以適應特徵選擇與支持向量機實現心電圖辨識系統

Abstract

心電圖提供了診斷心臟病病和心血管病症的功能，為了能夠及時的監控病人的生理狀態，有時候必須持續長時間且連續的記錄病患所產生的心電圖資料，採用更多的心電圖資訊來判斷波形的物理變化，藉此可以較正確的評估病患目前生理情況，但是通常所得到的心電圖資料必須由專業的醫護人員來解析判讀。

本研究所提出了一個新的心電圖分析演算法，使用小波轉換分析頻帶來擷取心電圖特徵值，包含了改善特徵選擇和分類系統的設計，所擷取出的特徵向量作為心電圖辨識系統中最重要的特徵。而在心電圖辨識系統中較特別的特徵為QRS複合波組，這是含有極高頻的成份且能量較大的峰值波形。在辨識系統中採用支持向量機作為辨別不同種類心臟疾病的分類器。

Electrocardiogram signal (ECG) provides the functional aspects of the heart and cardiovascular system. In order to monitor the real-time evolution of the patients, the ECG signal is sometimes recorded continuously for one or more days. The availability of more and more information on the physical status and evolution of the patient is always desirable, but usually the information needs to be assimilated and evaluated by doctors or nurses.

We propose a new wavelet transform based ECG analysis algorithm with improving the feature extraction and classifier design. Inherited from the properties of WT, the extracted vectors can represent the most important features for ECG signals. It is particularly true for the QRS complex the can be recognized as the high frequency and high energy components. The system adopts support vector machines (SVM) to differentiate the types of heart diseases.