運用類神經網路評估缺血性腦中風病患於靜脈內血栓溶解劑治療預後

Prognosis Assessment of Ischemic Stroke Patients after rt-PA Treatment by Using Artificial Neural Network Methods

林宜菁(Yi-Ching Lin 林宜菁)

指導教授 ： 邱泓文

臺北醫學大學/醫學科技學院/醫學資訊研究所/碩士(2013年)

https://doi.org/10.6831/TMU.2013.00066

摘要

腦中風在台灣俗稱中風，美國稱為Stroke，醫學上稱為腦血管疾病。在美國每年有700,000人發生腦中風，梗塞性腦中風佔了80%，臺灣佔75%。在美國每年在腦中風的照護上需花費將近400億美元，而在台灣每個月要花費1.5-4萬，其住院天數長，這都造成國家、病患及家屬心理上及經濟上的負擔，而不能達到雙贏之狀態，因此若能將資訊科技的技術運用至預測缺血性腦中風患者上，相信會有更好的效果。
2002年台灣正式開始使用靜脈內血栓溶解劑治療（以下簡稱rt-PA）治療急性缺血性腦中風。缺血性腦中風病患rt-PA在符合準則下使用rt-PA，是目前治療腦梗塞最有效以及安全的方法。而在準則規範後使用rt-PA的死亡率均有下降，不過對於預後行為能力好壞並無法有效預測需要即早提供訊息，以提供建立後續照顧者照護模式。
類神經網路是人工智慧的一支、一種模仿生物神經網路的資訊處理系統、大量簡單的相連人工神經元來模仿生物神經網路的能力、藉以處理高複雜度、非線性之問題。本研究即是希望以類神經網路來分析缺血性腦中風病人使用rt-PA的危險因子及預後兩種方式交叉預測的關係，希望由這些安全又經濟的預測就可以提早檢測出缺血性腦中風病人之預後狀況、進而提早治療和預防。
本次研究利用教學醫院蒐集範圍為 2005年07月至2012年06月間，共收集82位缺血性腦中風病患靜脈內血栓溶解劑治療的資料，病患性別、年齡、高血壓、心臟相關疾病、糖尿病、吸菸及NIHSS量表分數，運用類神經網路評估於病患治療後3個月之狀態。研究方法利用SPSS18與statistica10分類出測試組與訓練組的資料進行比較，statistica10之類神經網路方式選出最佳訓練模型，預測結果評估方法，如敏感度、特異度、精確度、及測試組與訓練組、全體樣本之ROC curve，評估比較分類器之好壞或檢驗或診斷之正確性，面積越大表示效果越佳。
研究結果發現，類神經網路模型比邏輯迴歸模型有更好的辨識能力。本研究所以使用之八項變數的類神經網路預測模型最佳模型，以測試組類神經網路模型的曲線下方面積AUC =0.85與訓練組神經網路模型的曲線下方面積AUC=0.98，面積皆大於各個變項在預測預後狀況好與不佳之AUC，可得知在利用類神經網路模型預測缺血性腦血管病患於使用溶栓治療預後狀況好與不佳。
本研究所建構之類神經網路預測模型-缺血性腦中風病患於靜脈內血栓溶解劑治療預後有較好的預測能力，於預後功能評估推斷，當輸入變數呈現非線性組合時，能更準確提供預測評估缺血性腦中風病患於靜脈內血栓溶解劑治療預後。醫療領域找出因子與因子間與病症間關係是重要的。與其他文獻中提出之影響因子分析的找出相關影響程度，也得到不錯的結果。提供臨床醫事人員和腦中風病患（及家屬）更多治療及預後評估的參考與建議。

