Automatic Detection Of Paroxysmal Atrial Fibrillation Using The Scatter-plot-based Feature And ECGs

Atrial fibrillation(AF)is a common arrhythmia disease.When AF occurs,the effective atrial contraction is lost,which cause blood stasis in the atrium,further resulting thrombus formation.When thrombus falls off with blood flow to the brain,it is easy to cause ischemic stroke,even endangering life.Therefore,the timely diagnosis and treatment of paroxysmal atrial fibrillation(PAF)is very important and necessary.The traditional detection of PAF mainly depends on the visual inspection to check whether AF waveforms exist in ECGs,and needs long-term monitoring of patients,which makes the traditional visual detection method very time-consuming and the workload of manual analysis is huge.Therefore,developing automatic detection of PAF not only effectively overcomes the limitation of traditional detection methods,but also has a great significant in clinics.This paper proposes a simple and rapid method for automatic detection of PAF based on scatterplot.Firstly,several appropriate filters are applied to denoise the original ECG.Secondly,we denoised the ECGs are segmented,and the measurement indexes which can describe the changes in scatterplot of each segment are designed.These indexes are then defined as the extracted features,including the area of confidence ellipsoid,the scatter degree of confidence distance,the scatter degree of confidence angle.Finally,the extracted features are fused as the input of extreme learning machine to complete the automatic detection of PAF.The concrete structure of this article is as follows:Chapter 1 describes the background and significance of automatic detection of PAF.The research status of automatic detection of PAF based on atrial activity and RR intervals,as well as the basic ideas and steps of the proposed method in this paper are introduced systematically.Chapter 2 firstly introduces the basic knowledge about ECG signal.Then the mechanism and types of AF are stated.Finally,the characteristics of atrial fibrillation in ECG are shown.Chapter 3 introduces various background noises during the acquisition process,and then three kinds of main noise and corresponding filters are introduced systematically.Finally,the design principle and specific steps of the proposed automatic detection method are described in detail.Chapter 4 applies the proposed method to MIT-BIH atrial fibrillation database.The feasibility and effectiveness of the proposed method are verified by numerical experiments.