Optimization Driven Variational Autoencoder GAN for Artifact Reduction in EEG Signals for Improved Neurological Disorder and Disability Assessment

Abstract This study introduces an innovative method for minimizing artifacts in electroencephalography (EEG) signals by integrating brainstorm optimization (BSO) with a variational autoencoder generative adversarial network (VAE-GAN), resulting in the BrOpt_VAGAN model. EEG signals are critical for the diagnosis of neurological disorders, for brain-computer interface (BCI) applications, and for the monitoring of neurological disabilities. However, EEG data often contains artifacts from physiological sources — such as electrooculographic (EOG), electromyographic (EMG), and electrocardiographic (ECG) signals — which can distort the accuracy of brain activity readings. Our proposed BrOpt_VAGAN model combines BSO with a VAE-GAN framework to more effectively remove these artifacts, thus improving the clarity and accuracy of EEG signals. In this model, the VAE first reduces the raw EEG signals into a lower-dimensional representation that captures the essential signal patterns while filtering out the noise. The GAN component then refines this representation via adversarial training, effectively minimizing artifacts and improving the quality of the processed EEG data. BSO optimally adjusts the encoding and decoding parameters within the VAE-GAN structure, enabling the model to handle different noise levels and helps to find different neurological disorders. Preliminary results show that BrOpt_VAGAN performs significantly better with an accuracy of 98.5 % and an error rate of 11.23 %, enabling a clearer and more precise EEG signal reconstruction.