A High-Precise Arrhythmia Detection Method Based on Biorthogonal Wavelet and Fully Connected Neural Network

A novel approach for detecting feature points of ECG signals is proposed, which is based on an adaptive wavelet-threshold-slope method. Biorthogonal spline wavelet is applied for QRS complex detection. The original signals are decomposed with the equivalent filters of a biorthogonal wavelet by Mallat algorithm. In addition, threshold and slope methods are used for auxiliary monitoring of QRS complex. A fully connected neural network is proposed for arrhythmia detection, after morphological and statistics features are calculated by the position of feature points. MIT-BIH Arrhythmia Databases are used to verify the recognition accuracy of the proposed method, and comparative experiments are conducted. Experimental results demonstrated that adaptive threshold and slope detection methods are robust against noise, and the fully connected neural network has high performance for arrhythmia detection.