Developing a prediction model of fatty liver on health examination data by Genetic Programming

• Main Authors: LIN, HUNG-CHUN, 林泓均
• Other Authors: HUANG, LIANG-TSUNG
• Format: Others
• Language: zh-TW
• Published: 2019
• Online Access: http://ndltd.ncl.edu.tw/handle/k82a5t

碩士 === 慈濟大學 === 醫學資訊學系碩士班 === 107 === The diagnosis of steatosis is made when fat in the liver exceeds 5% by weight or fat vacuoles exceed 10%. With 6-35% prevalence rate, the nonalcoholic fatty liver is a quite common disease in the world. Liver diseases, including liver cancer and cirrhosis, have been top ten leading causes of death in Taiwan. Therefore, the prevention of fatty liver is very important. The main diagnosis method of fatty liver is abdominal ultrasound and liver biopsy. Since machine learning has been applied to medicine studies successfully in recent years. This study aims to develop prediction models of discriminating fatty liver from health examination data of Hualien Tzu Chi Hospital by using machine learning methods. We used four machine learning methods, including evolutionary Genetic Programming (GP), SVM, J48, and PART. The experimental architecture is to build models by four methods, select import features, add Genetic Programming as a new variable to the training data, compare the various results and find the best model. Finally, the results show that the performance of SVM with the variable of Genetic Programming is better than other methods . The accuracy is 76.76%, the AUROC is 73.52%. After feature selection, The accuracy is 70.06%, AUROC is 74.44%.