透過您的圖書館登入
IP:52.91.255.225
  • 學位論文

膽結石病患伴隨中風之評估研究

A Study on the Assessment of the Gallstone Disease Associated with Stroke

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


膽結石疾病和中風都是近代常見的公共健康問題。膽結石疾病與中風的相互關係很少有研究做分析,但卻是有共同的危險因素,過去的研究都認為膽固醇累積在膽囊中以及動脈壁上的硬化,過多的膽固醇、高膽固醇血症和高血脂症以及其它條件如高血壓、糖尿病、周圍血管疾病、慢性腎病、酗酒和慢性肺阻塞疾病的條件是公認的危險因素。 本研究以某醫療機構資料庫中膽結石病患為研究對象,透過文獻探討與醫師訪談,篩選出增加伴隨中風之重要因子後,運用人工智慧中粒子群最佳化演算法、基因邏輯斯迴歸演算法、交叉熵演算法計算因子權重值,以建構預測模型與案例式評估系統,評估是否後續會有中風伴隨的風險。 研究結果顯示,六項預測模型經由統計檢定以交叉熵演算法、基因邏輯斯迴歸演算法與粒子群最佳化驗算法結合支援向量機表現較為優異,模型之間經傅利曼檢定預測模型準確度方與ROC曲線下面積為皆為顯著差異,再利用成對樣本T檢定來判別模型的優異性,經判別後粒子群最佳化演算法結合支援向量機為最佳預測模型,然而K疊驗證後其平均準確度與ROC曲線下面積皆有89%與0.89以上;結合最佳參數進行中風分類以交叉熵演算法結合倒傳遞類神經與支援向量機具有較佳的準確率與ROC曲線下面積98%與0.95,最適合進行中風分類;在是否中風風險評估系統方面,經K疊驗證後三種演算法結合案例式推理其平均準確度及平均ROC曲線下面積皆有93%與0.85以上。評估模型之間經傅利曼檢定準確度及ROC曲線下面積為顯著差異,表示模型間存在差異性。 經判別後粒子群最佳化演算法結合案例式推理為最佳,平均準確度與ROC曲線下面積皆有94%與0.91以上。 本研究結果可提供醫療機構及臨床工作者,做為疾病診斷評估之參考依據,並可指導病患正確治療與後續療育以減輕病患疾病負擔。

並列摘要


Gallstone disease and Stroke are public health problems of the modern times. Research analysis on the relationship between gallstone disease and Stroke remains scarce. Past studies have considered the cholesterol deposit in the gallbladder, hardening of the arterial wall, excessive cholesterol such as hypercholesterolemia and hyperlipidemia, and other conditions such as hypertension, diabetes, peripheral vascular disease, chronic kidney disease, alcoholism, and chronic pulmonary disease conditions to be recognized risks factors for Stroke. In this study, gallstone disease patients from the database of an anonymous medical institution were adopted as the research participants. Through literature reviews and interviews with physicians the important factors contributing to increased risk of accompanying Stroke were screened. Then, using the artificial intelligence particle swarm optimization algorithm, genetic logistic regression algorithm cross-entropy algorithm, and case-based reasoning system, the factor weights were calculated in order to construct predictive models and case-based assessment system and assess whether or not the risk of accompanying Stroke disease existed. Findings show that among the six predictive models that underwent statistical testing, particle swarm optimization algorithm, the genetic logistic regression algorithm, cross-entropy algorithm coupled with support vector machine was the best combination. The Friedman test verified the predictive model accuracy between the models was significant, indicating difference existed between the models but a significant area under ROC curve. The paired sample t-test was then conducted to determine the superiority of the models. After determination, particle swarm optimization algorithm coupled with the support vector machine was the best predictive model. However, through the k-fold validation, the average accuracy and the area of ROC under curve all reached over 89% and 0.89. The optimized parameters were conjunctively used to carry out stroke classification. The cross-entropy algorithm, coupled with back-propagation network, BPN and support vector machine, SVM derived at a better accuracy rate and the area under the ROC curve of 98%, 98% and 0.95 respectively, which were deemed most suitable for stroke classification. As for the accompanying Stroke disease system assessment, though the k-fold validation, three algorithms coupled with the case-based reasoning had average accuracy and average area under ROC curve reaching above 93% and 0.85. The assessment models verified by the Friedman testing had significant accuracy and area under ROC curve, indicating differences existed between the models. The paired sample t-test was then adopted to analyze the superiority of each model set. Among them, particle swarm optimization algorithm was the best. The average accuracy and the area of ROC under curve all reached over 94% and 0.91. The research results shall serve as a reference for medical institutions or clinical workers during disease diagnosis and assessment and provide patients with early treatment and prevention in order to lessen the load of patients with disease.

參考文獻


19. 孫瑛霙、王明庸、藍天雄(2015),「應用倒傳遞神經網路於負載啟斷開關接觸電阻預測之研究」,管理資訊計算,4卷,1期,頁142-149。
8. 古國宏(2014),「類神經網路在運動科學界的相關探討」,休閒運動保健學報,7期,頁106-115。
10. 李維平、王雅賢、江正文(2008),「粒子群最佳化演算法改良之研究」,科學與工程技術期刊,4卷,2期,頁51-62。
40. 賴春生、呂佩穎、陳自諒(2016),「外科的臨床推理教學」,台灣醫學, 20卷,3期,頁290-296。
34. 楊菀婷、賴美淑、高嘉宏(2016).,「以全民健康保險研究資料庫探討臺灣慢性 C 型肝炎患者之抗病毒藥物治療之利用」,臺灣醫學,20卷,2期,頁 124-133.

延伸閱讀