Detection Atrial Fibrillation Based On Rr Interval And Deep Belief Networks

Atrial fibrillation is one of the most common cardiac arrhythmia, with its prevalence increasing each year. AF, which is associated with other types of cardiac conditions, can have dangerous implications, including an increased risk of stroke, heart disease, mortality and thus threatening the health of human beings. Therefore, the development of atrial fibrillation detection system can detect atrial fibrillation. And detection of AF is of great significance of improving patient care, improving the quality of treatment, reducing the disease incidence and mortality in critically ill patients. However, the accuracy of previous AF detection algorithms is still insufficient. Through the survey, it is difficult of detection atrial fibrillation to stable and can effectively distinguish the characteristics of atrial fibrillation. According to the above insufficiency in AF method studies,this thesis will focus on the study of detection method of AF in ECG and propose a novel, high-perfomance method suitable for the ECG analysis.The contents of this thesis mainly include:1, the RR interval data processing. RR intervals of the Electrocardiograph(ECG) were pretreated with two kinds of analysis methods. Firstly, the difference sequence of RR interval is obtained based on RR interval sequence of ECG. Then the difference sequence of RR interval is transformed into histogram sequence and Shannon entropy. Secondly, the difference sequence of RR interval is transformed into the sign sequence based on symbolic dynamics to get Shannon entropy of the probability distribution of substring length. If the RR interval data as input,it can be data migration, but processed RR interval data can solve this problem.2, using deep belief networks(Deep Belief, Networks, DBNs) to detect the atrial fibrillation task. DBNs is developing rapidly in decades, which is a machine learning method combines unsupervised learning process and supervised learning process. Because the ECG signal is complex and changeable, and interfered by various noise.So the detectionof atrial fibrillation are difficult to extract the feature stability and effectively distinguish atrial fibrillation. And sometimes doctors can not determine the diagnosis. This paper use the deep belief network to detect the atrial fibrillation,which is based on a large amount of data and feature extraction and classification.The database that MIT-BIH signals is studied to evaluate the accuracy of the method for AF detection. Experiment results demonstrate that the proposed method has the ability to detect of AF based on the signals.