運用臨床檢驗值建立判斷脂肪肝之類神經網路模型

Designing an Artificial Neural Network Model for the Prediction of Fatty Liver Disease

本研究主要是利用類神經網路建立模型，探討影響模型預測分析脂肪肝與肝生化異常的相關性，使用SPSS 20；t-test統計分析有、無脂肪肝參數差異，同時採用腹部超音波（Abdominal Ultrasonographic survey）作為診斷有無脂肪肝之依據，進行預測評估罹患脂肪肝。其中變項年齡（Age）、腹圍（Abdominal girdle）、三酸甘油脂（Triglyceride）、空腹血糖（Glucose AC）、血清麩氨基酸草醋酸轉氨基酵素（SGOT-AST）、血清麩丙酮酸轉氨基酵素（SGPT-ALT）（p＜0.001）有顯著差異；而收縮壓（Systolic blood pressure）、舒張壓（Diastolic blood pressure）（P＞0.05）無顯著差。運用類神經網路模型預測有無脂肪肝，所獲得結果測試組的特異度（specificity）90.0%，敏感度（sensitivity）86.4%, ROC area:0.919。訓練組加測試組的特異度（specificity）82.1%，敏感度（sensitivity）89.7%, ROC area: 0.907。此將有助於醫師判斷有無脂肪肝以及預防肝臟產生後續之病變，並可達到早期發現及早治療之目標。

In recent studies, fatty liver disease (FLD) has been shown to be closely related to the Biomarker, Abnormal liver biochemistry, Central Obesity, Dyslipidemia and Metabolic Syndrome. Especially for people who are obese, the chance of being diagnosed with fatty liver disease has been increased. Metabolic syndrome is a medical disorder which will increase the risk of cardiovascular disease and diabetic development. People who have large abdominal girdle are also in high risk of metabolic syndrome. This study is using the data of liver biochemistry, Utilized SPSS, and t-test to make a statistical analysis of parameters between fatty liver and non-fatty liver disease. The following variables which are selected based on the Age, Abdominal girdle, Triglyceride, Glucose AC, SGOT-AST, and SGPT-ALT. Although statistically, all of their P Values are shown less than 0.001, but, for systolic and diastolic blood pressure were both show P＞0.05. Statistically show no significant difference. In this study, Artificial Neural Networks (ANNs) were applied to predict fatty liver disease. By using abdominal ultrasonography, for gold standard reference test, we found in testing group: 90.0% of specificity, 86.4% of sensitivity, ROC area 0,919; and in overall group: 82.1% of specificity, 89.7% of sensitivity, ROC area 0.907. The contribution of this study is to prevent the liver disease by early detection and treatment.