心電圖訊號分析演算法與硬體架構設計

Abstract

心電圖是心臟搏動相關的電位變化圖，而且對於醫師診斷出不同的心臟疾病和來監測、評估病人的病情是非常重要的。從兩段心跳之間所取出的心電圖樣本形狀通常用來辨識心臟方面的疾病。本篇論文中我們提出一個高效能的心電圖辨識系統，此系統可以適應性地從446個參數中選擇最重要的特徵出來，再利用支持向量機來辨識這些心電圖。而我們測試MIT-BIH Arrhythmia database中的心電圖辨識率可達98.09%，在已發表的論文中是效果最好的方法之一。

另外，我們也設計一個硬體的工具來完成抽取小波轉換的係數以及支持向量機的辨識器。此硬體可以增加辨識過程的速度以及嵌入成可攜式的元件。

Electrocardiogram (ECG) is a representation of the electrical activity of heartbeats and it is quite an important signal for doctors to diagnose cardiac disease and monitor patient conditions. The shape of each ECG beat cycle as well as the interval time between two successive beats is commonly used for identifying the types of heart diseases. In this thesis, we propose a high performance ECG recognition system which adaptively selects the most important features from 446 candidate parameters and identifies the heart condition based on modified support vector machines (SVM). With tested by MIT-BIH Arrhythmia database, the final classification result can achieve 98.09% which is believed to be the best one in the published literatures.

On the other hand, we also design a hardware engine dedicated for extracting wavelet transform based features and classification by SVM. The engine may help to speed up the recognition process and integrated into a portable device.