關鍵字

缺血性腦中風,靜脈血栓溶解治療,美國國家衛生生院中風量表,類神經網路

並列摘要

Commonly known as stroke, stroke in Taiwan, the United States called Stroke, medically known as cerebrovascular disease. There are 700,000 people in the United States each year occur stroke, ischemic stroke accounted for 80% and Taiwan accounted for 75%. Stroke each year in the United States takes on the care of nearly $ 40 billion, to spend a month in Taiwan 1.5-4 million, the length of their hospital stay, which all contribute to national, patient and family psychological and economic burden, but can not achieve win-win state, therefore, it would be the use of information technology techniques to predict ischemic stroke patients, I believe there will be better results.
2002 Taiwan officially started using intravenous thrombolysis (hereinafter referred to as rt-PA) for acute ischemic stroke. Ischemic stroke patients met criteria for rt-PA using rt-PA, is currently the most effective treatment of cerebral infarction and secure method. And after the standards governing the use of rt-PA were decreased mortality, but for good or bad prognosis incapacitated and unable to effectively predict immediate early to provide information needed to establish subsequent caregivers provide care model.
Neural networks are an artificial intelligence, an imitation of biological neural network information processing system is connected to a large number of simple artificial neurons to mimic the ability of biological neural networks, in order to deal with highly complex, non-linear 'questions. This study is hoped to neural networks to analyze ischemic stroke patients with rt-PA risk factors and prognosis are two ways to cross-prediction and hope that by the security and economic forecasts can be detected early ischemic The prognosis of stroke patients, and thus early treatment and prevention.
The study, teaching hospitals to collect range from July 2005 to June 2012, a total collection of 82 ischemic stroke patients intravenous thrombolysis information, patient gender, age, hypertension, heart related disease, diabetes, smoking and NIHSS scale scores, the use of neural network evaluation in patients after 3 months of treatment status. Research Methods SPSS18 the test group with statistica10 classified information and training group comparison, statistica10 elect the best neural network training model to predict the outcome evaluation methods, such as sensitivity, Specificity, precision level, and the test group and training group, all samples ROC curve, comparative assessment of the quality classifier or inspection or the accuracy of diagnosis, the larger the area the better the results indicate.
The results showed that the neural network model is better than the logistic regression model has better recognition ability. Therefore, this study used eight variables neural network forecasting model best model to test the neural network model set of area under the curve AUC = 0.85 with the training group neural network model area under the curve AUC = 0.98, area are greater than the individual variables in predicting the prognosis is good and poor condition of the AUC, can be learned in the use of neural network model to predict ischemic cerebrovascular use of thrombolytic therapy in patients with good and poor prognosis.
By this research neural network forecasting model - Ischemic stroke patients for intravenous thrombolytic agents have a better prognosis predictive ability in functional outcome assessment concluded that when the input variables that showed nonlinear combination, can provide a more accurate assessment of ischemic stroke patients predict intravenous thrombolytic agents in treatment and prognosis. Medical field to identify factors and factor between the relationship between the illness is important. And the impact of other factors proposed in the literature to identify relevant impact analysis, but also get a good result. Provide clinical medical personnel and stroke patients (and their families) more treatment and prognosis of reference and suggestions.

並列關鍵字

ischemic stroke, intravenous thrombolytic therapy（rt-PA）, National Institute of Health Stroke Scale（NIHSS）, Artificial neural networks（ANNs）

參考文獻

江俊宜、賴昱樺、熊光華、黃建華、彭家勛、陳建生、陳穎信（2010）。整合到院前及到院後急性缺血性腦中風溶栓照照護以北臺灣為例：前趨研究。Journal of Taiwan College of Emergnecy Physician,3（2），74-82。

李雅珍、吳姿誼、邱恩琦、涂富籌、謝清麟（2012）。中風病患整體健康相關生活品質之關聯因素探討。職能治療學會雜誌，30（1），53-67。

林俊杰（2013）。老式抗生素在急性缺血性腦中風的角色。台灣腦中風學會會訊，20（1），15-18。

林景嘉、鄭信鈴、王泰裕（2010）。以類神經網路建構再發性腦中風之預測模式。醫療資訊雜誌，19（2），19-38。

孫穆乾（2011）。急性腦中風的靜脈血栓溶解治療 。臺灣醫學，15（5），535-542。