透過您的圖書館登入
IP:54.163.14.144
  • 學位論文

心房顫動患者罹患缺血性中風之評估研究

A Study of Ischemic Stroke Patients with Atrial Fibrillation

指導教授 : 張俊郎
若您是本文的作者,可授權文章由華藝線上圖書館中協助推廣。

摘要


腦中風是腦血管堵塞或破裂,造成腦細胞受到傷害而引發神經症狀。中風成因甚廣,約10-15%與心臟疾病直接相關。而心房顫動是臨床上最常見的一種心律不整,依盛行率調查,估計約2%的民眾曾經有過心房顫動的發生且心房顫動的病人發生腦中風的機會是正常人的五倍以上。故本研究利用台灣近年某醫療機構資料庫,心房顫動患者罹患缺血性中風之病歷資料做為研究對象,藉由蒐集相關文獻及與專業醫師訪談挑選出影響缺血性中風之因子變數,利用粒子群演算法、交叉熵演算法及基因邏輯斯迴歸演算法分別結合倒傳遞類神經網路、支援向量機建構心房顫動患者罹患缺血性中風的六種預測模型;再利用三種演算法之權重值導入案例式推理技術建立心房顫動患者罹患缺血性中風評估系統。研究結果顯示,預測模型方面,模型間存在顯著差異,其中以粒子群演算法結合的兩種模型最佳,平均準確率及平均ROC曲線下面積皆有88%及0.85以上;評估系統方面,三種演算法之間無顯著差異,皆適合做為評估系統之權重值,平均準確率及平均ROC曲線下面積皆有85%與0.80以上。本研究建構之預測模型及評估系統,能提供醫療機構與相關醫療人員做為輔助診斷與評估之參考依據,從預防醫學之角度對於早期發現疾病以避免醫療資源的耗用上,將會有所助益。

並列摘要


Stroke is a cerebrovascular obstructive or cerebrovascular rupture, resulting in damage to brain cells and neurological symptoms. Cause of stroke is very wide, about 10-15% is directly related to heart disease. The atrial fibrillation is clinically the most common type of arrhythmias. According to the prevalence survey, it is estimated that about 2% of the population have gotten atrial fibrillation. The chance of atrial fibrillation patients get stroke is more than five times the normal. Therefore, in this study, patients with atrial fibrillation in the database of an anonymous medical institution in Taiwan were adopted as research participants. Through the collection of relevant literature and interview with professional physicians to select the factors that affect the ischemic stroke, using particle swarm optimization, cross entropy and genetic algorithms logistic regression combined with back propagation neural network and support vector machines to construct six predictive models of ischemic stroke patients with atrial fibrillation. In addition, using weight of three algorithms combined with case-based reasoning technique to construct evaluation system of ischemic stroke patients with atrial fibrillation. Research results show that there are significant differences among six predictive models. Among these models, the best two models are constructed by particle swarm optimization that the average accuracy rate and the average area under the ROC curve are both over 88% and 0.85. For the evaluation system, there are no significant differences among three algorithms. Thus, three algorithms are all suitable for the weight of the evaluation system. The average accuracy and average area under the ROC curve are both over 85% and 0.80. The prediction models and evaluation system constructed in this study can provide medical institutions and relevant medical personnel as a reference for assisting diagnosis and evaluation. From the perspective of preventive medicine, there will be a help to early detection of diseases to avoid the consumption of medical resources.

參考文獻


陳信源、葉鎮源、林昕潔、黃明居、柯皓仁、楊維邦(2009),「結合支援向量機與詮釋資料之圖書自動分類方法」,資訊科技國際期刊,3卷,1期,頁2-21。
李天行、邱群穎、呂奇傑(2016),「應用支援向量機於相關性製程監控階梯式干擾」,Journal of Data Analysis,11卷,4期,頁1-20。
謝日章、吳浩佑(2007),「應用案例式推理法建構財務危機預警系統」,萬能商學學報,12期,頁151-162。
趙子凡、段大全、江志桓(2011),「心房顫動藥物治療新曙光」,台北市醫師公會會刊,55卷10期,頁30-32。
林盈利、陳清埤、余昭宏、林益卿(2012),「心房顫動與抗血栓治療」,家庭醫學與基層醫療,27卷,6期,頁202-206。

延伸閱讀