以人工智能建构自动化心律不整分类系统.pdf

中文摘要 本文目的在提出自動化心律不整判讀系統。其所面臨的問題有 2 項目，第一是如何由原始的心電圖訊號，找尋出 QRS 波位置，第二 個問題則是由所定位的 QRS 波 ，進一步判斷此心跳是屬於正常、心 律不整節律或僅是雜訊所造成。由於在 QRS 波的標記上不可能做到 100%的準確率，因此錯誤標記就必定會影響到心跳節律的判斷。本 文將分三個部分1.) 利用 一導程Lead II 為輸入訊號自動化找尋 QRS 波位置。2.) 利用所標記 QRS波位置，自動截取 二導程Lead II及 V1 波形為輸入訊號，以圖像辨識的方法，辨識正常節律 、心律不整節律 或雜訊。 3.) 利用所標記 QRS波位置，尋找出 RR間隔，並以此時序 的變化為輸入資料，透過時序辨識的方法，辨識正常節律 、心律不整 節律或雜訊。分類器辨識波形包含正常節律、早發性心室縮、早發性 心房收縮、左束枝傳導阻滯和右束枝傳導阻滯。心電訊號資料是來自 MIT-BIH Arrhythmia資料庫，選用 33個檔案資料 12776個心跳週期。 圖形特徵心律不整的辨別正確度達 99.2% ，而時序特徵在雜訊的影響 下正確率還是可達 92.6% 。 關鍵字 : 心電圖、心律不整、早發性心室縮、左束枝傳導阻滯 I Abstract In this study, we propose an automatic system of arrhythmic classification which has to solve two major problems. First, the position of the QRS complex has to be detected from the electrocardiogram (ECG) signal accurately. Second, each QRS complex will be classified as a normal beat, an arrhythmic beat, or a noise only. Because the accuracy of QRS complex detection not is 100%, the mistaken mark of QRS complex would affect the classification result. In this study, there are three parts: one, using Lead II signal automatically detect QRS complexes. Two, the morphological features of Lead II and V1 were used to classify the normal sinus rhythm, cardiac arrhythmias and noise. Three, time sequence features of heart rate were used to classify the same problems. There are five categories of heart beats including normal sinus rhythm, premature ventricular contraction, premature atria contraction, right bundle ranch block and left bundle branch block. We used the 33 files of MIT-BIH arrhythmic Data Base that are selected 12776 QRS complex. The accuracy of morphological method ach