Predicting Liver Fibrosis Severity Using Machine Learning Models

Abstract

Introduction: The diagnosis of NAFLD typically involves the use of the FibroScan test, which can be costly. More affordable options, like liver enzyme and hematology tests, cannot diagnose fatty liver disease; they only serve as preliminary tools for its diagnosis.

Methods: In this study, a machine-learning model was developed to diagnose fatty liver disease using demographic information, liver enzymes, and hematology tests. Data was extracted from the records of 1078 patients who visited Haj Marafi Hospital between 2018 and 2023, encompassing 25 dependent variables. After preprocessing, the data was reduced to 531 records. A multi-objective particle swarm optimization algorithm was used to impute missing data. Following preprocessing, a support vector machine (SVM) algorithm was applied to the data, and the performance of the proposed algorithm was compared and evaluated against similar algorithms.

Results: During preprocessing, records with more than 20% missing data were removed, and the remaining data were imputed. The data was then divided into training and testing sets (70-30 split). The radial basis function (RBF) SVM was applied to the training data, resulting in sensitivity, specificity, and accuracy of 96.24%, 90.86%, and 93.55%, respectively. For the test data, these rates were 80%, 77.22%, and 78.62%.

Conclusion: This study demonstrated that machine learning algorithms can diagnose NAFLD more cost-effectively.