Enhanced Noiseremoval Tool for ECG Signals Using Deep Learning Techniques

This work concerns noise reduction for biomedical signals (e.g. ECG signals) that is distorted by one of the most undesired noises, baseline wander noise. Noise reduction is categorized into parametric approach, which relies on structuring models based on knowledge of signal and noise nature, and non-parametric approach, which is data-driven algorithm with no assumption about the signal nor the noise. In this paper, we propose a hybrid approach combining the well-structured models of the parametric approach and trying to enhance the training process using non-parametric approaches, such as principal component analysis (PCA) and singular value decomposition (SVD). These proposed approaches are trained and tested against the QT Database and the MIT-BIH Noise Stress Test Database from Physionet. Additionally, the proposed modified models are compared against their original ones. The proposed system for ECG denoising has a comparable performance in noise removal compared to other classical techniques. However, it has faster training convergence (around 11 epoch instead of 50 epoch of classical techniques) The simulation results verify the performance of the proposed models in ECG denoising as it yields comparable accuracy and training convergence better more 40% of the time.