Date of Award

January 2015

Document Type

Thesis

Degree Name

Master of Science (MS)

Department

Electrical Engineering

First Advisor

Reza Fazel-Rezai

Abstract

A dramatic growth of interest for wearable technology has been fostered by recent technological advances in sensors, low-power integrated circuits and wireless communications. This interest originates from the need of monitoring a patient over extensive period of time. For cardiac patients, wearable heart monitoring sensors have already become a life-saving intervention ensuring continuous monitoring during daily life. Therefore, it is essential for an accurate monitoring and diagnosis of heart patients. Patients can be equipped with wireless, miniature and lightweight sensors. The sensors temporarily store physiological data and then periodically upload the data to a database server. These recorded data sets are then analyzed to predict any possibility of worsening patient's situation or explored to assess the effect of clinical intervention. To obtain accurate response with less computational complexity as well as long battery life time, there is a demand of developing fast and accurate algorithm and prototypes for wearable heart monitoring sensors. A computationally efficient QRS detection algorithm is indispensable for low power operation on electrocardiogram (ECG) signal.

In need of detecting QRS complex, most of the early works were proposed based on derivatives of ECG signal. They can be easily implemented with high computational speed. But owing to the inherent variability in ECG, these methods are highly affected by large derivatives of baseline noises. Algorithms based on neural network (NN) showed relatively robust performance against noise but requires exhaustive training and estimation of model parameter. On the other hand, wavelet based methods have the choice problem of mother wavelet. Hence, none of these methods is suitable for giving a long battery performance in wearable devices with high accuracy.

Recently, Wang et al. proposed a novel dual slope QRS detection algorithm which has less computational complexity as well as high accuracy. Considering that the width of the QRS complex is relatively fixed, this algorithm is based on the fact that the largest change of slope usually happens at the peak of QRS complex. The hardware requirement is also low. However, the method has a set of time consuming slope calculations on both sides of each sample. To avoid such time consuming slope calculation, only one sample on each side can be highlighted. In addition, the multiplication of the left and right hand side slope should give us a very high value in QRS complex.

The goal of this thesis is to develop a new computationally efficient method to detect QRS complexes and compare with the other renowned QRS detection algorithms. MIT-BIH arrhythmia database based on patients of different heart diseases and database containing ECG from healthy subjects are used. To analyze the performance, false negative (FN) and false positive (FP) are evaluated. A false negative (FN) occurs when algorithm fails to detect an actual QRS complex quoted in the corresponding annotation file of the database record and a false positive (FP) means a false beat detection. Error rate (ER) , Sensitivity (Se) and Specificity (Sp) are calculated using FP and FN.

Share

COinS