EOG ARTIFACT CLEANING IN EEG SIGNAL USING DENOISING AUTOENCODER

• Main Author: Fauzy Perdhana, Hasbian
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/63549
• Institution: Institut Teknologi Bandung

Electroencephalography (EEG) is a recording technique to record electrical activity on the brain using electrodes attached to the head scalp. Electrooculography (EOG) artifact is one of the artifacts that are prone to appear on EEG due to eye movement and cause EEG signals to deform. To fix the EEG signal, we have to remove the artifact while conserving EEG information. In this research, we detect EOG artifactual signal using ICA and peak detection and used a generative model Denoising Autoencoder (DAE) to reconstruct clean EEG by using EEG artifact-corrupted signal. Our artifact detection method scores 85% sensitivity and 83% Positive Predictive Value on the secondary dataset and 82% sensitivity on the primary dataset. We train the DAE model with 10-fold crossvalidation and got 0.007 ± 0.008 Mean Squared Error (MSE). We demonstrated DAE on its ability to generate a clean EEG segment by feeding it contaminated EEG segment.