Comparative analysis of deep learning models for various nonalcoholic fatty liver disease datasets

Abstract

Fatty liver disease is caused by increased liver buildup or weight above 5-10%. This disorder is widespread in people with diabetes, overweight persons, and metabolic syndrome patients. Clinical decision support systems can improve liver failure diagnosis and prediction to reduce this situation. Many liver failure models have drawbacks, and liver failure prediction is still a problem. This work uses four large open-access critical care patient datasets to create and verify liver failure risk prediction models. This study aims to construct a clinically applicable diagnostic and predictive model that evaluates the probability or risk of liver failure in intensive care unit (ICU) patients using extreme gradient boosting (XGBoost), artificial neural networks (ANN), multi-layer perceptron (MLP), Modular Neural Network (MNN), and generalized feed forward (GFF). We evaluated performance metrics using these models: accuracy, sensitivity, specificity, and predictive accuracy.