基於希爾伯特黃轉換之非穩態醫學訊號之特徵擷取與量化評估法-以心血管循環系統為例

Abstract

心血管疾病在全球死亡率位居高位，且於台灣現居前十大死因其第二名，心血管系統的疾病可及早透過檢測的方式被診斷，現今對其監測與評估越來越受到重視，加上無線傳輸技術的快速發展、分析晶片微小化、感測技術輕便化，心血管照護系統之開發所需技術越來越趨於成熟。 既有的非侵入式心血管照護系統中，深入研發且趨於成熟的主要是心電圖的分析，可用於心功能的監測。目前已開發至能藉由特殊的晶片或連至手機進行隨身監測，並提供心功能與心臟疾病相關之參考指標，但對於血液脈波的分析，尤其心血管循環系統之調節與血管相關特性評估的研究卻很少。現有的心血管調節評估在頻譜之頻域的分析上，主要採用心律變異度對應之自律神經活性的頻譜進行量化；而現有之血管特性的評估方法則是透過血液脈波之波形計算。心率變異度的頻譜分析主要採用傅利葉轉換(以弦波為基本組成的前提假設)或是小波轉換(以各種小波方程式為基本組成的前提假設)來進行分析與量化評估，但結果受限於各轉換的前提假設，無法完整呈現醫學訊號非穩態的特性變化。 近年，黃鍔院士提出一個擁有高自適性的分析法，稱為希爾伯特黃轉換，此方法有別於傳統的轉換分析法，對非穩態的訊號能在無失真的前提下，將訊號內生性的特徵進行不同時間尺度的拆解，其中，拆解是透過經驗模態拆解法。此拆解法未有數學模型的前提假設，取而代之，透過波前與載波的物理理論，以訊號或資料本身的變化做為內生性特徵擷取的基礎，因此對不同資料的適應性極佳，此方法對非穩態的資料分析有相當重大的突破，目前已廣泛應用於各個領域，尤其是醫學分析應用。 本研究針對血液脈波的訊號，採用希爾伯特黃轉換中的經驗模態拆解法，將血液脈波的非穩態變化進行內生特徵的定性與擷取，再對應各不同時間尺度下的內生特徵，與相對應的臨床指標進行交叉驗證與深入探討。 本研究針對血壓訊號的波形與短期趨勢，分別進行擷取與更進一步的驗證與探討。波形分析的結果顯示，內生特徵的擷取可強化血壓反彈波的變化，並得到與傳統血壓波形分析一致的結果；而脈波透過正規化希爾伯特轉換，將脈波的內生性特徵對應到正交之複數平面上，並得到訊號的瞬時頻率與瞬時脈率，臨床驗證顯示，瞬時脈率可輔助呼吸與血壓調節狀況的評估；在短時趨勢分析中，血壓訊號經由經驗模態拆解法，根據不同內生特徵變化，被拆解為若干個短時趨勢，分別對應到不同的生理調節頻帶上，經由與臨床血壓變異度的分析與驗證，發現此方法能在頻譜上提供自適性頻帶的量化評估，對個體差異下之個人生理調節頻帶的探討提供額外的切入點。整體上，透過血壓訊號的分析與臨床驗證，血液脈波結合希爾伯特黃轉換的技術，能進一步計算出可適用於心血管系統調節的評估指標，並在非侵入式的心血管照護上，具有很大的潛力。 本研究同時以此分析法為核心，透過可程式化之邏輯電路板實作出心血管調節照護的系統雛形，包含血液脈波感測、分析電路、網路傳輸介面、遠端伺服器，建構出此分析法之應用實境範例。未來的延續研究方面，瞬時脈率的變化，其隱含的資訊相當複雜，包含血管的特性、週邊組織的響應、自律神經調節、心功能、…等，目前已有不少非線性分析法應用於心率變異度之分析(例:去趨勢波動分析、多尺度熵)，且於心血管疾病之評估上有顯著的參考性，瞬時脈率所計算出的量化指標與各心血管疾病的關聯性(死亡風險評估)需更進一步的探討。而自適性頻帶的評估在個人生理調節狀態的表現，也需設計對應的臨床試驗進行更進一步的驗證。最後，在照護應用方面，由於目前為了對應臨床的驗證，採用的訊號源為連續時間之血壓訊號，此訊號目前僅能在臨床端取得，若要拓展對應至方便照護使用的血液容積感測或血氧脈波訊號，則需做同步量測與各種不同生理狀態下的交互驗證，若能對應至適合隨身量測之血液脈波量測技術，在心血管調節照護系統與脈診定性分析的研發將有相當大的幫助。

In 2008, world health organization estimated that there are 17.3 million people died from cardiovascular diseases (CVDs) and CVDs is one of the ten leading causes of death in Taiwan. CVDs is preventable compared to cancers and can be detected by cardiovascular monitoring. The health care in cardiovascular circulation becomes important now a day. The portable healthcare device becomes mature owing to the developments of several techniques, including wireless data transfer, system on chip, and wearable sensor network. The requirement of health care device becomes huge owing to aging population. Recent non-invasive cardiovascular monitoring system focuses on the development of electrocardiogram, including the specific process chip and the evaluation index of cardiac function, but the research of cardiovascular auto-regulation is relatively rare. Traditional studies investigated that the auto-regulation could be assessed by heart rate variability (HRV). Most of the HRV studies adopted Fourier transform or wavelet transform as spectral analysis method. These methods have good properties to extract and present the characteristics of stationary signal based on their own mathematical fundamental basis, such as sinusoid function or wavelet function, but the characteristics of medical signal are usually non-stationary. Recently, N.E. Huang proposed a novel adaptive method, called Hilbert-Huang transform (HHT). HHT has good capability for non-stationary characterization without information loss and extracts the intrinsic features in multiple scales. The extraction method of HHT, called empirical mode decomposition (EMD), extracts the intrinsic features of signal based on signal's fluctuations, such as the envelop of signal, and is adaptive for different target signals. The aim of this research is to 1) extract the intrinsic features of blood pulse signal by HHT; 2) quantitatively assess the non-stationary features in multiple time scales; 3) examine the usefulness of the assessment in clinic. The results showed that the blood pulse signal could be decomposed into four different intrinsic features in four physiological time scales, including noise, pulse wave morphology, short-term trend, long-term trend. By analysis of arterial blood pressure (ABP), the reflection wave could be enhanced by EMD and the results were consistent with traditional studies. The pulse wave could be extracted from ABP by EMD, and instantaneous pulse rate (iPR) was estimated by normalized Hilbert transform. The results showed that the iPR presents the characteristics of respiration and cardiovascular auto-regulation. In short-term ABP trend study, the individual ABP regulation was extracted adaptively by EMD. This method helps for the exploration of individual optimal frequency band in auto-regulation assessment. This research also designed and implemented the prototype of cardiovascular auto-regulation monitoring system based on embedded system development and network programming. The progress of this research contains several parts. Currently, there are several non-linear approach for cardiovascular auto-regulation analysis, such as detrend fluctuation analysis and multiscale entropy. Though, the iPR can be used as the estimator of cardiovascular auto-regulation function, the relationship between iPR and CVDs needs further investigation. Besides, the ABP signal is hard to get in daily life, the replacement of ABP signal by blood pulse sensing needs further examination and validation.