A machine learning-based framework for the diagnosis and grading of non-alcoholic fatty liver disease using anthropometric features

Background and aims: Non-alcoholic fatty liver disease (NAFLD) is a kind of accumulation offat in liver cells, which if not controlled, goes through a cirrhosis process towards fibrosis of livertissue and cell destruction. In recent years, the prevalence of NAFLD has increased. Therefore,the creation of a simple and low-cost method to classify NAFLD patients in the population isneeded. The aim of this study is to diagnose fatty liver and distinguish grades of fatty liver usingdifferent machine learning models, which can be used for accurate screening of a large numberof people.Method: For this purpose, the anthropometric features of ۶۵۱ people over the age of ۱۸ without ahistory of continuous alcohol consumption and underlying liver disease in two southern and easternprovinces of Iran, ۲۷۸ women and ۳۷۳ men with a mean age of ۳۷.۴۶ years, were recorded.۳۶۱ subjects without fatty liver, ۲۹۰ subjects had different grades of fatty liver. Different machinelearning algorithms such as Support Vector Machine, k-Nearest Neighbor, Random Forest, LogisticRegression, Naive Bayes, and Multilayer Perceptron Neural Network were used to detect fattyliver and determine the grade of fatty liver. ۵۲۰ subjects were used to train the models and ۱۳۱subjects were used to test the models. To evaluate and compare the performance of the createdmodels, accuracy, precision, recall and f۱-score criteria were calculated.Results: Six models based on machine learning were developed and showed good performancein predicting NAFLD. Among these models, the random forest method and the number of ۲۰ decisiontrees showed the best performance with accuracy ۹۹.۲%, precision ۹۸%, recall ۱۰۰%, andf۱score ۹۹%. To detect fatty liver grade, the support vector machine algorithm with linear kernelfunction and gamma parameter equal to ۰.۰۰۱ and c parameter equal to ۵ with ۹۵% accuracy hadthe best performance compared to other methods.Conclusion: Machine learning classifiers can help in the medical field to achieve early detectionand classification of NAFLD. The proposed models can achieve high accuracy without relyingon laboratory measurement parameters, especially in areas with poor financial situations and highepidemiology.