基於階層迴歸方法運用光體積描記訊號與心電圖估測血壓與脈波傳導速度;Estimation of Blood Pressure and Pulse Wave Velocity from Photoplethysmography and Electrocardiography Using Hierarchy Regression

NCUIR > College of Electrical Engineering & Computer Science > Graduate Institute of Electrical Engineering > Electronic Thesis & Dissertation > Item 987654321/89979

Please use this identifier to cite or link to this item: http://ir.lib.ncu.edu.tw/handle/987654321/89979

Title:	基於階層迴歸方法運用光體積描記訊號與心電圖估測血壓與脈波傳導速度;Estimation of Blood Pressure and Pulse Wave Velocity from Photoplethysmography and Electrocardiography Using Hierarchy Regression
Authors:	李御銓;Li, Yu-Chuan
Contributors:	電機工程學系
Keywords:	血壓估測;脈波傳導速率估測;心電圖;光體積描記訊號;Blood Pressure Estimation;Pulse Wave Velocity Estimation;Electrocardiography;Photoplethysmography
Date:	2022-08-10
Issue Date:	2022-10-04 12:06:26 (UTC+8)
Publisher:	國立中央大學
Abstract:	血壓以及脈波傳導速率為心血管疾病重要的指標，血壓量測上多仰賴脈壓式血壓計，而脈波傳導速率則是需要在醫院量測，大多儀器是以壓脈帶方式量測。近年來，利用穿戴式裝置量測生理訊號，推算血壓與脈波傳導速率技術逐漸成熟，具有可長時間監測的優點以及對遠距醫療有所幫助。本論文利用心電訊號及光體積描記訊號來估測血壓和脈波傳導速率。從裝置取得手指及手腕心電訊號與光體積描記訊號後，對波型進行前處理，去除60Hz市電雜訊及基線飄移，並對訊號不良的波型進行初步的過濾，透過特徵萃取程序從心電訊號與光體積描記訊號取得特徵，並對光體積描記訊號進行波型拆解，將波型拆解成5個高斯成份波，並取其振幅、寬度與中心位置當作特徵。在資料處理的階段，篩選可用的特徵與波型段，並對不同血壓及脈波傳導速率的區間補足，以維持訓練資料集區間的平衡。而基於不同區間的血壓或是脈波傳導速率所看重的特徵不一定相同的觀察，因此提出了分層迴歸的方法，讓估測結果可以更加準確。使用手腕訊號特徵估測脈波傳導速率，演算法使用極限梯度提升(XGBoost)並直接做全域迴歸，女性均方根誤差為183.73(cm/s)，男性均方根誤差為188.80(cm/s)，如果採用先分類再迴歸，可以使得女性均方根誤差進步到149.72(cm/s)，男性均方根誤差進步到160.15(cm/s)。而手指訊號特徵估測血壓，參考了國際量測血壓標準，以符合區間分布數量要求的85人255筆量測資料作為測試集。使用神經網路做全域迴歸，收縮壓均方根誤差為10.71(mmHg)，若使用分層迴歸的技術，收縮壓均方根誤差可達到9.79(mmHg)。;Blood pressure and pulse wave velocity are important indicators of cardiovascular diseases. Blood pressure measurement mostly relies on the mercury sphygmomanometer and pulse wave velocity usually needs to be measured in hospitals. Most instruments measure pulse wave velocity by means of pressure pulse band. In recent years, the technology of measuring physiological signals by using wearable devices becomes popular, which has the advantages of long-term monitoring and is helpful for remote medical treatment. In this thesis, Electrocardiography (ECG) signals and photoplethysmography (PPG) signals are used to estimate blood pressure and pulse wave velocity, and the feature properties of blood pressure or pulse wave velocity in different zones may not be the same, so a hierarchical regression method is proposed to make the estimation results more accurate. The PPG signal from wrist was used to estimate the pulse wave velocity. We used eXtreme Gradient Boosting (XGBoost) for global regression. The root mean square estimation error (RMSE) for female and male was 183.73(cm/s), and 188.80(cm/s). The root mean square error can be improved to 149.72(cm/s) for female and 160.15(cm/s) for male if we applied classification before regression. The PPG and ECG signals from finger were used to estimate blood pressure according to the international standard of blood pressure measurement. The test set were composed of 255 measurements from 85 people which met the requirements of BP distribution. The root mean square error of systolic blood pressure was 10.71(mmHg) when neural network was adopted for global regression. The root mean square error of systolic blood pressure was 9.79(mmHg) when residual neural network-based hierarchical regression was used.
Appears in Collections:	[Graduate Institute of Electrical Engineering] Electronic Thesis & Dissertation

Files in This Item:

File	Description	Size	Format
index.html		0Kb	HTML	58	View/Open

社群 sharing

Loading...