近年來全球罹患末期腎臟病(ESRD)的人數明顯增加，值得注意的是在各個年齡層的病患人數都呈現逐年成長的趨勢。血液透析是目前臨床上最多末期腎臟病患選擇的治療方法，透過血液透析機中的滲透膜，將血液中的新陳代謝廢物和雜質去除，再將淨化後的血液輸回體內。在血液透析進行前，會先由醫師透過外科手術在病患手臂建立血管通路，一般而言可分為動靜脈廔管(AVF)、動靜脈人工血管通路(AVG)以及靜脈導管三種類型，而其中最常使用的為AVF。然而廔管在長時間進行血液透析治療下，容易造成血管壁彈性受損進而發生阻塞情形。這是由於廔管反覆穿刺而導致的血管壁增厚，或是血管內部充滿鈣化的血塊導致血壓異常的阻塞現象。當血管阻塞程度相當嚴重時，將導致血流量不足，此時就需要進行廔管重建手術。因此，為了延長廔管使用壽命，避免完全阻塞，病患維持定期的廔管檢查並使用正確的廔管照護方式是相當重要的一環。雙通道PPG是一種非侵入式的同步光測量技術，可用於檢測身體不同部位的脈搏與血管健康狀況。本論文招募了臨床受試者，使用糾錯輸出編碼多類別分類支援向量機技術進行動靜脈廔管的阻塞程度評估。非同步雙通道PPG訊號，是根據左手與右手手指作為時域訊號輸入並計算其差異。雙通道PPG訊號的斜率作為特徵值進行輸入，再透過大間隔最近鄰居演算法(LM-NN)進行AVF阻塞程度之分類。本論文為了改善分類器的準確度，也根據適當的特徵值輸入是否會造成影響進行探討。

Recently, the number of patients with end-stage renal disease (ESRD) prevalence has significantly increased around the world. Within each age group of ESRD sufferers, the trend is increasing every year. Hemodialysis is the most considered treatment used by a patient with ESRD which filters out unused product from the blood using a device called Dialyzer. In the case of hemodialysis treatment, three vascular access models are well known: Arteriovenous Vistula Fistula, Arteriovenous Venous Graft and Cental Venous Catheter respectively. The Arteriovenous Vistula Fistula (AVF) becomes a vascular access model that is most often recommended by a nephrologist for a hemodialysis patient. There are two general conditions that often occur in the treatment of hemodialysis: vascular access stenosis and impaired vessel wall elasticity. Stenosis is a physiological deformation of blood vessels caused by calcification which leads to narrowing vessel or can be defined as vessel wall thickening caused by new material which lead to abnormal blood pressure. Stenosis causes a decrease in blood flow, so the amount of blood required for the hemodialysis process is reduced. In order to avoid total occlusion caused by this fact, regular monitoring is needed to prolong the life of fistula, otherwise re-making the fistula must be done. Bilateral Photoplethysmography (PPG) is a non-invasive synchronous electro-optic measurement technique for detecting the cardiovascular pulse and physiological information about vascular health at different sites of the body. In this thesis, the degree of stenosis (DOS) of AVF in the recruited subjects was evaluated with the technique called error correcting output coding support vector machine one versus rest (ESVM-OVR). The bilateral differences (asynchronous) of PPG signal were measured and calculated from the left and right fingers in time domain as the features input. The other work represented in this thesis is AVF stenosis classification with Levenberg-Marquardt Neural Network (LM-NN) Algorithm where the slope of bilateral PPG shape were used as the input features. In addition to the techniques used, the proper input features ware investigated in this work to improve the classifier performance.