運用人工智慧於孕婦早產現象之研究

在台灣近年來每年約有20多萬名新生兒（據內政部統計處公佈96年出生人口數為20萬3711人），但每年卻有約 3097 ~ 4130 名新生兒死亡，早產的發生率約佔所有懷孕的5~10%，但卻佔新生兒死亡的80%。伴隨早產而來的各種急、慢性問題，使得早產家庭窮於應付。比起一般新生兒而言早產兒需耗用更多醫療資源例如：新生兒加護病房或是保溫箱。面對這樣常見而影響深遠的問題，如何充分利用現有的知識，儘量降低早產的發生，值得大家深思。本研究運用人工智慧建立醫療診斷上的模型以提供醫師做全方位的判斷，藉由多次產檢資料來預測是否會發生早產現象，提供一個合理良好的安胎環境，進而研究出更積極的、完整性的輔助診斷系統使醫師、孕婦和早產兒之間做更緊密的資訊交流與療育規劃，降低早產兒出生死亡率。本研究採用決策樹和類神經網路來作為探討孕婦早產現象之研究工具，研究結果顯示，在本胎具有早產現象的領域中以倒傳遞類神經網路作分析有較佳的預測效果，其測試結果為93.34%，在住院安胎現象中以決策樹C5.0正確分類率93.44%為最佳，依此建立診斷系統以利在產前及早找出關鍵因素進行有效防治與療育。

關鍵字

早產；危險因子；決策樹；類神經網路

並列摘要

According to the data of Ministry of Interior, there are approximately 200,000 neonates in Taiwan per year, but the death number of neonates is about 3097 to 4130. The happening rate of premature is 5~10% among the pregnant, but among death of neonates, the happening rate of premature accounts for 80%. Every kind of acute or chronic diseases accompanying with premature are hard to handle for those premature families. A premature infant needs to consume more medical resource than a normal neonate such as intensive care unit or incubator. Hence, how to utilize available knowledge in order to lower down the premature rate is considerable. This research applies artificial intelligence to establishing model of medical diagnose in order to offer physician to do omnibus diagnosis. We analyze the data of repeated antenatal examinations to predict whether the premature happens or not so as to provide a good environment to stabilize fetal position. Furthermore, we deliberate a positive and integrity auxiliary diagnosing system in order that the physician, expectant mother, and premature infant could have closer information communication, therapy and birth plan to decrease death rate of premature infant. This study applies decision tree and neural network to explore premature phenomenon. The research indicates that in the field of premature phenomenon, back-propagation N.N analysis has better predictive effect, and the testing result is 93.34%. Besides, in the field of being hospitalized, according to decision tree C5.0, the direct classifying rate 93.44% is the best. We can use it to establish diagnosing system in order to find out the main factor before parturition and then to undertake prevention and treatment.