Liver fibrosis stage classification in stacked microvascular images based on deep learning

Area of Science:

• Hepatology
• Medical Imaging
• Artificial Intelligence

Background:

• Monitoring liver fibrosis is crucial for managing chronic liver disease (CLD).
• Stacked microvascular imaging (SMVI) is a novel technique for assessing fibrosis.
• Previous studies demonstrated SMVI's high sensitivity and examiner scoring effectiveness.

Purpose of the Study:

• To evaluate the effectiveness and objectivity of SMVI in diagnosing liver fibrosis stages using artificial intelligence (AI).
• To compare AI-based SMVI analysis with human judgment for liver fibrosis assessment.

Main Methods:

• A convolutional neural network model was developed to analyze SMVI from 517 CLD patients.
• Liver fibrosis was classified into five stages (F0-1Low, F0-1High, F2, F3, F4) based on liver stiffness measurements.
• Human judgment focused on microvessel morphological changes (narrowing, caliber irregularity, tortuosity) in SMVI.

Main Results:

• AI achieved 83.8% accuracy in a 2-class classification (F0-1 vs. F2-4), outperforming human assessment (81.6% for ≥F2).
• AI demonstrated higher sensitivity (84.2%) and specificity (83.5%) compared to human judgment.
• AI analysis showed a superior F1 score in the 2-class classification, indicating better overall performance.

Conclusions:

• AI-based SMVI assessment is more accurate than human judgment for detecting significant liver fibrosis (≥F2).
• AI analysis of SMVI reduces subjectivity and provides objective fibrosis development assessment.
• AI-powered SMVI represents a significant advancement in diagnosing liver fibrosis in CLD patients.