Automatic Eyeblink Artifact Removal from Single Channel EEG Signals Using One-Dimensional Convolutional Denoising Autoencoder

• Published in: International Conference on Computer, Electrical & Communication Engineering (Online) pp. 1 - 7
• Main Authors: Acharjee, Raktim, Ahamed, Shaik Rafi
• Format: Conference Proceeding
• Language: English
• Published: IEEE 02.02.2024
• Subjects: Automatic Artifact Removal; Brain modeling; Brain-Computer Interface (BCI); Convolution; Convolutional Denoising Autoencoder; Correlation coefficient; Decoding; Electroencephalogram (EEG); Electroencephalography; Eyeblink Artifact; Noise reduction; Single Channel EEG; Training
• ISSN: 2768-0576
• DOI: 10.1109/ICCECE58645.2024.10497290

Eyeblink artifacts are common and significant disruption that occurs when the subject blinks during the recording of the Electroencephalogram (EEG) signals. Removing this artifact is essential to ensure the precision and reliability of the recorded brain activity. Compared to multichannel EEG systems, the limited spatial resolution of single channel EEG makes it more challenging to identify and eliminate eyeblink artifacts. In this work, we proposed a one-dimensional Convolutional Denoising Autoencoder (CDAE) architecture to efficiently remove the eyeblink artifacts from the single channel EEG signals. The publicly available "EEGdenoiseNet" dataset was used to synthetically generate the eyeblink-contaminated noisy EEG signals which were fed to the encoder to generate a compressed representation and capture essential features. From there, the decoder reconstructed the clean EEG signal. We calculated the Relative Root Mean Square Error (RRMSE) and Correlation Coefficient (CC) to assess the effectiveness of our proposed method and achieved an RRMSE of 35.4% and CC of 0.92. Our proposed method efficiently removes eyeblink artifacts from the single channel EEG signal and performs better in terms of CC compared to any other state-of-the-art method.