Towards Automatic Eeg Signal Denoising By Quality Metric Optimization

Electroencephalography (EEG) signals are widely used in areas such as: mental disease research, psychological evaluation and Affective Computing. One of the obstacles faced by researchers is related to EEG noise filtering. Involuntary muscle activity, such as eye-blinks and mandibular movements, insert noise into the signal which has a negative impact on its quality thus may result in misleading conclusions. Consequently, this study proposes an approach to remove noise in EEG signals based on a deep learning strategy that optimizes quality assessment algorithms. Furthermore, our methodology trains a model that learns how to optimize algorithms of quality assessment. In such manner, EEG signal users will not need human interference to extract noise which saves time and resources. To evaluate the robustness of our approach, we compare it with a baseline band-pass filter in terms of Peak Signal-to-Noise Ratio (PSNR), Structural Similarity index (SSIM) and quality scores. Our approach has demonstrated superior performance over the baseline technique in terms of PSNR and considering the second quality score applied. These are still preliminary results, yet they show great potential for continued development. Furthermore, this approach provides a new perspective on how to build deep learning methodologies in order to remove noise in EEG signals.