ABSTRACT
The electrocardiogram (ECG) is a necessary diagnostic test in the medical industry that graphs minute changes in the regular functioning of the heart to detect abnormalities. However, the odds of certain noises (aka undesirable signals) accumulating throughout the ECG recording process are considerable that obstructs the correct interpretation of the signals, which highlights the importance of the removal of these noises. Empirical Mode Decomposition (EMD), Non-Local Mean, Discrete Wavelet Transform and Wiener filter are among the various ECG denoising approaches available. An Ensemble Convolutional Neural Network – Bidimensional EMD (ECBEMD) model is proposed in this work which combines the BEMD technique with a deep learning model that consists of convolution layers with successive pooling, batch normalization and a fully connected layer for denoising ECG images. The proposed Denoising Convolutional Neural Network (CNN) model is trained where it learns the noisy features which are mapped and later removed. When compared with few of the most used, prominent and efficient denoising techniques, the proposed model is proven to be a better denoising technique. On evaluation, it is observed to have lesser latency and better outcomes that preserve the morphological structure of the ECG image.
