適用於可移動心電圖之心房震顫檢測法:使用支持向量機之學習法則

Abstract

近20年來，心房震顫已經在西方國家成為最重要的健康問題之一，在2014年，心房震顫影響了歐洲及北美大約2%~3%的人口。為了達成長期照護的目標，我們團隊設計了一個無線多導極心電圖系統附有即時偵測心律以及警示系統。基於兩個原因，我們選擇心房震顫當做我們檢測的目標病徵: (1) 心房震顫在已開發國家以及年長族群中是非常普遍的疾病，因為全球人口老化的關係，在未來心房震顫將會成為一個重要議題。 (2) 心房震顫雖然沒有立即致死性，但是其與某些重症的疾病息息相關，例如：缺血性中風。病患的ECG心電圖訊號可以用來判別病患是否有心房震顫，P波消失以及不規律的RR區間是可以用來判別心房震顫的兩個ECG波形。在此篇論文中，我們撰寫MATLAB程式語言並且使用LIBSVM來模擬心房震顫演算法，LIBSVM是一個免費的開發工具，用於實作支持向量機(SVM)。演算法總共擁有19個特徵，包含時域特徵、頻域特徵以及非線性特徵。訓練資料以及測試資料皆來自MIT-BIH裡面的心房震顫資料庫。我們也實作了另一個由C. Huang等人所研究的心房震顫偵測演算法為了比較我們演算法的性能。兩個演算法有相同的輸入。實驗結果顯示我們演算法的精確率達97.57%，優於C. Huang等人的90.06%。在未來，這個演算法會實作到雲端平台上，以達到可以即時偵測心電訊號的用途。

In the last 20 years, atrial fibrillation has become one of the most important health problems in western countries. In 2014, atrial fibrillation affects about 2% to 3% of the population in Europe and North America. In order to monitor heart wave signal for a long time, our team designed a wireless multi-lead ECG system for long-term heart monitoring including real-time detecting and alarming system. We choose Atrial Fibrillation (AF) as the target arrhythmia of our detecting algorithm due to two reasons: (1) AF is a prevalent arrhythmia in the developed country and the group of elders. Owing to the global population aging, it will be a significant issue in the future. (2) AF is not related to imminent mortality but it is related to some highly morbid conditions like embolic stroke. The patient’s ECG signals can be used to determine whether the patient has AF or not, the absence of P wave and irregular RR interval are two ECG traces can be used to determine AF. In this thesis, we used MATLAB to simulate an AF detecting algorithm implemented by LIBSVM which is a free tool can implement support vector machine (SVM). The algorithm has totally 19 features, including time domain, frequency domain, and nonlinear features. The training data and the testing data are both from the MIT-BIH atrial fibrillation database. We also implemented another AF detection algorithm made by C. Huang et al for the purpose of comparing with our algorithm’s performance. Two algorithms were used the same input. The experimental result manifested that the accuracy of our algorithm is 97.57%, it is better than C. Huang et al’s 90.06%. In the future, the algorithm will be implemented on the cloud platform to realize the real-time heart wave signal detecting.