Spatial AI quantification of inflammatory and architectural features differentiates pathologist-defined active ulcerative colitis from histological remission

PAIR-IBD spatially quantifies inflammatory and crypt-architecture features that distinguish pathologist-defined active ulcerative colitis from histological remission, including 2.1-fold higher LP neutrophils.

• Why it matters: Histology in ulcerative colitis is often quantified without anatomical location, leaving gaps for resolving discordant or inconclusive biopsies and supporting consistent endpoint scoring.
• What they did: A deep-learning ensemble segmented mucosal compartments and quantified lymphocytes, neutrophils, eosinophils, plasma cells plus crypt injury metrics (mucin depletion, solidity, roughness, branching, abscesses) across three IBD datasets.
• The result: Active UC showed increased mucin depletion, crypt branching, and crypt abscesses with reduced crypt solidity (p<0.0001 for all), metrics correlated with manual scores (rs=0.50–0.73), and up to 89% of inconclusive biopsies matched remission-like profiles.