Introduction
The Banff classification is the gold standard for histopathologic assessment of transplant kidney biopsies. It consists of 17 lesion scores, 10 of which focus on the presence and extent of inflammatory cells in different kidney compartments. As this scoring suffers from subjectivity and is very time consuming, development of automated biopsy assessment holds great potential to reduce pathologist's workload and increase scoring consistency.
Material and methods
The MONKEY challenge focuses on detection of inflammatory cells, specifically lymphocytes and monocytes, in PAS-stained kidney transplant biopsies. It is run on the Grand Challenge Platform (monkey.grand-challenge.org) with leaderboards for individual tasks. The dataset consists of annotated regions from a multi-centric cohort of 120 WSIs. To ensure reliable annotations, the slides are re-stained with antibodies against CD3, CD20, and PU.1 to identify lymphocytes and monocytes, respectively. Algorithm performance will be evaluated on a test set from a separate institution.
Results and discussion
The challenge is planned to open before summer 2024. It will result in a best performing algorithm to detect inflammatory cells in PAS-stained biopsies. This algorithm will be accessible for the research community and will be further incorporated in AI for automated Banff lesion scoring. The MONKEY dataset will remain publicly available for research purposes.
Conclusion
Several of our previous challenges (CAMELYON, PANDA and TIGER), where such wisdom-of-the-crowd approach was applied for urgent clinical applications, have produced highly successful algorithms, sometimes even surpassing experienced pathologists. The MONKEY challenge is highly ambitious as it aims to differentiate between lymphocytes and monocytes, bringing us a step closer to automated Banff lesion scoring.