Robust Processing of Visual Evoked Potentials

Abstract

Electrophysiology measurements attempt to get information about neural pathological conditions by measuring electrical cortical responses to appopriate stimuli.

Visual Evoked Potentials (VEPs) are the responses of occipital visual cortex to time-varying images captured by human eyes. They consist of a quasi-deterministic signal embedded in a strong background noise, produced by Electro-EncephaloGraphic (EEG) activity, sensors and amplifiers.

The VEP analysis is useful for early diagnosis and monitoring of diabetes, neural diseases such as multiple sclerosis, vascular diseases of the brain, check of the visual acuity in non-collaborative patients (children, old-aged people with degenerative neural diseases), and as a support for medico-legal investigations.

The signal-to-noise ratio is usually very low and a statistical averaging on multiple stimuli is required in order to recover an acceptable VEP waveform.

However the background noise is highly non stationary, clearly not Gaussian distributed and in some realization there is a strong quasi-periodic interference due to cerebral α-waves.

Moreover, the loss of attention and blinking may cause the lack or the weakening of the VEP signal in some trials. This impacts the reliability of standard (unweighted) average commonly used to reduce noise effects. So estimators that are resistant to the presence of outliers in the data set should be used for VEP preprocessing.

In this work, traditional robust estimators, such as the trimmed mean, and the median are compared with a modern neural version of the Iteratively Reweighted Least Squares (IRLS) which uses the recently developed Block Recursive Least Squares (BRLS) learning algorithm.

The same IRLS is used for the final deconvolution of brain spikes, that are the main target of VEP analysis.