以資料探勘技術探討顱內動脈及顱外動脈狹窄的相關因子之研究

Applying Data Mining Technology to Exploring the Relevant Factors of Intracranial and Extracranial Arterial Stenosis

李文雄(WEN-HSIUNG LEE)

指導教授 ： 林昭維 ； 阮金聲

國立中正大學/管理學院/資訊管理系醫療資訊管理研究所/碩士(2016年)

摘要

腦血管疾病(簡稱腦中風)一直為臺灣的主要死因，因為腦中風的治療與照顧的進步，腦中風的致死率逐漸下降，顱內動脈及顱外動脈狹窄的成因是粥狀硬化斑塊沉積使得血管內膜增厚，進而阻塞血管或形成血栓剝落，是造成腦部缺血引起中風的重要原因。而顱內動脈及顱外動脈狹窄多是因動脈粥狀硬化所引起，顱內動脈及顱外動脈狹窄引起腦梗塞部分是因為嚴重狹窄導致腦血流不足，或是因為動脈硬化斑破裂產生的栓子阻塞遠端腦血管。 本研究來自中部某區域教學醫院，對象為101年曾於腦血管超音波室，執行腦血管超音波檢查之民眾。運用檢查資料與生活習慣資料，藉由嘗試多種資料探勘技術的分析，以分類技術來建立顱內動脈及顱外動脈狹窄的預測模型，進而採取因應之道以減少腦中風的機會，並協助醫師執行臨床醫療決策時的參考。
研究結果，所建構預測顱內動脈狹窄預測效能最佳的是決策樹模式為最佳，其正確率為78.41%(ROC曲線為82.2%)，其次為邏輯斯迴歸正確率為69.78%(ROC曲線為65.6%),最後為貝氏網路正確率66.18%(ROC曲線為63.9%)。而所建構預測顱外動脈狹窄預測效能最佳的是決策樹模式為最佳，其正確率為92.78%(ROC曲線為81,1%)，其次為邏輯斯迴歸正確率為89.42%(ROC曲線為77.0%),最後為貝氏網路正確率88.94%(ROC曲線為80.3%)。決策樹的分類器分析結果則較符合實際狀況且分類規則易懂。顱內動脈狹窄過程中，服降血壓藥物的治療扮演重要角色；而顱外動脈狹窄過程中，空腹血糖扮演著重要的角色。

關鍵詞:腦血管疾病、貝氏網路、邏輯斯迴歸、決策樹

關鍵字

邏輯斯迴歸 ； 貝氏網路 ； 腦血管疾病 ； 決策樹

並列摘要

Cerebrovascular disease (stroke for short) has been the main cause of death in Taiwan, because of the improvement of treatment and caring for stroke, the death rate has been decreased gradually. The stroke followed by brain ischemia is related to blood vessel intimal thickening due to blocking vascular thrombosis or peeling which is caused by atherosclerotic plaque deposition from intracranial and extracranial artery stenosis. The intracranial and extracranial artery stenosis results from atherosclerosis. Severe vessel stenosis leads to insufficient blood flow or cerebrovascular occlusion due to arteriosclerotic rapture, both are the main causes for intracranial and extracranial artery stenosis .
In this study, the data source comes from a central teaching hospital, a cerebrovascular ultrasound room. Data sampling period is within year 2012. By utilizing life-style questionnaire, checking results and various data mining, we attempt to establish predictive models for classification intracranial and extracranial artery stenosis to reduce the chances of stroke. The result also can be reference for clinical decisions.The best predictive model for intracranial and extracranial artery stenosis is decision tree, the correct rate of 78.41% (ROC curve of 82.2%), followed by logistic regression accuracy was 69.78% (ROC curve 65.6%), and finally the Bayesian network accuracy rate 66.18% (ROC curve of 63.9%). The forecast extracranial artery stenosis constructed to predict the effectiveness of a decision tree model is the best for the best, the correct rate of 92.78% (ROC curve was 81,1%), followed by logistic regression accuracy was 89.42% (ROC curve 77.0%), and finally the Bayesian network accuracy rate 88.94% (ROC curve of 80.3%). Decision tree classifier analysis results are more in line with the actual situation and the classification rules to understand. Intracranial arterial stenosis process, serving antihypertensive drug therapy plays an important role; and extracranial artery stenosis during fasting glucose plays an important role.
Keywords: cerebrovascular disease, Bayesian network, logistic regression, decision trees

並列關鍵字

Logistic Regression ； Bayesian Network ； Cerebrovascular Disease ； Decision Trees

參考文獻

張永昌 (民99)。醫學資料分析之研究。中興大學資訊科學與工程學研究所博士論文。台中市。

曾麗文 (民100)。貝氏分類器於醫學數據資料探勘之研究。成功大學工業與資訊管理研究所碩士論文。台南市。

鄭建興 (民94)。腦中風與頸動脈粥狀硬化的非傳統危險因子與存活分析研究。臺灣大學流行病學研究所博士論文。台北市。

劉建宏 (民101)。奠基於貝氏網路的中醫證型辨識。國立中正大學資訊工程研究所碩士論文, 嘉義縣。

郭丞軒 (民104)。植基於增量技術與決策樹分類系統在失智症篩檢之研究與應用。中興大學資訊科學與工程學研究所碩士論文。台中市。