Digital Hepatic Iron Content: An Artificial Intelligence Model for Spatially Resolved Histologic Iron Quantitative...

Area of Science:

• Digital pathology
• Artificial intelligence in medicine
• Hepatology

Background:

• Accurate hepatic iron content (HIC) evaluation typically requires destructive laboratory methods.
• Routine histologic stains offer potential for non-destructive iron assessment.

Purpose of the Study:

• To develop and validate an artificial intelligence (AI) model for recognizing and quantifying liver iron using routine histologic stains.
• To assess the AI model's correlation with established quantitative and semi-quantitative iron assessment methods.

Main Methods:

• A supervised deep learning AI model was trained on digitized Pearl Prussian blue stained whole slide images of liver samples.
• The AI model was validated using 98 liver samples with available inductively coupled plasma mass spectrometry (ICP-MS) data.
• Correlation analyses were performed between AI-derived % iron area and HIC, hepatic iron index (HII), and histologic scoring systems.

Main Results:

• The AI model showed high correlation with ICP-MS for HIC (Rs=0.93 for core biopsies, Rs=0.86 for all samples).
• The AI model accurately predicted high hepatic iron index (HII > 1, AUC=0.93; HII > 1.9, AUC=0.94).
• Spatially resolved iron analysis by AI identified hereditary hemochromatosis mutations (AUC=0.65) and correlated strongly with histologic scores (Rs=0.87).

Conclusions:

• The AI model provides accurate, non-destructive, and spatially resolved quantification of liver iron content.
• This AI approach enhances the utility of routine histologic stains for diagnosing and managing iron overload disorders.
• The AI model offers advantages over traditional tissue-destructive methods for HIC assessment.