| 摘要: | 隨著臨床醫學的快速進步，醫學的最終目標是希望可以針對不同病人施行個體化的醫療，以取代目前忽略個體異質性的標準化疾病療法。雖然分子層次的生化指標，尤其是經由高通量技術所產生的，已取得一定程度的成功，不但可以提供疾病當前所處的嚴重度、甚至可分類單一疾病的亞群體，但因其需龐大之資源或是技術，多數的生化指標還是難以獲得臨床廣泛的使用，且以此類參數來監測或追蹤疾病的進程也不切實際，因此發展一些方法能探究病人在不同疾病病程時的特徵，補足目前方法之不足，便可能使個體化的醫學能加速達成。生理訊號是由生物體特定的控制系統在維持內在系統恆定時所產生的輸出，分析此種訊號應為一個可用於探究個體在不同疾病及疾病進程中的特徵，並可提供生理或病理機轉上重要的訊息。此外多數的生理訊號，如心電圖、腦波或呼吸訊號，都可以由簡單且非侵入的方式取得，並藉由數位化長久的保存於便宜的儲存裝置中。也因此對於分析此類訊號的需求也就日益增大，但對此類充滿非線性及非穩態特質的訊號(因為要持續與改變的環境互動所造成)，對目前現存的分析方法是一大挑戰。希爾伯特黃轉換為一個創新的方法，其演算法由經驗模態分析及希爾伯特轉換所組成。經驗模態分析為當中最關鍵的步驟，可將訊號解構為多個不同尺度、隨時間變化的內在模態函式，相較於用先驗的基底組成現有的訊號，此方法可以對非線性與非穩態系統產生的訊號在對時間上的頻率及振幅能有更好的表示，進一步，從希爾伯特黃轉換所產生的瞬時頻率，可以提供對分析訊號所內存訊息更佳的解釋，尤其是對多個系統交互相連、且需在不斷改變的環境下調控，並具非線性及非穩態的生理訊號。但當前只有非常少數的研究應用非線性與具適應性的訊號分析的概念，因此本研究的目的在於應用希爾伯特黃轉換於多種不同的生物醫學訊號相關主題，包括：(1)用心率變化於探索老化及疾病變化的動態特徵、(2)發展由訊號本身特質所萃取的去趨勢演算法，用於協助易受趨勢影響之非線性分析法、(3)由自動體外電擊器中具高度非穩態心電圖提取主要的特質。此研究發現，由希爾伯特黃轉換可以在分析生理訊號中的相對應機轉能有更好的結果，且訊號中動態的訊息可能提供一個替代性評估老化或是一些特定疾病的指標，同時，希爾伯特黃轉換也可應用為一個可適性的濾波法，不但可留取訊號中重要的特徵或是去除訊號中的趨勢，且取出的訊號，除可提供一些臨床重要議題更寶貴的資訊外，也可以加強一些既存非線性方法的敏感度。As rapid development in clinical medicine, the ultimate goal of clinicians turns to tailor an individualized treatment for each patient rather than using standardized therapy which neglects the heterogeneity of subjects. Although molecular biomarkers, especially generated from high-throughput screening technologies, have achieved a certain degree of success in providing useful information about disease status or finding the subgroups of patients with same diseases, most of those biomarkers are still difficult to gain widespread acceptance because of the resource-consuming analysis and procedure. Furthermore, it is impractical for monitoring or following up the changes of disease processes. Complementary ways to explore the characteristics of patients in different statuses may expedite the progress in achieving the goal.Analysis of physiological signals, the outputs of specific control systems of biological entity in order to maintain internal homeostasis under the environments, should be a promising way to probe the characteristics of the subjects during normal and diseased conditions and may provide crucial information of the underlying physiological and pathological mechanisms. Moreover, most of the physiological signals, such as electrocardiogram (ECG), electro-encephalogram (EEG), and respiratory signal, can be non-invasively and easily acquired and the digitalized signals can be stored for a long time by in-expensive data storage devices. The demands of approaches to analyze those signals inevitably become stronger by clinical practice. However, the major challenge of contemporary methods is the daunting nonlinearity and nonstationarity of those physiological signals which continuously interact with the varying environment.An innovated method, Hilbert-Huang Transform (HHT), consists of Empirical Mode Decomposition (EMD) and Hilbert Spectral Analysis. EMD, the key step of HHT, can adaptively decompose the signal into many different intrinsic mode functions (IMFs) operated in different time-scales. Instead of integral of a priori basis to reconstruct the signal, it gives better time-frequency-energy representation for the nonlinear and nonstationary systems. Moreover, the instantaneous frequency derived from HHT can provide a better route to analyze the underlying information of the signal, especially for physiological systems that are nonlinear by nature and interconnect with other systems under the perturbation of the ever-changing environments. However, only limited studies adopt the concept of nonlinear and adaptive data analysis. The aims of the study are, therefore, trying to apply HHT in several biomedical topics including (1) probing the dynamical characteristics of aging processes or pathological changes of heart rate dynamics; (2) developing a data-driven detrending method to assist the existed nonlinear analysis that is vulnerable to nonstationary trends; (3) extracting the main features of the highly nonstationary ECG signals from automated extracorporeal defibrillator.It is suggested that the underlying mechanisms of physiological signals are better described by HHT and dynamical information of the signals may serve as alternative biomarkers for aging or some diseases. Moreover, HHT can also be applied as adaptive filter to extract the important features or eliminate the unwanted trends of the signals. The elicited components of the signals can provide more crucial information of clinical relevant issues and the sensitivity of the existed nonlinear analysis for the reconstructed signals can be improved. |
