Semi-Supervised Fatty Liver Classification Using Attention-Based Graph Neural Network Models.

Area of Science:

• Medical Informatics
• Machine Learning
• Graph Neural Networks

Background:

• Fatty liver disease is prevalent and linked to serious health issues like cirrhosis.
• Accurate diagnosis is vital, but deep learning faces challenges with limited labeled clinical data.
• This study explores graph-based deep learning with attention for fatty liver disease prediction.

Purpose of the Study:

• To evaluate attention-based Graph Neural Networks (GNNs) for fatty liver disease prediction.
• To assess the efficacy of GNNs in semi-supervised learning with scarce labeled data.
• To identify key predictors and patient subgroups for individualized risk stratification.

Main Methods:

• Utilized a clinical dataset of 7,953 individuals with health check-up variables.
• Applied GNNs with attention mechanisms in a semi-supervised learning framework.
• Employed GNNExplainer for feature importance and conducted subgroup analysis.

Main Results:

• Attention-based GNNs significantly outperformed logistic regression in fatty liver disease prediction (P < 0.05).
• High AUCs (e.g., 0.7893) were achieved with only 100 labeled samples.
• Key predictors included HbA1c, body fat, and glucose; two distinct patient clusters were identified.

Conclusions:

• Attention-based GNNs offer strong predictive performance for fatty liver disease with minimal labeled data.
• This method leverages relational data structures for efficient, individualized risk assessment.
• Graph-based learning shows promise for label-constrained clinical settings.