A method for reconstruction of visually evoked potentials from limited amount of sweeps

Visually evoked potentials (VEPs) are signals evoked by a visual stimulus. They consist of brief discrete deflections embedded in background electroencephalographic (EEG) activity, which often has larger amplitude. Background EEG cancelation is a major part of VEPs analysis algorithms often realized by coherent averaging or other methods requiring large minimal amount of registered sweeps. In some cases, especially for pediatric patients, or in poor patient compliance cases, long procedure duration and fatigue might cause impaired attention and non-steady target fixation, affecting the quality of recorded VEPs. The possibility to reconstruct VEPs in every single sweep from limited size ensembles opens new diagnostic possibilities and shortens the registration procedure improving its quality. A proposed method is based on truncated expansion (Karhunen-Loève transform) of VEP signals applying generalized universal basis functions (eigenvectors of covariation matrix) calculated from learning set of sweeps, i.e. an ensemble of collected typical recordings. It realizes the possibility to reconstruct a signal from every single sweep even in limited size ensembles of registered sweeps. Application of adaptively time-shifted basis functions enables optimal reconstruction of the signal with latency shift or jitter.