利用心電圖與光體積變化描計圖辨識嚴重心律不整之研究

Recognition of Serious Arrhythmia Using Electrocardiogram and Photoplethysmography

沈大祐(Ta-Yu Shen)

指導教授 ： 余松年

國立中正大學/工學院/電機工程學系/碩士(2016年)

摘要

本研究提出使用心電圖(Electrocardiogram, ECG)與光體積變化描計圖(Photoplethysmprgraphy, PPG)來建構一套嚴重心律不整之辨識系統，辨識四種疾病，心搏過慢(Bradycardia)、心搏過速(Tachycardia)、心室頻脈(Ventricular Tachycardia)以及心搏停止(Asystole)。
提出的辨識方法分為兩個部分，第一個部分為使用最簡單且直接的方式，以連續心率與波寬特徵來直接辨識嚴重心律不整。第二個部分為使用直接辨識搭配分類器辨識，使用到的分類器有兩種，支持向量機(SVM)與倒傳遞類神經網路(BPNN)。在特徵擷取中，本研究發現當疾病發生時，ECG訊號與PPG訊號的波形與振幅皆會產生變化，故提出波形相關特徵期望能提升辨識率，從結果得知，本研究所提出的波形相關特徵對於整體辨識率是有所幫助的。另外，為了能在更有效地提升整體辨識率與減少辨識所花費之時間，研究中亦提出了階層式的分類方法，利用直接辨識(Direct, D)與分類器辨識(Classifier, C)作搭配，分別為DC、DDC與DCC，並使用留一交叉驗證的方式得到辨識結果，從研究結果得知，不管在任何訊號組合下，皆以DDC系統架構(搭配BPNN)擁有較佳之辨識率，使用兩種訊號辨識率為99.24%，使用ECG訊號辨識率為99.24%，使用PPG訊號辨識率為94.66%。
在本研究中，雖然以使用兩種訊號與單一使用ECG訊號搭配DDC系統架構辨識效果最佳，但本研究最終目標是期望僅用PPG訊號來辨識嚴重心律不整，由結果顯示，單一使用PPG訊號搭配DDC系統架構辨識效果可達到94.66%之辨識率，且平均辨識一筆所花費時間約為0.11秒，此結果亦證明了辨識嚴重心律不整演算法的高效能和即時辨識的可行性。

關鍵字

嚴重心律不整 ； 心電圖 ； 光體積變化描計圖 ； 波形相關特徵 ； 階層式分類 ； 支持向量機 ； 倒傳遞類神經網路

並列摘要

In this study, we proposed a serious arrhythmia recognition system based on two physiological signals. Electrocardiogram (ECG) and Photoplethysmprgraphy (PPG) were used to recognize four kinds of disease, including bradycardia, tachycardia, ventricular tachycardia, and asystole.
The proposed method was divided into two parts. The first part was the direct identification method that recognized serious arrhythmias by a continuous heart rate and pulse width characteristics. The second part used direct identification combined with a classifier for recognition. Two classifiers, support vector machine (SVM) and back-propagation neural network (BPNN) were used. For the feature extraction, we discovered that when the disease occurred, the waveform and amplitude of the ECG signal and PPG signal would change. Therefore, we hope to use the waveform relevant features to improve the accuracy. The results showed that the waveform relevant features proposed in this study raised the overall recognition rate. Moreover, in order to more effevtively enhance the overall accuracy and reduce the recognition time, we proposed multi-level classification methods that combined direct recognition method (D) and classifier recognition method (C), including DC, DDC, and DCC. We also employed leave-one-out scheme for cross validation. According to the results, regardless of signal combination, DDC system with BPNN classifier had best the accuracy. The accuracy was 99.24% when using two signals, 99.24% when using ECG signal alone, 94.66% when using PPG signal alone for recognition.
According to the result, although the use of two signals and ECG signal with DDC system have the best performance, our final goal is to use PPG signal alone for serious arrhythmia recognition. The accuracy could reach 94.66% when using PPG signal alone with DDC system for recognition, and the average time was about 0.11 seconds to recognize one record. The result demonstrated the high-capability and real-time recognition feasibility of the proposed serious arrhythmia recognition system.

並列關鍵字

Serious Arrhythmia ； Electrocardiogram (ECG) ； Photoplethysmorgraphy (PPG) ； Waveform Relevant Features ； Level Classification ； Support Vector Machine (SVM) ； Back-Propagation Neural Network (BPNN)

參考文獻

[3] N. V. Thakor, Y. S. Zhu, and K.Y. Pan., “Ventricular tachycardia and fibrillation detection by a sequential hypothesis testing algorithm,” IEEE Transactions on Biomedical Engineering, Vol. 37, No. 9, pp. 837-843, September. 1990.

[4] X. S. Zhang, Y. S. Zhu, N. V. Thakor, and Z. Z. Wang, “Detecting ventricular tachycardia and fibrillation by complexity measure,” IEEE Transactions on Biomedical Engineering, Vol. 46, No. 5, pp. 548-555, May. 1999.

[5] M. A. Arafat, A. W. Chowdhury, and Md. K. Hasan. “A simple time domain algorithm for the detection of ventricular fibrillation in electrocardiogram,” Signal, Image and Video Processing, Vol. 5, Issue. 1, pp.1-10, March. 2011.

[6] Z. Hou, Y. Zhang, “Ventricular Fibrillation Detection by An Improved Time Domain Algorithm Combined with SVM,” Medical Biometrics, pp. 189-194, 2014.

[7] Q. Li, C. Rajagopalan, “Ventricular Fibrillation and Tachycardia Classification Using a Machine Learning Approach,” IEEE Transactions on Biomedical Engineering, Vol. 61, pp. 1607-1613, June. 2014.