適用於心電圖分析之訊號處理技術

Abstract

關於人類心臟狀態，心電圖訊號提供重要資訊。心臟的資訊，像是正常或不規則的心律，心跳率及心臟運作方式，都可以用來解釋心臟健康或不健康的情況。一個高精準度和運算效率的自動化心電圖波形分析演算法對於心臟疾病診斷和心臟健康持續監護是有助益的。 一個典型的心電圖波形包含P、Q、R、S、T點，最重要的一個為R點。當找到R點位置，就可以參考相對於R點的位置來決定出P、Q、S、T點。在找到P、Q、R、S、T點後，這些點與點之間的位置、高度、寬度及距離都可以被擷取出來成為基本特徵供分類病徵用途。精準的心臟疾病問題分析，例如心室早期收縮、心房早期收縮和心房顫動等疾病相當依賴精準分析心電圖訊號的方法。 在這篇論文中，我們提出以時域分析為基礎、精準且有效率的演算法來分析心電圖訊號，供診斷心臟疾病及心臟健康持續監護使用。藉由使用很多訊號處理技巧，例如時變式梯度權重函數來移除造成心電圖飄移的訊號，相似哈爾的匹配濾波器、比例變化測試的方法來移除心電圖中類似雜訊的波峰、可調變式門檻技巧來偵測找尋R點，墨西哥帽匹配濾波器來偵測找尋P、Q、S、T點，從心電圖訊號中擷取出高識別度的基本式特徵點、組合式特徵點以利於心臟疾病的分析，規則式分類器採用乘法形式權重函式、比例變化假設性測試法、及基尼系數雙類別群聚分類法，來對心室早期收縮、心房早期收縮和心房顫動心臟等疾病做偵測。 我們所提出的即時偵測演算法可分析雙頻道心電圖訊號，並測試於MIT-BIH ARR、AF、QT與AHA等資料庫中，得到相較於其它已發表方法更好的準確率及更低的錯誤率。藉由我們所提出的訊號處理技巧來分析心電圖訊號，對於心室早期收縮、心房早期收縮和心房顫動心臟等疾病的偵測能得到精準的分析。

The electrocardiogram (ECG) signals provide important information about human heart status. The information of the human heart, such as the normal or irregular heartbeat rhythm, the heartbeat rate, and the working behaviors of heart, can be used to interpret healthy or unhealthy states of heart. An automatic ECG waveform analysis algorithm with high accuracy and efficiency is helpful for cardiac disease diagnosis and health monitoring. A typical heartbeat consists of the dominant points of P, Q, R, S, and T peaks. The most important one is the R-wave peak. When the position of the R-wave peak is found, P, Q, S, and T peaks can be determined according to the relative positions to the R-wave peak. After detecting P, Q, R, S, and T peaks, their locations, heights, widths, and distances are extracted as the basic features for heartbeat classification. The accuracy of cardiac disease problem analysis, such as premature ventricular contraction (VPC), atrial premature contraction (APC), and atrial fibrillation (AF) analysis, significantly depends on whether the features of an ECG signal can be extracted accurately. In the dissertation, we propose a time-domain-based algorithm, which is very effective and efficient, to analyze an ECG signal for heart disease diagnosis and health monitoring. Based on the signal processing techniques of the gradient varying weighting function for baseline subtraction of an ECG signal, the Haar-like matched filter, noise-like peaks removal by the variation ratio test, adaptive thresholds for R-wave peak sifting, and the Mexican-hat matched filter for detection P, Q, S, and T peaks, the intra-heartbeat and inter-heartbeat features can be extracted precisely. Moreover, a rule based weighted classifier with product-form score functions, a ratio variation hypothesis test method, and a two-class cluster splitting method by the Gini index are also applied for VPC heartbeat, APC heartbeat, and AF episode classification. The proposed real-time detection algorithm is tested in the MIT-BIH arrhythmia database, the atrial fibrillation database, the QT database, and the AHA database, which consist of two-lead ECG signals. Simulations show that the proposed algorithm achieves higher sensitivity value (SE), positive prediction rate (+P), detection error rate (DER), and specificity value (SP) than those of other existing algorithms. With the proposed signal processing techniques for ECG signal analysis, the PVC heartbeats, APC heartbeats, and AF episodes can be determined in an accurate way.