Robust Feature Extraction from Noisy ECG for Atrial Fibrillation Detection
In this article, we propose a novel methodology for detecting 1D periodical time series (single lead Electrocardiograph, ECG) by classification from QRS features extracted from the signal.Our approach starts by dividing the input signal in windows of a number of periods.The boundaries of a period are detected using the state of the art QRS detector.For each period, the input signal is filtered and the S, T, P and Q points are localized.Next we define two sets of classification features, one measuring the distances between the identified points and one considering the differences between the ECG values corresponding to these points.These features are further combined with the features from the remaining periods from the same window.This is the input data for the classification stage, where we use a decision tree.The process is repeated for each window extracted from an ECG and the final result is decided by a voting mechanism.
Related projects
No projects linked.
Attachments
No attachments yet.