實踐經驗模態分解於高度非穩態生理訊號之訊號特徵擷取;Harnessing Empirical Mode Decomposition on Feature Extraction of Highly Nonstationary Biomedical Signals

NCUIR > college of Health Sciences and Technology > Institute of Biomedical Engineering > Electronic Thesis & Dissertation > Item 987654321/85757

Please use this identifier to cite or link to this item: http://ir.lib.ncu.edu.tw/handle/987654321/85757

Title:	實踐經驗模態分解於高度非穩態生理訊號之訊號特徵擷取;Harnessing Empirical Mode Decomposition on Feature Extraction of Highly Nonstationary Biomedical Signals
Authors:	王子文;Wang, Zih-Wen
Contributors:	生物醫學工程研究所
Keywords:	經驗模態分解;峰值檢測;自適應性;晝夜節律;睡眠分析;Empirical Mode Decomposition (EMD);beat detection;adaptive;circadian rhythm;sleep analysis
Date:	2021-07-23
Issue Date:	2021-12-07 11:21:45 (UTC+8)
Publisher:	國立中央大學
Abstract:	隨著科技的進步與蓬勃發展，穿戴式裝置、微型化裝置以及智慧型手機已然成為普及世界的存在，藉由這些便攜型的非侵入式與智慧型裝置來長期量測生物醫學訊號是未來醫療領域的發展主軸之一，除有利於長期監測及個人化醫療，對於遠距醫療與居家檢測的助益也十分可觀；有鑑於日常生活中擷取的生理訊號具有非穩態與非線性特質，因此對該類訊號進行分析甚至是特徵擷取與指標量化計算有其難度，本研究目的在於應用經驗模態分解於數種生物醫學訊號相關主題，包含：(1)發展普適性峰值檢測演算法並應用於心電圖、光體積變化描記圖、動脈血壓波以及遠端光體積變化描記圖與 (2) 應用日常手機使用量於作息節律監控與睡眠分析。經驗模態分解是一種多尺度、自適應性的訊號處理技巧，由於該方法的基底函數取決於數據本身的波動特性，所以可保留訊號的局部特性，因此被廣泛應用於非穩態與非線性訊號解析；透過經驗模態分解，訊號可被解構成數個在不同時間尺度下的本質模態函數，而每一個本質模態函數都具有不同的物理意義，針對不同主題的研究而言，則可進一步篩選出特定的本質模態函數以利後續分析。傳統上既有的峰值檢測演算法大多奠基於設定自適應性閾值並結合繁複的篩選條件，而現今以機器學習為基底的演算法會受到訓練資料集多樣性與跨資料庫問題；此外，現有演算法中仍無可適用於各類生理訊號峰值檢測的統一框架，因此本研究回歸生理訊號的類週期性本質，將此特質與經驗模態分解結合以發展出能夠應用於多種生理訊號的檢測演算法。而手機使用行為分析則是以手機為媒介紀錄日常手機使用行為，並量化該紀錄成數位訊號，透過經驗模態分解來萃取出隱藏在資料中的晝夜節律與行為模式，從中推算出睡眠相關指標。本方法克服了以往訊號特徵擷取時會遇到的斷點問題，在提取主要成分的過程中不須仰賴繁複的閾值設定與條件式，而是自適應性的決定濾波器頻帶並擷取出訊號的主要成分，此方法除可去蕪存菁的萃取出特定頻帶的成分外，也比現有方法更能完整呈現該成分在頻率與振幅的非穩態特性，此獨特性使本研究中的兩個主題在臨床上的應用價值與助益大幅增加。 ;With the global prevalence of the wearable and portable devices in recent years, we can achieve long-term monitoring, telemedicine or even individualized treatment with the assistance of cloud computing. Nevertheless, the biomedical signals gathered from these wearable and portable gadgets are prone to be contaminated by various kinds of perturbation, such as motion artifacts, intermittency, baseline wandering or white noise, thereby causing further analysis, including feature extraction and finding indicators, to be more complicated. In this study, we develop a distinctive method based on the quasi-periodicity and Empirical Mode Decomposition (EMD)-based method, Uniform Phase EMD (UPEMD), to adopt the nonlinearity and adaptiveness on biomedical signal analysis. This study aims at applying our proposed method in two biomedical topics, the first one is to develop a universal beat detection algorithm that is suitable for different types of biomedical signals, such as electrocardiography (ECG), photoplethysmography (PPG), arterial blood pressure (ABP) and remote photoplethysmography (Remote PPG). The second topic is harnessing our proposed technique on the smartphone usage records to estimate the sleep-wake cycles as well as the sleep parameters. In view of the non-stationary property in the physiological signals, EMD tends to be a satisfactory choice that can accommodate the unpredicted variations. The rationale of EMD is to segregate the signal into a finite number of Intrinsic Mode Function (IMF) through the iterative sifting processes, meanwhile considering the time scale characteristics of the data. To put it bluntly, EMD serves as a data-driven technique rather than relying on a priori basis. Our proposed method not merely avoid the cumbersome and rigorous statements and threshold settings of the existing algorithms but also automatically and adaptively decide the filter bank to extract the specific component, meanwhile retaining the nonstationary and nonlinear properties of the physiological signal without being interfered by the intermittency.
Appears in Collections:	[Institute of Biomedical Engineering] Electronic Thesis & Dissertation

Files in This Item:

File	Description	Size	Format
index.html		0Kb	HTML	127	View/Open

社群 sharing

Loading...