Quantitative morphometric analysis via artificial intelligence-driven image analysis in autoimmune hepatitis: a multi-center study.

Clinical-pathological correlations in autoimmune hepatitis (AIH) remain challenging due to significant variability in clinical and pathological presentations and diverse subtypes. This study employed computer vision on digital slides to derive quantitative morphometric parameters, aiming to improve diagnostic accuracy and strengthen correlations with clinical markers. Clinical data and hematoxylin and eosin (H&E)-stained histological slides from diagnosis were collected from two Italian and one German tertiary referral centers. Images were processed with tissue and portal tract segmentation, and a nucleus classifier to differentiate hepatocytes and inflammatory cells. For each case, ten morphometric parameters, six classical histological metrics (mhAI and Ishak fibrosis scores), and over fifty clinical parameters were analyzed (Figure). The study includes a cohort of 128 patients with confirmed AIH; mean age at diagnosis was 54±17 years, with 66% females. After feature selection, the following morphometric parameters were prioritized for clinico-pathological correlations: 'inflammatory cell-to-hepatocyte ratio', 'mean distance between inflammatory cells', 'functional liver area' and 'cell density in portal tracts'. The 'inflammatory cell-to-hepatocyte ratio' in the lobule (p<0.01) and the 'mean distance between inflammatory cells' (p<0.01) showed strong correlations with mhAI score. When compared to traditional histological mHAI and Ishak Fibrosis scores, morphometric parameters demonstrated stronger correlations with AST and ALT levels in multiple linear regression. Patients with higher values of AST and ALT at diagnosis showed significantly higher values of 'inflammatory cell-to-hepatocyte ratio' (p<0.001) and lower values of 'mean distance between inflammatory cells' (p=0.02). The 'functional liver area' and 'cell density in portal tracts' were higher in patients with lower platelets (p<0.01 and p=0.01, respectively). Artificial intelligence-driven image analysis enables more reproducible quantification of histological parameters and enhances the correlation with clinical findings, complementing existing semi-quantitative scoring systems. [ABSTRACT FROM AUTHOR]