試管嬰兒(In Vitro Fertilization, IVF)治療是昂貴的,且對受療者的身心,都是一項考驗,IVF病人最常問的問題,就是「我成功的機率是多少?」。醫師通常根據病人的年齡、卵泡刺激素(follicle stimulating hormone)及不孕症的診斷,決定成功懷孕機率的大小,但有許多其他因素也會影響IVF的成功率,如卵子及精子的品質、殖入胚胎的數量及是否使用顯微注射技術(ICSI)等,這些複雜的因素使臨床醫師很難針對每一組IVF受療者擬定個別的治療決策,以達到最佳的成功率。本研究的目的在於應用資料探勘(Data Mining)技術於IVF資料庫之分析,構建出一個IVF結果的預測模式;同時找出預測IVF成功或失敗的規則,藉以探討影響IVF結果的因素之間的關係。為了達到較好的探勘結果,本研究使用三種混合(hybrid)資料探勘方法分析IVF資料庫:(1)結合複迴歸(multi-regression)與決策樹(Decision Tree)學習;(2)結合主成份分析(Principal Component Analysis)與決策樹學習;(3)結合基因演算法(Genetic Algorithm)與決策樹學習。比較四種模型後,GA結合DT之準確率為74.8%優於其它三種。由決策樹模型所得到的規則經與醫生確認後在臨床上是有診斷的有效性並與文獻吻合。希望藉由人工生殖技術的關係來幫助醫生與病人在溝通上、倫理上有共識,增進醫生與病患之間的和諧。本研究建構之模型對於醫生之臨床診斷將有實質之助益。
In vitro fertilization (IVF) treatment is expensive. Doctors are usually based on the patient age, follicle stimulating hormone (FSH) and infertility diagnosis, decision pregnancy rate. However, many more parameters are know to impact the IVF success rates. These complex factors cause clinician to be very difficult in view of each group of IVF to draw up curing the individual treatment decision-making. The purpose of this study is to apply data mining technology in IVF database analysis, build a forecast model of IVF results. In order to achieve the good exploration result, this study aims to use three kinds to mix (hybrid) in IVF database analysis:(1) multi-regression combine Decision Tree;(2) Principal Component Analysis combine Decision Tree;(3) Genetic Algorithm combine Decision Tree. Contrast with four models, the accuracy of genetic algorithm combine decision tree is 74.8% which is superior to the other three kinds. After confirmed by doctor, the rule from decision tree is equipped with clinical diagnostic variables and corresponds with literature. The hope helps doctor and the patient because of the assisted reproduction technology in the communication, ethics has the mutual recognition, promotes between doctor and patient harmony. The model constructed is beneficial to doctors’ clinical diagnosis.