| A Deep Learning Based Framework for Denoising EEG Signals |
| Paper ID : 1103-ICEEM2025 (R2) |
| Authors |
|
zeinab mohammedelsherbieny Elsherbeny * FACULTY OF ENGINEERING |
| Abstract |
| Electroencephalography (EEG) signals play a critical role in neuroscience, clinical diagnostics, and brain-computer interface (BCI) systems. However, the presence of noise and artifacts such as muscle activity, eye movements, and environmental interference often limits the accuracy and reliability of EEG-based analyses. Traditional denoising techniques, including filtering, blind source separation (ICA, PCA), and wavelet transforms, provide partial solutions but struggle to balance noise removal and signal preservation. Recently, Artificial Intelligence (AI) methods have emerged as powerful alternatives for EEG signal enhancement. Deep learning architectures like convolutional neural networks (CNNs), recurrent neural networks (RNNs), autoencoders, and generative adversarial networks (GANs) have demonstrated superior performance in isolating and suppressing noise while preserving vital neural patterns. This survey reviews traditional and AI-based denoising approaches, evaluates their effectiveness using public EEG datasets and metrics, and highlights key applications of enhanced signals. Challenges and future directions, including real-time processing and explainable AI, are also discussed to guide further advancements.Furthermore, we propose a Denoising Autoencoder (DAE) framework designed to enhance EEG signal quality by learning nonlinear mappings from noisy to clean data. Experiments conducted on a subset of the EEGdenoiseNet dataset demonstrate that the proposed model effectively increases the Signal-to-Noise Ratio (SNR) from 7.21 dB to 15.87 dB, reduces the Root Mean Square Error (RMSE) from 0.145 to 0.062, and improves classification accuracy from 81.4% to 92.7%. These results highlight the capability of AI-driven solutions to suppress noise while preserving critical EEG features, paving the way for more reliable clinical and brain–computer interface applications. |
| Keywords |
| EEG Denoising, Quality Enhancement, Artificial Intelligence, Deep Learning, Artifact Removal |
| Status: Accepted |