ground truth for training of artificial intelligence models. The project has developed four large, systematically annotated data collections that are shared through the AIDA Data hub, in total about 800 GB of image data with 25,000 annotations.
Several clinical applications of AI in a human-in-the-loop setting are being explored: Increasing precision and efficiency in breast cancer histological grading, the same for cancer detection in lymph nodes, and smart assistance in skin cancer diagnostics. We also develop methods to accelerate the cumbersome process of creating training data for initial AI development and for tuning AI models in clinical use. Furthermore, methodology to validate digital clinical work is developed and deployed. A common ground for these research projects is the focus on finding digital solutions that will work in the clinical setting.
A noteworthy result from the project is also a solution for visualization of 3D histology, handling these very large data sets at high speeds and with ample interaction possibilities.
The CMIV pathology group consists both of medical and technical researchers from the university and pathologists and lab assistants from the clinical pathology department at the hospital, all working together on research and development efforts close to clinical practice. The group is very active in the international community to promote advances in the field. Examples include the organization of the Nordic Symposium on Digital Pathology and contributions to the Computational Pathology Symposium at the European Congress of Pathology.