Motion Artifact Removal from EEG Signals Using the Motion-Net Deep Learning Algorithm
Year: 2025
Type: article
Abstract: Electroencephalography (EEG) is vital in brain signal analysis, with mobile EEG (mo-EEG) extending itsutility to movement scenarios. However, mo-EEG is highly susceptible to motion artifacts, degrading signalquality. Existing methods for motion artifact removal often exhibit limitations: subject-specific approacheslack accuracy, while machine learning models trained on heterogeneous datasets risk overfitting. This studyintroduces Motion-Net, a novel Convolutional Neural Network (CNN)-based one-dimensional signal reconstructionnetwork for subject-specific motion artifact removal. Motion-Net is the first machine-learning studydedicated to subject-specific removal and the first to incorporate visibility graph (VG) features, enhancinginput data and mitigating small dataset challenges. Evaluated through three experimental approaches usingan EEG dataset with motion-artifacted signals, reference ground truth, and accelerometer data, Motion-Netdemonstrated superior performance when VG features were integrated, achieving an average motion artifactreduction percentage (η) of 86% ± 4.13, a Signal to Noise Ratio improvement of 20 ± 4.47 dB, and a MeanAbsolute Error (MAE) of 0.20 ± 0.16. These findings underscore the effectiveness of integrating additionalfeatures to enhance model performance, promising advancements in mo-EEG data processing and analysis.
Language: English
Subfield: Cognitive Neuroscience
Field: Neuroscience
Domain: Life Sciences
Open Access status: hybrid
Estimated APC paid: $2,590
OpenAlex ID: W4415646119