Analyzingobstructive sleep apnea(OSA)using machine perception and wavelet taransforms

Abstract

Obstructive Apnea is a respiratory based sleeping disorder where throat tissues falls back towards airways which it partially or completely blocks the airflow during the sleep. Due to the lack of airflow, oxygen levels in blood will drop suddenly which it increases blood pressure and strains cardiovascular system. This leads to increase the risk of Cardiovascular diseases, Stroke, Obesity, Diabetes, Hypertension etc. One of the most commonly diagnosed methodology for sleeping disorders is Polysomnography (PSG) which is expensive and takes much effort, due to these reasons in most of the cases sleeping disorders were undiagnosed. To overcome the drawbacks of PSG, practical and recent systems concentrate on the usage of electrocardiogram (ECG) for detection of OSA. To get precise ECG interpretation is essentially needed in order to evaluate the useful information inside the ECG signal. The standard method of visual analysis to evaluate the ECG signals by physicians are ineffective as well as time consuming. Therefore, an automated system which includes digital signal integration as well as evaluation is needed This paper proposes a system which it uses machine perception for analyzing ECG images and to detect the OSA abnormalities in ECG. Here the input data is taken in form of images instead of signals for accurate ECG interpretation. This can be done by using wavelet transforms, which is utilized to extract the coefficients of each ECG Simultaneously, auto regressive model (AR) is used to acquire the temporal structures of ECG wave forms, by using AR fit method. Based upon both wavelets transform as well as AR model their coefficients can be taken and integrated with each other to form a 1-D eigen vector where each vector represents a point in space. Based upon improvised Classification Algorithm, it can able to distinguishes between apnea and no apnea. Improvised classification Algorithm involves the combination of both K-Means as well as improvised KNN classifier is utilized to decrease the computation complexity and to increase the accuracy by using the hyper tuning parameters.