Automated multi-scale computational pathotyping (AMSCP) of inflamed synovial tissue

Abstract

AbstractRheumatoid arthritis (RA) is a complex immune-mediated inflammatory disorder in which patients suffer from inflammatory-erosive arthritis. Recent advances on histopathology heterogeneity of RA pannus tissue revealed three distinct phenotypes based on cellular composition (pauci-immune, diffuse and lymphoid), suggesting distinct etiologies that warrant specific targeted therapy. Thus, cost-effective alternatives to clinical pathology phenotyping are needed for research and disparate healthcare. To this end, we developed an automated multi-scale computational pathotyping (AMSCP) pipeline with two distinct components that can be leveraged together or independently: 1) segmentation of different tissue types to characterize tissue-level changes, and 2) cell type classification within each tissue compartment that assesses change across disease states. Initial training and validation were completed on 264 knee histology sections from mice with TNF-transgenic (n=233) and injected zymosan induced (n=32) inflammatory arthritis. Peak tissue segmentation performance with a frequency weighted mean intersection over union was 0.94 ± 0.01 and peak cell classification F1 was 0.83 ± 0.12.We then leveraged these models and adapted them to analyze RA pannus tissue clinically phenotyped as pauci-immune (n=5), diffuse (n=28) and lymphoid (n=27), achieving peak cell classification performance with F1 score of 0.81 ± 0.06. Regression analysis demonstrated a highly significant correlation between AMSCP of lymphocyte counts and average Krenn Inflammation Score (rho = 0.88; p<0.0001). While a simple threshold of 1.1% of plasma cells demonstrated the phenotyping potential of our automated approach vs. a clinical pathologist with a sensitivity and specificity of 0.81 and 0.73. Taken together, we find AMSCP to be a valuable cost-effective method for research. Follow-up studies to assess its clinical utility are warranted.