AI Models the Body as a Network to Predict Fatty Liver Disease

Hepatic steatosis, or fatty liver disease, affects a third of the global population, yet accessible tools for early, non-invasive screening are lacking. Addressing this challenge, scientists have developed a novel artificial intelligence model that learns to predict liver fat from simple body measurements.

The study, involving 705 participants, used a graph neural network (GNN) – a type of AI that conceptualises the body as an interconnected graph. This approach allows the model to capture the complex physiological relationships between different parts of the body, using 27 features like physical exam results and body composition data.

This new GNN model significantly outperformed traditional machine learning methods in predicting a key measure of liver fat called the CAP score. An attention-guided analysis revealed that waist circumference, trunk fat mass, and neck circumference were the most influential predictors. This graph-based modelling offers a promising new direction for reliable, non-invasive screening for fatty liver disease.