A novel adaptive artifacts wavelet Denoising for EEG artifacts removal using deep learning with Meta-heuristic approach

Electroencephalogram (EEG) is said to be a common tool to control neurological disorders, performed medical diagnoses, and cognitive research. But, EEG is generally polluted through various kinds of artifacts, which further causes complexities in interpreting the EEG data. These artifacts affect the efficiency of the wearable or portable EEG recording systems. Further, it results in more difficulties in the implementation of “neurologically-oriented mobile health solutions”. While comparing with the five most common neurological disorders, epilepsy has more dependency on EEG for diagnosis. Recently, a combined form of “Independent Component Analysis (ICA) and Discrete Wavelet Transform (DWT)” has been developed and is said to be a standard technique for EEG artifact removal. Moreover, the wavelet-ICA procedure depends on the requirements of arbitrary thresholding or visual inspection for finding the artifactual components in the EEG signals. This task introduces a deep learning and heuristic-based adaptive artifacts wavelet denoising approach for making epilepsy detection the most accurate. Initially, the EEG signal is decomposed by the Empirical Mode Decomposition (EMD) approach. In the decomposed signal, the first level of adaptive artifacts wavelet denoising is based on DWT. The second level of artifact removal is performed by a deep learning approach termed Improved CycleGAN (I-CycleGAN) with parameter optimization by the Opposition Searched-Elephant Herding Optimization (OS-EHO). With a deeply learned wavelet coefficient, the adaptive artifacts wavelet denoising is performed with OS-EHO-based thresholding. Further, the adoption of inverse DWT is applied followed by the signal reconstruction to generate the final artifacts removed signal. From the simulation findings, the proposed OS-EHO-CycleGAN secures 28%, 5.12%, 28.12%, and 24.2% improved than PSO-CycleGAN, GWO-CycleGAN, WOA-CycleGAN, and EHO-CycleGAN in terms of PSNR values with other existing meta-heuristic algorithms. The experimentation is evaluated on various biological artifacts such as “ECG, EMG, and EOG”, and the results reveal the superior efficiency of the suggested method while comparing with an existing approach based on different quality measures.