Machine Learning-Enabled Quantification of Hepatocellular Necrosis in the Liver After Lethal Marburg and Ebola Virus...

Area of Science:

• Pathology
• Virology
• Artificial Intelligence

Background:

• The liver is a primary target in severe filovirus infections.
• Accurate quantification of liver pathology in filovirus disease is currently limited.

Purpose of the Study:

• To develop and validate deep learning models for quantitative analysis of liver pathology in Marburg virus (MARV) and Ebola virus (EBOV) infections.
• To compare liver injury patterns between MARV and EBOV in a non-human primate model.

Main Methods:

• Deep learning models were trained for pixel-level segmentation of digitized liver pathology slides from rhesus monkeys infected with MARV, EBOV Makona, or EBOV Kikwit.
• The model quantified hepatic necrosis, arteries, veins, and bile ducts, with performance assessed against pathologist variability.
• Liver enzymes (AST, ALT, GGT, ALP) were measured and correlated with histopathologic findings.

Main Results:

• The deep learning model achieved necrosis segmentation comparable to expert pathologists.
• DL-quantified liver necrosis was significantly correlated with the infecting virus, with MARV-infected monkeys exhibiting the highest necrosis (11.0%).
• Alanine aminotransferase (ALT) levels correlated with necrosis severity, and necrosis localization differed between MARV (portal tract-proximate) and EBOV (central vein-proximate) infections.

Conclusions:

• Deep learning provides a robust method for quantitative histopathologic analysis of liver injury in filovirus infections.
• MARV infection induces more severe liver necrosis than EBOV infection in rhesus monkeys.
• This approach enables large-scale retrospective analyses of filovirus-induced liver pathology.