Laboratory variables-based artificial neural network models for predicting fatty liver disease: A retrospective study

Area of Science:

• Medical Informatics
• Machine Learning in Medicine
• Hepatology

Background:

• The predictive capability of artificial neural network (ANN) models for fatty liver disease (FLD) using laboratory data is not well-established.
• Accurate prediction of FLD is crucial for timely intervention and management.

Purpose of the Study:

• To develop and validate ANN models for precise FLD prediction.
• To evaluate the impact of incorporating demographic variables (age and gender) on model performance.

Main Methods:

• Utilized a dataset of 7,990 eligible participants for FLD screening.
• Developed and validated ANN models using training, validation, and testing sets.
• Assessed model performance using the area under the receiver operating characteristic curve (AUROC).

Main Results:

• ANN models with varying numbers of laboratory variables demonstrated robust FLD prediction (AUROCs > 0.89).
• A 2-variable model achieved 80.3% accuracy and 89.7% positive predictive value.
• Incorporating age and gender significantly improved AUROCs across all tested ANN model configurations.

Conclusions:

• ANN models are effective tools for predicting FLD.
• The inclusion of age and gender enhances the predictive accuracy of ANN models for FLD.