基於毫米波雷達偵測脈搏運動之血壓預測;Blood Pressure Prediction Based on Pulse Motion Detected by Millimeter-Wave Radar

NCU Institutional Repository > 資訊電機學院 > 資訊工程研究所 > 博碩士論文 > Item 987654321/98631

請使用永久網址來引用或連結此文件: https://ir.lib.ncu.edu.tw/handle/987654321/98631

題名:	基於毫米波雷達偵測脈搏運動之血壓預測;Blood Pressure Prediction Based on Pulse Motion Detected by Millimeter-Wave Radar
作者:	廖聿程;Liao, Yu-Cheng
貢獻者:	資訊工程學系
關鍵詞:	毫米波雷達;脈搏;Millimeter-Wave Radar;Pulse
日期:	2025-08-28
上傳時間:	2025-10-17 13:01:44 (UTC+8)
出版者:	國立中央大學
摘要:	本研究提出一套基於毫米波 FMCW 雷達的非接觸式血壓預測系統，分別針對胸部與手腕進行量測，以擷取脈搏引起的皮膚微小位移訊號，進而預測收縮壓與舒張壓。此系統無需傳統充氣式袖帶，具備舒適、連續且低干擾的監測優勢，特別適用於居家照護、高齡者與慢性病患的長期健康監控需求。為提升訊號穩定性並抑制來自呼吸與身體動作的偽訊號干擾，本系統首先對每幀毫米波雷達回波進行脈衝雜訊去除，並搭配帶通濾波器初步擷取與脈搏相關的有效訊號。隨後擷取長度為 10 秒的訊號，並再度套用帶通濾波器以進一步強化脈搏成分。從處理後的訊號中擷取七項脈搏波形特徵作為模型輸入，包括：最大峰值、第一峰值、最大與最小峰值之比、最大與第一峰值之比、峰對峰平均間隔、期望值與變異數。最終，透過隨機森林回歸模型進行血壓預測，並以 leave-one-out 交叉驗證（LOOCV）驗證模型效能。本研究共招募 6 位受試者進行實驗分析。結果顯示，胸部訊號預測收縮壓與舒張壓之平均誤差均為 -0.10 mmHg，標準差分別為 4.54 mmHg 5.07 mmHg；手腕訊號預測收縮壓與舒張壓之平均誤差則均為 -0.13 mmHg，標準差分別為 6.33 mmHg 與 5.78 mmHg。上述結果皆符合美國食品藥物管理局（FDA）所採用之 AAMI 標準，驗證本系統具備應用於非接觸式血壓估測之可行性與潛力。 ;This study proposes a non-contact blood pressure estimation system based on millimeter-wave frequency-modulated continuous-wave (FMCW) radar. Measurements are conducted at both the chest and wrist to capture micro skin displacements induced by arterial pulsation, which are then used to estimate systolic blood pressure (SBP) and diastolic blood pressure (DBP). Without the need for traditional inflatable cuffs, the system offers a comfortable, continuous, and low-interference monitoring solution, particularly suitable for long-term health management in home care, elderly individuals, and patients with chronic conditions. To enhance signal stability and suppress artifacts caused by respiration and body movements, each radar frame undergoes impulsive noise removal, followed by bandpass filtering to preliminarily extract pulse-related signal. A 10-second segment of radar data is then selected and further processed with another bandpass filter to enhance pulse components. From the filtered signal, seven pulse waveform features are extracted as model inputs, including: maximum peak (MP), first inflection peak (FIP), maximum to minimum ratio (MMR), maximum to inflection ratio (MIR), peak-to-peak interval (PPI), expectation, and variance. A random forest regression model is used for blood pressure prediction, and model performance is evaluated using leave-one-out cross-validation (LOOCV). A total of six participants were recruited for the experimental analysis. Results show that predictions based on chest signals yielded a mean error of -0.10 mmHg for both SBP and DBP, with standard deviations of 4.54 mmHg and 5.07 mmHg, respectively. Predictions from wrist signals resulted in a mean error of -0.13 mmHg for both SBP and DBP, with standard deviations of 6.33 mmHg and 5.78 mmHg, respectively. These outcomes comply with the Association for the Advancement of Medical Instrumentation (AAMI) standards adopted by the U.S. Food and Drug Administration (FDA), validating the feasibility and potential of the proposed system for non-contact blood pressure estimation.
顯示於類別:	[資訊工程研究所] 博碩士論文

文件中的檔案:

檔案	描述	大小	格式	瀏覽次數
index.html		0Kb	HTML	8	檢視/開啟

在NCUIR中所有的資料項目都受到原著作權保護.

社群 sharing

資料載入中.....