Органоспецифічний патоморфоз печінки при генералізованій вродженій цитомегаловірусній інфекції: розробка моделі на основі штучного інтелекту

Cytomegalovirus infection (CMV) remains the leading cause of congenital infectious pathology in children, accounting for up to 20-30 % of antemortem diagnostic discrepancies due to its ability to mimic other acute conditions, including multiple congenital malformations and bacterial sepsis. The liver, as a primary target organ in generalized congenital CMV infection, demonstrates both non-specific and pathognomonic histomorphological changes, making it an ideal substrate for machine learning-based pattern recognition. The aim of the study was to develop an artificial intelligence (AI) classification model for improving the criteria of pathohistological verification of generalized congenital CMV infection through machine learning analysis of organ-specific hepatic changes, enabling differential diagnosis between CMV infection, age-matched normal liver tissue, and competing destructive pathologies. The study was based on two databases: Database 1 comprised 25 autopsy protocols from the Vinnytsia Regional Pathological Anatomical Bureau of Vinnytsia Regional Council (VRPAB) involving children aged 12 days to 10 years (2012-2025); Database 2 included data from 160 cases derived from peer-reviewed publications sourced from PubMed, Google Scholar, and Dove Medical Press. Three histomorphological classes were digitized for model training: Class 0 – age-matched normal liver (absence of giant cell metamorphosis, preserved lobular architecture, no inflammatory infiltrates); Class 1 – competing pathologies (bacterial sepsis, biliary cirrhosis), characterized by non-specific dystrophic changes and microcirculatory thrombosis without viral inclusions; Class 2 – generalized congenital CMV infection, defined by the presence of “owl-eye” cytomegalic inclusion cells (intranuclear inclusions ~10 µm in diameter with characteristic nuclear halo), specific canalicular cholestasis, and vacuolar dystrophy of hepatocytes. The AI model was implemented in the Jupyter Notebook environment using a Linear Discriminant Analysis (LDA) classifier integrated with the Gemini neural network for dataset processing. The model achieved a classification accuracy of 0.9037, approaching near-perfect diagnostic performance. The developed “one-button” principle allows any clinician to upload a histological image of a liver specimen and obtain an automated differential diagnosis verdict. The proposed AI-based model significantly reduces subjectivity in pathohistological examination, provides rapid organ-specific verification of congenital CMV infection, and demonstrates strong potential for broad clinical implementation as a digital diagnostic assistant in pediatric pathology.