Related Experiment Video
Updated: Jul 1, 2026
08:41
Novel In Vivo Micro-Computed Tomography Imaging Techniques for Assessing the Progression of Non-Alcoholic Fatty Liver Disease
Published on: March 24, 2023
1.7K
MMD-Net: a weakly supervised solution for quantification of nonalcoholic fatty liver biopsies
Ming Yu1, Tao Jiang1, Hongsheng Zhou2
1Department of Biomedical Engineering, School of Information Science and Technology, Fudan University, Shanghai, China.
Quantitative Imaging in Medicine and Surgery
|January 12, 2026
Summary
A new weakly supervised deep learning model reduces pathologist workload for diagnosing nonalcoholic fatty liver disease (NAFLD). This AI approach enables efficient and standardized NAFLD assessment using multi-instance learning.
Area of Science:
- * Hepatopathology
- * Artificial Intelligence in Medicine
- * Computational Pathology
Background:
- * Nonalcoholic fatty liver disease (NAFLD) impacts 25% of the global population, often leading to cirrhosis.
- * Liver biopsy, the gold standard for NAFLD diagnosis, uses the labor-intensive Kleiner scoring system.
- * Current deep learning (DL) methods for NAFLD assessment require extensive pathologist annotations, limiting scalability.
Purpose of the Study:
- * To develop a weakly supervised framework using multi-instance learning (MIL) for NAFLD assessment.
- * To reduce the annotation burden on pathologists in NAFLD diagnosis.
- * To create a clinically applicable diagnostic approach for NAFLD.
Main Methods:
- * Utilized a publicly available hepatocellular pathology dataset.
- * Developed MMD-Net, a weakly-supervised framework integrating MIL with multi-task learning (MTL).
- * Evaluated model performance using accuracy, precision, recall, F1 score, and Cohen's κ.
Main Results:
- * MMD-Net demonstrated strong agreement with ground truth in NAFLD assessment.
- * Achieved high quadratic weighted Cohen's κ coefficients: 0.932 for ballooning, 0.836 for inflammation, and 0.766 for steatosis.
- * The mean Cohen's κ across all metrics was 0.845.
Conclusions:
- * Established a novel paradigm for standardized NAFLD histopathological assessment.
- * Eliminated the need for pixel-level annotations, reducing pathologist workload.
- * Paved the way for AI-driven histopathological analysis in clinical NAFLD assessment.
