目的:隨著醫學科技的進步,許多醫學資料與日俱增地累積下來,成為一個龐大的資料庫。面對這大量未經整理的資料庫,我們可經由資料探勘找尋深藏在資料內容的資訊,再經由專家作更進一步之解釋研究。方法:本文以類神經網路、分類迴歸樹演算法,找出診斷肝癌患者的分類模式。研究樣本為2005年仁愛院區共978筆資料,其中包含326筆肝癌患者與652筆非肝癌患者的檢驗資料。先以T檢定與卡方檢定篩選有顯著差異之變數,再使用類神經網路與分類迴歸樹進行建構其分類模式。結果:研究結果發現:1.整體而言分類迴歸樹演算法敏感度83.8%優於類神經網路(72.8%)。2.經由統計結果顯著的變項包含(Alkaline Phosphatase、AST(GOT)、ALT(GPT)、r-GT、Total Bilirubin、Gender、AGE)。結論:經由這些變項便可提供大部分的分類模式,並可提供醫界與未來研究者進一步研究發展。
Background and Purpose: According to the advancement of the medical technology, lots of medical data has been accumulated with each passing day. To face there large amount of data, data mining technology has become an essential instrument for hospital management and medical research. The possible meaningful information is found out from the database by data mining. And further research continued by physician to find out the mechanism of it. The aim of this study is to look for the determine factor of Liver Cancer. The data is collected from Ren-Ai Hospital from Jan. 2005 to Dec. 2005. We find out the significant different variables by statistical analyses first. And seven significant different variables are loaded into the decision tree algorithm and Artificial Neural Network to conclude determine factor of Liver Cancer. Methods: The CART methodology find that GOT variable is the major determine factor of Liver and having 39.0 U/L as cut point. Results: In other words, decision tree algorithm is able to conclude rules compatible with medical knowledge. In the validation data set, the ANN Model demonstrated accuracy and sensitivity were 81.9% and 86.9%, the CART Model showed accuracy and sensitivity were 82% and 81.3%. Conclusion: CART analysis to the liver cancer decision results in a significant reduction in unnecessary variables while retaining a high degree of sensitivity when compared with the Artificial Neural Network.