Purchase this article with an account.
M.C. Brown, A.C. Fisher, R.P. Hagan; Elimination of Artefacts From Visual Electrodiagnostic Recordings Using Adaptive Rules . Invest. Ophthalmol. Vis. Sci. 2006;47(13):1667.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Signal recovery in electrodiagnostic recordings by coherent averaging requires that the artefact sources (noise) are stationary, of zero mean and adequately sampled. This requirement is usually satisfied in the clinical situation for uncorrelated noise sources of the same order of magnitude and morphology as the signal to be recovered: however, it is likely to fail for noise sources arising from eye movements (ocular movements and blinks). These signal artefacts are spontaneous and heteromorphic. Here a post hoc analysis is used to identify and remove individual noise–corrupted epochs using intuitive heuristic rules which adapt continuously to the probability density functions (PDFs) of the artefactual noise sources using an identification strategy based on first order statistics.
To remove signal artefacts arising from eye movements and blinks from electrodiagnostic recordings without reliance on the continuous data record.
Synthetic, and conventional ERG and PERG recordings were made. Data were stored as arrays of discrete epoch segments to allow stimulus–by–stimulus analysis. The parametric distributions for a range of heuristic models were extracted for each epoch and a weighting applied based on their PDF prior to inclusion in the ensemble average: arbitrarily, 60% of the samples, selected on the basis of the 20th to 80th percentile range were attributed the weight of 1, with the remainder weighted adaptively between 1 and zero.
Improvement in SNR ranged from 10 to 60% depending on the level of eye movement artefact. This improvement was characterised formally as the reduction in the time required to converge the average (recovered) signal to within 0.63 (i.e. 1–1/exp) of its final (asymptotic) value.
The use of an adaptive heuristic model to minimise the inclusion of eye movement noise artefacts in the signals recovered by coherent averaging improves the SNR and decreases the recording time to a degree useful in clinical recordings. This method deals with artefacts of magnitudes below those detected by conventional ‘hard limit’ rejection schemes. It is applicable post hoc to discrete data epochs.
This PDF is available to Subscribers Only