A time domain constrained subspace-based estimator for extracting a visual evoked potential (VEP) from a highly noisy brain activity is proposed. Generally, the desired VEP is corrupted by background electroencephalogram (EEG) behaving as colored noise, making the overall signal-to-noise ratio as low as -10 dB. The estimator is designed to minimize signal distortion, while keeping residual noise below a specified threshold. Also, the algorithm applies a Karhunen-Loeve transform to decorrelate the corrupted VEP signal and decompose it into two parts called signal and noise subspace. Before an inverse Karhunen-Loeve transform is applied, the noise only subspace is discarded. VEP enhancement is therefore achieved by estimating the desired VEP only from the signal subspace. The performance of the filter to detect the latencies of P100's is comprehensively assessed using realistically simulated VEP and EEG data. Later, the effectiveness and validity of the algorithm are evaluated using real patient data recorded in a clinical environment. The results from both experiments show that the estimator generates reasonably low errors and high success rate.
[1]
Yariv Ephraim,et al.
A signal subspace approach for speech enhancement
,
1995,
IEEE Trans. Speech Audio Process..
[2]
P M Rossini,et al.
Clinical utility of evoked potentials.
,
1996,
Electroencephalography and clinical neurophysiology. Supplement.
[3]
Xiao-Hu Yu,et al.
Time-varying adaptive filters for evoked potential estimation
,
1994,
IEEE Transactions on Biomedical Engineering.
[4]
Saeed Gazor,et al.
An adaptive KLT approach for speech enhancement
,
2001,
IEEE Trans. Speech Audio Process..
[5]
R. Fitzpatrick.
eMedicine
,
2002,
Medical reference services quarterly.
[7]
Nidal S. Kamel,et al.
Single-trial extraction of visual evoked potentials from the brain
,
2008,
2008 16th European Signal Processing Conference.