AutoFibroNet: A deep learning and multi-photon microscopy-derived automated network for liver fibrosis quantification...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Hepatology

Background:

• Liver fibrosis is a critical predictor of complications and mortality in metabolic dysfunction-associated fatty liver disease (MAFLD).
• Label-free imaging techniques like SHG/TPEF offer promising avenues for assessing liver fibrosis.
• Accurate staging of liver fibrosis is crucial for managing MAFLD patients.

Purpose of the Study:

• To develop and validate AutoFibroNet, an automated quantitative histological classification tool.
• To combine multi-photon microscopy (MPM) with deep learning for accurate liver fibrosis staging in MAFLD.
• To create a novel AI-driven system for objective and precise fibrosis assessment.

Main Methods:

• Development of AutoFibroNet using deep learning models (VGG16, ResNet34, MobileNet V3) on a training cohort of 203 Chinese adults with biopsy-confirmed MAFLD.
• Integration of deep learning features with clinical and manual features using multi-layer perceptrons for a joint model.
• Validation of the AutoFibroNet model in two independent cohorts to assess its generalizability.

Main Results:

• AutoFibroNet demonstrated excellent discrimination in the training set, with AUROCs of 1.00 (F0), 0.99 (F1), 0.98 (F2), and 0.98 (F3-4).
• Validation in independent cohorts showed strong discriminatory ability, with AUROCs for fibrosis stages ranging from 0.80 to 1.00.
• The tool achieved high accuracy in classifying different stages of liver fibrosis across diverse patient groups.

Conclusions:

• AutoFibroNet is a validated automated quantitative tool for accurate histological staging of liver fibrosis in MAFLD patients.
• The AI-based approach offers a reliable method for assessing liver fibrosis, potentially improving patient management.
• This technology shows significant promise for non-invasive or minimally invasive fibrosis assessment in MAFLD.