基於心率變異性之高階統計與非線性方法辨識充血性心臟衰竭的研究

High Order Statistics and Nonlinear Methods for Congestive Heart Failure Recognition Based on Heart Rate Variability

李銘元(Lee Ming-Yuan)

指導教授 ： 余松年

國立中正大學/工學院/電機工程學系/博士(2012年)

摘要

充血性心臟衰竭 （congestive heart failure, CHF）是一種常見而又嚴重的心臟疾病。然而，充血性心臟衰竭在早期階段容易被忽略，需藉輔助醫療服務系統 (Auxiliary medical service system, AMSS)，協助診斷。對於檢測心臟的電性活動，心電圖是一種低成本與便利的方式，可協助識別早期充血性心臟衰竭的病患。
心率變異性 (Heart rate variability, HRV) 是一種基於心電圖來分析心臟節律的方法。近年來，幾個研究報告顯示，利用HRV已經成功分辨正常竇房律動與充血性心臟衰竭。但這些方法的辨識率仍有待改進地方。
在這篇論文中我們首先提出了簡單而有效異位點與趨勢去除方式如同預處理。此預處理方法被證明在頻率領域特徵，其改善了充血性心臟衰竭分類上的效能。基於此預處理，我們探討了幾個利用HRV改善CHF辨識率的方法。首先，雙頻譜 (bispectral) 特徵的應用提升了敏感率 (sensitivity) 、確認率 (specificity) 與正確率 (accuracy)。相反的，當加入越多種類的特徵時，其計算量也隨之增加。為了解決此問題，基於互訊息理論，我們提出了一個特徵選取器(mutual information) 來降低特徵集合的維度。而所提的特徵選取器被驗證能提升分類效能並有意義的降低特徵維度。
此外，另一令人感興趣的研究，應用多尺度 (multiscale) 分析方法在非線性與渾沌HRV特徵對於CHF的辨識。一個新多尺度方法基於離散小波轉換 (discrete wavelet transform, DWT)，探討非線性與渾沌 (chaotic) 特徵之特性。 同時，我們所提出方法拿來比較其他傳統多尺度分析方法與分類CHF文獻，其結果顯示其方法勝過其他文獻方法。最後，我們整合了雙頻譜、渾沌、非線性特徵與互訊息特徵選取器為一個系統。其整合系統的辨識CHF能力達到不錯的效果。

關鍵字：充血性心臟衰竭、心率變異性、異位點、去趨勢、雙頻譜、互訊息理論、混沌理論、多尺度分析。

關鍵字

充血性心臟衰竭 ； 心率變異性

並列摘要

Congestive heart failure (CHF) is a common yet serious heart disease. However, in the early stage of CHF, the symptoms can be easily ignored and might need auxiliary medical service system (AMSS) to assist in diagnosis. Electrocardiography (ECG) is a low cost and convenient means for detecting the electrical activity of the heart and may assist the patients identify CHF in its earlier stages.
Heart rate variation (HRV) is an approach to analyze heart rhythm based on ECG. In recent years, several studies have reported success of using HRV for recognizing CHF from normal sinus rhythm (NSR). However, the recognition rates of these methods still leave room for improvement. In this thesis, we first propose a simple and effective ectopic and trend removal method as preprocessor. This method is demonstrated to improve the performance of CHF classifiers with frequency-domain features. Based on this preprocessor, we investigate several methods to improve the performance of CHF recognition based on HRV. Firstly, the application of bispectral features raises the recognition rates of sensitivity (SEN), specificity (SPE), and accuracy (ACC). However, the inclusion of multi-category features also raises computation load. To tackle this problem, we, secondly, propose a feature selector base on mutual information to reduce the dimension of the feature set. The proposed feature selector is demonstrated to enhance the performance of the classifier with significantly reduced feature dimension.
Besides, we explore another interesting issue of applying multiscale analysis method to nonlinear and chaotic features for characterizing HRV for CHF recognition. A novel multiscale method based on discrete wavelet transform (DWT) is proposed to uncover the properties of chaotic and nonlinear features in the multiscale domain. The performance of the proposed method is compared to that of traditional multiscale analysis method and other CHF classifiers in the literature. The results demonstrate the superiority of the proposed method over other methods. Finally, we integrate the three parts of the study, including bispectral features, multiscale chaotic and nonlinear features, and feature selector based on mutual information, into a system. Impressive discrimination power of the integrated system in CHF recognition is observed.

Keywords: Congestive heart failure, heart rate variation, ectopic, detrend, bispectral, mutual information, chaotic, multiscale.

並列關鍵字

Congestive Heart Failure ； Heart Rate Variation

參考文獻

[56] N. Kwak and C.-H. Choi, Input feature selection for classification problems, IEEE Transaction Neural Network, 5 (4) (2002) 143-159.

[2] N. Guo, Z. Lu, X. Xue, J. Shu, S. Liu, Assessment of autonomic function in patients with acute myocardial infarction or diabetes mellitus by heart rate variability, ventricular late potential and QT dispersion, Hypertension Research: the Official Journal of the Japanese Society of Hypertension, (23) (2000) 4 367-370.

[3] J. T. Bigger, J. L. Fleiss, R. C. Steinmann, L. M. Rolnitzky, W. J. Schneider, P. K. Stein, RR variability in healthy, middle aged persons compared with patients with chronic coronary heart disease or recent acute myocardial infarction, Circulation 91, (7) (1995) 1936–1943.

[5] L. Fauchier, D. Babuty, P. Cosnay, M.L. Autret, J.P. Fauchier, Heart rate variability in idiopathic dilated cardiomyopathy: characteristics and prognostic value, Journal of the American College of Cardiology, 30 (4) (1997) 1009–1014.

[6] M. H. Asyali, Discrimination power of long-term heart rate variability measures, in: Proceeding of the 25th Annual International Conference of the IEEE EMBS, Cancum, Mexico, 1 (2003) 200-203.