運用人工智慧於高尿酸血症現象之研究

高尿酸血症現象的引發狀況，近幾年來已被視為診斷慢性或退化性疾病的重要合併症因素之ㄧ，同時根據衛生署國人營養的調查，台灣地區成人罹患高尿酸血症的比率，男性達1/4、女性達1/6，尿酸過高的人數更高達500萬人。而高尿酸血症(hyperuricemia)是早期引起痛風(gout)疾病的有效生化數據檢測的指標與必要條件，高達九成七以上的痛風患者均有高尿酸血症，但並非所有的高尿酸血症患者均會發生痛風，當高尿酸血症嚴重時，則會造成痛風、腎臟功能衰退等嚴重疾病，若不加以治療，甚至會引發多重併發症，導致重大疾病的發生，危害到患者的生命安全。本研究運用人工智慧的方法，以個案醫療機構的疾病防治中心資料庫為研究對象，分別使用倒傳遞類神經網路、決策樹與決策樹結合倒傳遞類神經網路，三種模型相互評估比較，建構引發高尿酸血症現象之輔助診斷預測模型，研究結果顯示決策樹結合類神經網路模型在分類預測上有較佳的分析成效，其平均測試準確率為90.28%，顯示決策樹結合類神經網路模型對引發高尿酸血症現象之預測有較佳的解釋能力。本研究透過人工智慧的方法，挖掘資料間的關聯性與分類規則，以協助醫師臨床診斷之參考依據，並提供一般民眾對於高尿酸血症的認知，瞭解自己的身體健康狀況，有效控制生活型態的改變，定期進行高尿酸血症的檢查，早期發現、早期治療，降低罹患高尿酸血症之發生率，對於預防慢性疾病的發生與提升生活品質更有實質助益。

關鍵字

高尿酸血症；痛風；資料探勘；決策樹；類神經網路

並列摘要

Hyperuricemia is nowadays considered one of the important complications to diagnose chronic or regressive illnesses. According to the public nutrition investigation conducted by the Health Department in Taiwan, one out of four male adults and one out of six female adults developed hyperuricemia and the population with uric acid above the normal range reached 5 millions. Hyperuricemia is an essential condition and effective bio-chemical data for testing early stage of gout. Over 97% of gout patients also develop hyperuricemia; however, not all patients with hyperuricemia will necessarily develop gout. Patients with severe hyperuricemia symptoms will very likely develop other serious diseases such as gout, and/or kidney function decline (or renal failure). Without proper treatment, multiple complications might develop thus leading to the onsets of other major diseases and eventually put patients’ lives in jeopardy. In this study, artificial intelligence methodologies were utilized, using the data base of a case medical institution as the subject, we evaluated and compared the three different models using BPN network, decision tree, and decision tree with BPN networks respectively to construct a predictive model to help determine the causes of hyperuricemia. The results show that using decision tree combining with artificial neural network model has more optimal analytical performance in the classification prediction with a mean accuracy at 90.28%, indicating that the combination model has better explanatory ability to the causes of hyperuricemia. This study also attempted to explore the association and classification rules among data through artificial intelligence methods to help provide physicians the reference in clinical diagnosis as well as to promote health knowledge on hyperuricemia, and to increase public awareness of their own health conditions, hoping for them to effectively manage their changes of life styles and taking routine examinations for hyperuricemia. Early diagnoses and early treatments can then be provided to people at risk so to reduce the onset of hyperuricemia. The implementation of the system can provide more practical benefits to prevent chronic diseases from developing and to enhance the quality of life.