使用心電圖ST段位置估測與動態校正策略偵測 心肌缺血事件的研究

Detection of Myocardial Ischemia Episode Using ECG ST Segment Position Estimation and Dynamic Correction Method

陳維軒(Wei-Syuan Chen)

指導教授 ： 余松年

國立中正大學/工學院/電機工程學系/碩士(2015年)

摘要

本研究提出使用新研發的心電圖ST段估測策略與動態校正策略，來建立一個高正確率的心肌缺血心搏與心肌缺血事件辨識系統。所謂心肌缺血心搏，是指心電圖波形中的ST段，相較於正常心搏的ST段，其電壓值下降達一定程度，則判斷為心肌缺血心搏。而心肌缺血事件，是指心肌缺血心搏持續達到一定時間，則稱為心肌缺血事件發生。
本論文前半段的研究，著重於提昇ST段的位置估測及ST特徵擷取的正確性，在特徵擷取中，特別加入區域化特徵來降低一些變化性較大的心搏所造成的誤差影響。論文的後半段在探討如何有效地使用分類器在本辨識系統中，在分類器辨識的部分又分為兩個階段，第一階段為心肌缺血心搏的辨識研究，所使用的分類器為支持向量機(SVM)，實驗方式是使用留一驗證法(leave-one-out cross-validation)，試圖探討當個別測試資料不包含在訓練資料集中的系統表現；在第二階段提出動態校正策略，針對事件的發生進行視窗與閥值的校正，希望能提升心肌缺血事件的辨識率。方法是針對區域化校正後的結果，加入十個心肌缺血心搏發生率相關特徵，使用倒傳遞類神經網路(BPNN)建構出針對不同資料特性合適的校正策略，並驗證此動態校正策略的可行性。
結果顯示，在辨識心肌缺血心搏的研究中，當使用支持向量機分類器以及留一驗證法的平均辨識率可達94.82%。而在對心肌缺血事件的辨識上，當使用最佳化的區域校正參數後，其平均辨識率可達98.48%，而當使用神經網路將此最佳化參數建立動態校正策略，對原來的檔案的平均辨識率為96.77%，對新的檔案的平均辨識率可達97.06%，顯示在對心肌缺血事件的辨識上，所提出的方法可以建構強健的動態校正策略，且能將所有的測試資料中的心肌缺血事件都辨識出來，正確的事件覆蓋率也能達到98%以上。
關鍵字：心電圖、ST段、位置估測、心肌缺血、區域化特徵、動態校正策略

關鍵字

區域化特徵 ； 動態校正策略 ； 心肌缺血 ； 位置估測 ； ST段 ； 心電圖

並列摘要

This study proposes to use novel ECG ST segment position estimation and dynamic correction methods to build an effective system for recognizing myocardial ischemic heart beats and myocardial ischemic episodes. The myocardial ischemic heart beat refers to the heart beat that shows depressed ST segment level when compared to the normal heart beat. On the other hand, a myocardial ischemic episode refer to the continuing of myocardial ischemic heart beats for more than a certain time period.
This study is divided into two parts. The first half of the study focuses on how to increase the accuracy of ST segment position estimation and ST segment feature extraction. In feature extraction, we added regional features to reduce the minus effects from the heart beats that show significant variations. The second half of the study focuses on how to effectively use classifiers in the system. This part is further seperated into two stages. In the first stage, we explored the effectiveness of using the support vector machine (SVM) classifier in recognizing myocardial ischemic heart beats. The leave-one-out cross-validation method was used to explore the accuracy when the test data was not included in the test data set. In the second stage, we proposed dynamic segmental correction strategy with windowing and thresholding to improve the accuracy of myocardial ischemic episode recognition. Ten features associated with the occurance of myocardial ischemic beats were used to establish suitable strategy, based on the result from segmental correction, for ECG records with different attributes using the back-propagation-neural network (BPNN) and its feasibility was justified.
The results showed that, in the recognition of myocardial ischemic heart beats, an accuracy of 94.82% was achieved using the SVM classifier with leave-one-out cross-validation. In the recognition of myocardial ischemic epsodes, a accuracy of 98.48% was achieved with the optimal segmental correction parameters. When we used the optimal parameters to built the dynamic correction strategy using BPNN, accuracies of 96.77% and 97.06% were achieved when tested using the original training records and novel test records, respective. The results demonstrated the robustness of the established dynamic correction strategy in recognizing myocardial ischemic epsodes. Moreover, all the myocardial ischemic epsodes in the records were recognized with over 98% accuracy in correct epsode coverage.

Keywords: Electrocardiogram (ECG), ST segment, position estimation, myocardial ischemia, segmental feature, dynamic correction strategy

並列關鍵字

dynamic correction strategy ； segmental feature ； myocardial ischemia ； position estimation ； ST segment ； Electrocardiogram (ECG)

參考文獻

[3] P. Exarchos, P. Costas,D.I.Fotiadis ,” An Association Rule Mining-Based Methodology for Automated Detection of Ischemic ECG Beats” IEEE Transactions on Biomedical Engineering, vol. 53, no. 8, August 2006.

acquisition, preprocessing, parameter measurement, and recording,”

[6] G. Y. Jeong and K. H. Yu,“ Morphological Classification of ST segment using Reference STs Set”, Conference of the IEEE EMBS, August 23-26, 2007.

[7] K. Shiva, T. Mohammad, A.S. Mahdi ,” Probabilistic Neural Network Oriented Classification Methodology for Ischemic Beat Detection Using Multi Resolution Wavelet Analysis”, Proceedings of the 17th Iranian

[13] 徐佑，“ 利用型態學特徵偵測心肌缺血的發生”，國立中正大學電機所，101年