肝病,是台灣地區最主要的本土病。根據衛生署 2007年統計資料顯示,估計台灣每年約有一萬人以上死於肝炎、肝硬化、肝癌。因為肝病在初期症狀並不明顯,要等到病情相當嚴重時才會出現症狀,所以目前學者研究以建立一個輔助肝病診斷機制為主要議題之一。故本研究實際蒐集病患資料,使用多種資料探勘方法並整合專家們的意見,並運用基因演算法 (genetic algorithm, GA) 尋找最佳組合解的能力,建立一套最佳化的整合型肝病輔助診斷模式 (integrated liver diagnosis model, ILDM)。並且使用多元適應性雲形迴歸 (multivariate adaptive regression splines, MARS) 獲得較重要的肝病診斷變數,期望能建立更有效率的診斷模式。 研究結果顯示,整合資料探勘方法所建構的診斷模型表現優於單一方法,且 GA 能減少建構所有診斷模型所要耗費的與成本,快速找到最佳的診斷模型。另外,經由 MARS 所建立的診斷模型,表現優於未篩選變數的模型。及所建立的肝病輔助診斷模式,可以降低因誤判所造成的延誤就醫的可能性,並且節省醫療成本,減少不必要的檢查。
Liver disease is the most common local disease in Taiwan. According to the statistics from Department of Health in 2007, around ten thousand people die from liver cirrhosis, liver cancer and other liver diseases because the symptoms of liver disease are not obvious in the initial stage, and the condition is usually too serious to be treated when related symptoms make themselves felt. Developing an assisted liver disease diagnosis model has therefore become a major issue attracting growing attention from scholars and researchers. This study accordingly aims at constructing an optimal integrated liver disease diagnosis model (ILDM) by collecting patient data, using data mining techniques, integrating expert opinions, and utilizing genetic algorithm that is capable of finding best combination of diagnosis models. Moreover, MARS (multivariate adaptive regression splines) is adopted to obtain significant diagnosis variables, helping to construct a more efficient diagnosis system. As the results reveal, the integrated data mining techniques of diagnosis model outperforms the single data mining techniques of diagnosis model. Using GA helps reduce the time and cost spent on model construction and speed up the identification of the best combination of ILDM. In addition, the diagnosis model established by MARS outperforms the diagnosis model with no screening variables. ILDM can be expected to decrease the possibility of delays in medical treatment caused by wrong diagnosis and save medical costs by eliminating unnecessary inspections.