應用類神經網路於冠狀動脈支架置放後再狹窄的預測

Application of Artificial Neural Network Model for the Predication of Coronary In-Stent Restenosis

楊朝宏(Chao-Hung Yang)

指導教授 ： 徐建業

臺北醫學大學/醫學科技學院/醫學資訊研究所/碩士(2011年)

摘要

心臟疾病一直來都是十大死因的前幾名，而冠狀動脈心臟病（Coronary Artery Disease；CAD）佔了大多數。在臨床上目前皆是利用經皮冠狀動脈氣球擴張術來治療冠狀動脈心臟病，為了增加早期癒後利用冠狀動脈支架（Coronary Stent）來減少因血管彈性所造成的早期再狹窄（Restenosis）。但放置完冠狀動脈支架後血管細胞會進行修復，產生平滑肌細胞的增生，進而產生冠狀動脈內支架再狹窄（In-Stent Restenosis；ISR）。現階段診斷是否產生ISR皆是當病人產生不舒服就醫時，透過侵入性或非侵入性的檢查來診斷。非侵入性的檢查有核子醫學檢查、心電圖檢查、運動心電圖檢查、心臟超音波。侵入性的檢查有血管內超音波及最終診斷的冠狀動脈攝影。

本研究的目的在利用多層感知器的類神經網路建構類神經模組，用以偵測是否產生ISR，來判別是否進行冠狀動脈攝影。以及冠狀動脈心臟病治療的選擇。

本研究資料來源是回溯性收集汐止國泰醫院2006年10月至2010年6月心導管室接受冠狀動脈支架置放226名病人，共置放了475冠狀動脈支架資料；且於6個月後有再進行冠狀動脈攝影的病人。將資料依約三比一分為訓練組及測試組。進行比較類神經網路模組與邏輯斯回歸的結果。

在模型建構中，分成組合A、組合B、組合C、組合D、組合E等5種不同預測變數組合進行模型建構，選擇組合A利用所有預測變數組合所訓練出來的結果較佳，其類神經網路模組的AUROC為0.895、敏感度及特異性分別為92.2％％、80.6％、正確率為0.87。而邏輯斯迴歸在組合A的AUROC為0.663、敏感度及特異性分別為49.4％、75.8％、正確率為0.612。

關鍵字

冠狀動脈支架 ； 冠狀動脈支架再狹窄 ； 類神經網路

並列摘要

Heart disease is always the leading cause of mortality, especial in coronary artery disease. In clinical practice, it usual fix by balloon angioplasty. In order to improve immediate result and early restenosis, coronary stent was used. However, endothelial cell repair and smooth muscle proliferation induced in-stent restenosis. The present diagnosis of ISR are by clinical symptoms, non-invasive or invasive examination. Non-invasive examinations included Thallium re-perfusion scan, ECG, treadmill exercise test, and echocardiogram. Invasive procedures are intravascular ultrasound with confirmed by coronary angiography.

In our study, the data was collected from Oct. 2006 to June 2010 at Sijhih Cathay General Hospital. The population is the patients who had coronary artery stent deployment with coronary angiography follow up six months later. We collection 226 patients have 475 coronary stents.The data was separated as training and test group then analyzed by Artificial Neural Network (ANN) and logistic Regression Model.

In this modeling structure, 5 groups are divided based on different predicting variables and they are designated from combination A to E. The results from combination is A is the best in this modeling structure and the AUROC for this neural network module is 0.895, and sensitivity, specificity, and accuracy are 92.2% , 80.6%, 0.87 respectively. Logistic regression for the AUROC of combination A is 0.663, and the sensitivity, specificity, and accuracy are 49.4%, 75.8%, and 0.612 respectively.

並列關鍵字

Stent ； ISR ； Artificial Neural Network

參考文獻

行政院衛生署 (2009)。98年度死因統計完整統計表。

行政院衛生署國民健康局 (2003)。健康心生活:冠心病防治宣導手冊。台中市。

徐敏耀, 劉., 馬作鏹等人 (2007)。冠狀動脈心臟病危險因子之老年人心導管檢查預測模型研究 台灣老年醫學雜誌. 第3卷。

Baxt, W. G. (1995). "Application of artificial neural networks to clinical medicine." Lancet 346(8983): 1135-1138.

Dotter, C. T. and M. P. Judkins (1964). "Transluminal Treatment of Arteriosclerotic Obstruction. Description of a New Technic and a Preliminary Report of Its Application." Circulation 30: 654-670.