Deep-Learning-Aided Quantification of Steatohepatitis-Associated Pathological Findings in Liver Specimens

Area of Science:

• Hepatology
• Digital Pathology
• Artificial Intelligence in Medicine

Background:

• Semi-quantitative scoring of liver histopathology can lack objectivity and reproducibility.
• Developed a deep learning model to objectively quantify pathological findings in steatohepatitis.
• AI aims to improve the accuracy of assessing liver disease progression.

Purpose of the Study:

• To develop and validate a deep learning model for quantifying steatohepatitis-associated pathological findings.
• To assess the model's ability to measure percentages of steatosis, ballooning, and fibrosis in liver specimens.
• To correlate AI-quantified metrics with histological grades and ultrasound-based measurements.

Main Methods:

• Trained a convolutional neural network (AI) using 18 steatohepatitis and 8 chronic hepatitis liver specimens.
• AI model quantified %Steatosis, %Ballooning, and %Fibrosis in liver biopsy samples.
• Validated AI model performance in 233 patients by comparing with expert histological evaluation and ultrasound parameters.

Main Results:

• AI accurately measured %Steatosis, %Ballooning, and %Fibrosis.
• %Steatosis strongly correlated with histological steatosis grade (R=0.78) and ultrasound CAP (R=0.51).
• %Fibrosis in NASLD patients strongly correlated with histological fibrosis stage (R=0.72) and ultrasound elastography (R=0.7).

Conclusions:

• Deep learning-based pathology offers an objective method for quantifying steatohepatitis-associated findings in liver tissue.
• The AI model serves as a quantitative tool for pathological assessment, not a diagnostic algorithm.
• This technology has the potential to improve the consistency and accuracy of liver disease evaluation.