人體動脈硬化程度，直接關係著許多與心血管有關的慢性病的發生，脈波傳導速率是目前用以評估動脈硬化的重要指標之ㄧ，然而現今量測脈波傳導速率需在醫院量測，醫院所採用的非侵入式動脈硬化檢測，大都是以壓脈帶方式量測脈波傳導速率，其計算方式是以量測之間的距離與時間差下去評估，如果在家就能隨時量測，可增加便利性並且有助於病情的診斷。目前光容積波訊號的形態已得到廣泛研究，可以由光容積波訊號獲取與血液動力學狀態高度相關的各種特徵，加上最近穿戴式設備的快速發展，透過泛用性的生理訊號量測來得知血管變化的方式逐漸引起了人們的廣泛關注。 本論文使用心電訊號和光容積波訊號來估測脈波傳導速率。從穿戴式裝置量測手腕和手指心電訊號和光容積波訊號，對手腕和手指做同步和前處理，將量測產生的干擾或不良訊號移除。對光容積波訊號取特徵，針對特徵丟失和歧義性的問題，提出了解決方法來處理，利用三階微分和二階微分光容積脈衝波的極值來識別二階微分和一階微分的光容積脈衝波波形中缺失或模糊的特徵，可使特徵萃取達98.6%以上的比率。另外，針對光體積描記信號的加權脈衝分解分析，採用五個高斯波解析，並以加權最小平方法準則進行優化，我們將權重施加於被認為可提供血管年齡和血管僵硬度的脈搏段，另外適當設定高斯參數的邊界約束，從結果可以看出加權使分解穩定，高斯參數的平均方差減小。最後使用手腕特徵估測脈波傳導速率，採用極限梯度提升(XGBoost)的演算法，女性均方根誤差最好的結果可以達到177.70(cm/s)，男性均方根誤差最好的結果約為144.39(cm/s)。 ;The degree of arteriosclerosis is directly related to the occurrence of many chronic diseases. Currently, pulse wave velocity is one essential metricused to evaluate arteriosclerosis. However, the pulse wave velocity is usually measured in the hospital nowadays. The non-invasive method is mostly based on the pressure belt, and calculates their distance divided by the time difference. At present, morphology of the photoplethysmography signals has been widely studied so as to acquire various features that are highly related with hemodynamic states. Recently, the rapid development of wearable devices makes it possible to realize vascular conditions. This change has gradually aroused widespread concern. In this thesis, we use wrist electrocardiogram (ECG) signals and photoplethysmography (PPG) signals to estimate pulse wave velocity. The ECG and PPG of wrist and finger are measured from wearable devices. These signals are first synchronized and pre-processed to remove interference and bad-quality signals generated during the measurement. Considering the problems of missing and ambiguous features, we propose imputation and resolution techniques. The extrema of the third-order derivative and second-order derivative PPG waveforms are employed to identify the missing or ambiguous features of the second-order derivative and first-order derivative PPG waveforms. We propose weighted pulse decomposition analysis that emphasize the PPG pulse portion corresponding to point a to point f of the second-order derivative PPG. Five Gaussian waves are used for decomposition In addition, the boundary constraints for Gaussian wave parameters are properly set. From the results, we can see that weighting makes the decomposition stable and the variances of Gaussian parameters reduced. Finally, the wrist PPG and ECG features are employed to estimate the brachial ankle pulse wave velocity. The root mean square errors (RMSE) of female estimation results is 177.70(cm/s), and of male estimation results is about 144.39(cm/s), which outperform the result from the conventional multiple linear regression.
