基於離散小波轉換的心跳分類系統

Abstract

心電圖是心臟疾病臨床診斷的重要工具，透過心臟電生理活動紀錄，可以即時診斷防患於未然，其關鍵在於正確且快速地辨識心電圖中的心跳類型。本論文聚焦於心跳分類系統中的特徵擷取方式，主要是利用離散小波轉換與主成分分析取得心跳片段的型態特徵，加上相鄰心跳RR區間組成的動態特徵，並採用機率神經網路做為分類辨識器。本文針對MIT-BIH心律不整資料庫中十三種心跳類型的心電圖，探討不同的心跳片段長度及離散小波解析組合之效能，並增加心跳RR區間與小波係數的統計分析以加強辨識效果。系統模擬顯示本文提出的心跳分類方法之整體辨識率為98.3%，平均敏感性為91.5%，平均特異性為99.9%。

Electrocardiogram (ECG) is a pictorial representation of the electrical activity of the heart. ECG beat classification plays an important role in the clinical diagnosis of cardiac diseases, which can prevent potential patients from physical dangers at the early stage by immediate and accurate classification of beat types. This study focuses on the extraction of morphological and dynamic features of ECG signals. Specifically, discrete wavelet transform (DWT) and principal component analysis (PCA) are applied consecutively to each heart beat to extract morphological features. In addition, RR interval information is computed to provide dynamic features. These two different types of features are concatenated and a probabilistic neural network (PNN) classifier is utilized for the classification of heart beats into one of 13 classes. The proposed method is validated on the baseline MIT-BIH arrhythmia database and it yields an overall accuracy of 98.3%, sensitivity of 91.5%, and specificity of 99.9%.