Kates, H.; Lee, C.; Paul, A. S.; Ansari, I.; Tatke, A.; Lee, T.; Nguyen, M.-T.; Eadon, M. T.; Sarder, P.; Chen Wongworawat, Y.

Background: Tubulitis is a defining histologic feature of T cell-mediated rejection (TCMR), while glomerulitis is often characteristic of antibody mediated rejection (AMR). Histologic quantification of tubulitis and glomerulitis using Banff criteria is subject to interobserver variability. Bulk transcriptomic assays (e.g., MMDx) have introduced molecular correlations of tubulitis with TCMR and glomerulitis with AMR, but lack spatial resolution. Methods: We applied a web-based platform, FUSION (Functional Unit State Identification in Whole Slide Images), to a cohort of 8 cases (n=2 per condition) with kidney allograft biopsy samples acute TCMR, active AMR, chronic active AMR, and no rejection (control). The machine-learning (ML) platform enabled integrated visualization and analysis of spatial transcriptomics (10x Genomics Visium v2) together with high resolution whole slide histology. Results: Transcriptomics-derived immune cell proportions within AI-segmented tubular and glomerular regions were used to generate spatial Banff t- and g-scores. Derived t-scores showed full concordance with pathologist scores in both acute TCMR cases; g-scores showed concordance in 2 of 4 AMR cases, with discordant cases characterized by low absolute immune signal near the classification boundary. Conclusions: We demonstrate the feasibility of using AI-based FTU segmentation integrated with spatial transcriptomics-derived immune cell proportions to generate spatially informed t- and g-scores aligned with Banff criteria, with full concordance in severe rejection and partial concordance in mild rejection. This approach lays the foundation for validated, spatial transcriptomics-augmented t-scores and g-scores that enhance diagnostic precision, reduces inter observer variability among renal pathologists, and support potential clinical adoption.