Multimodal AI for Renal Allograft Biopsy Assessment

Approximately 40% of renal transplants fail within 10 years, and allograft rejection is a major contributing cause. The current standard of care for renal transplant rejection assessment is the manual histologic examination of biopsies of the allograft. This process is known to suffer from large inter- and intra-observer variability across tasks (κ=0.22). Moreover, the expertise needed for comprehensive evaluations is often not available in low-resource settings, which can result in delays in diagnosis and treatment. Here, we propose MANTA (Multimodal AI for Nephro-Transplant Assessment), an objective and automated method for analysis of renal allograft biopsies to assess rejection, subtyping, and the status of renal interstitial fibrosis. MANTA is a weakly supervised multimodal deep learning paradigm which fuses morphological features from Hematoxylin and Eosin, Periodic acid Schiff, Masson trichrome, and Jones methenamine silver stains to make diagnostic assessments. MANTA utilizes a multinational cohort of renal transplant data of, 12,998 gigapixel whole slide images across 1,140 cases from the USA and Turkey to deliver promising results across renal allograft assessment tasks. This large-scale study lays the foundation for a system to deliver reliable, accessible assessment of renal transplant biopsies in order to improve kidney transplant outcomes.