因為晚婚和壓力所造成的不孕症人口愈來愈多,台灣於2008年採用人工生殖的人數甚至已突破8000人次,活產嬰兒數也高達3000名,若以當年度新生兒總數約20萬人來估算,人工生殖治療的出生寶寳比率已佔了總出生率1.5%。雖然台灣人工生殖技術懷孕率高達36.5%,活產率則由民國87年的22.2%,逐年提升到97年的27.1%,但在不能保證懷孕的情況之下,動輒數十萬的治療費讓許多不孕婦女怯步,對許多不孕症夫婦而言,是否要進行這些須耗費時間與金錢的治療是個難題,且治療後能成功懷孕的比率又是多少?這些問題一直困擾著不孕症夫婦們。 在治療時,病患年齡與卵子及精子的品質等許多不同因素,都會對人工生殖的成功率造成影響。此外,進行人工生殖的整個流程步驟,其內容複雜且每一細部的環節皆會影響到懷孕率最終成果。因此,本研究將應用資料探勘(Data Mining)中的特徵選取(Feature Selection)技術,如主成份分析法(Principal Component Analysis, PCA)、線性區別分析(Linear Discriminant Analysis, LDA)及決策樹(Decision Tree, DT)結合台北某教學醫院生殖醫學中心所提供的資料,透過分析運算來瞭解人工生殖診療過程中,不同的醫囑診斷、身體素質等是否會成為對懷孕率造成影響的主要因素,希望能從大量的醫療診斷資料中篩選出有意義的屬性,並透過分類(Classification)方法中的類神經網路(Artificial Neuron Network, ANN)進行懷孕率預測,以提供醫師醫療診斷決策資訊,更精確地掌握及安排患者的療程。
Because of the reason by late marriage and pressure cause the infertility population getting more and more. In 2008, Taiwan has using the artificial reproduction technology to treatment infertility that has exceeded 8,000 people, and the number of live births as high as 3000, if we using the year neonatal terms of the total number of about 20 million people to estimate, using artificial reproductive technique of babies born accounted for a total birth rate of baby birth rate of 1.5%. Although Taiwan in the assisted reproductive technology has high pregnancy rates as 36.5%, and from 1998 to 2008, the live birth rate by year raise from 22.2% to 27.1%. However, due to the high cost of the assisted reproductive technology treatments, uncertainty of live birth rate and the pressure it causes to patients physically and mentally, how to predict the possible IVF outcomes becomes an important research topic. In the treatment process, the pregnancy rate of the artificial reproductive technology will be impacted through by the female patient age, ovum quality or the male patient sperm quality and many different factors etc. In addition, the entire process of artificial reproduction technology treatment, from the patient's basic information such as age, weight to into the operating room surgical ovum retrieval, fertilization, embryo culture and implantation, every single step will affect the final outcome. In this research, a hybrid data mining methodology is used to predict the treatment of the assisted reproductive technology outcome. Firstly, Principal Component Analysis (PCA), Linear Discriminant Analysis(LDA) and Decision Tree (DT) will be adopted respectively to select features. And then, the classification methods, Artificial Neural Network (ANN) is used to predict the outcome of IVF. Combination of each feature selection method as well as classification method is compared to test the accuracy of prediction. The IVF dataset used in this research is obtained from one of the teaching hospital in Taiwan. The results of this research will provide a good reference for doctors in ART unit when they are asked to give an answer of success rate to infertility couples.