Neural networks based approach to remove baseline drift in biomedical signals

Abstract

Nowadays there exist different approaches to cancel out noise effect and baseline drift in biomedical signals. However, none of them can be considered as completely satisfactory. In this work an artificial neural network (ANN) based approach to cancel out baseline drift in electrocardiogram signals is presented. The system is based on a grown ANN allowing to optimize both the hidden layer number of nodes and the coefficient matrixes. These matrixes are optimized following the simultaneous perturbation algorithm, offering much lower computational cost that the traditional back propagation algorithm. The proposed methodology has been compared with traditional baseline reduction methods (FIR, Wavelet-based and Adaptive LMS filtering) making use of cross correlation, signal to interference ratio and signal to noise ratio indexes. Obtained results show that the ANN-based approach performs better, with respect to baseline drift reduction and signal distortion at filter output, than traditional methods.