Purchase this article with an account.
D. L. McCulloch, M. S. Borchert, P. Garcia-Filion, C. Fink, M. J. Austin, A. C. Fisher; Human and Computer-Automated Systems for Measurement of Pattern Electroretinograms (PERGs) in Infants With Optic Nerve Hypoplasia (ONH). Invest. Ophthalmol. Vis. Sci. 2010;51(13):3270.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Pattern electroretinograms (PERGs) can reflect the extent of functional loss in optic nerve hypoplasia (ONH). As PERGs have inherently low signal-to-noise ratios (SNRs), they can be difficult to detect when degraded by pathology or noise. We compare a system for automated PERG analysis with expert human interpretation in data containing a range of clear to undetectable PERGs.
PERGs were recorded under chloral hydrate sedation in 35 children with ONH (aged 3.5-36 mo). Checkerboard stimuli (0.5-4 degree check widths) were focused uniocularly using an optical system incorporating the cycloplegic refraction. Data on 40 PERGs were analyzed; 20 selected at random from all eyes and 20 from eyes expected to have good vision (fellow eyes or eyes with mild ONH). For each PERG, raw data for 150 check reversals were reviewed manually deleting trials with movement artifact, slow-wave EEG (4-8 Hz) or other artifacts. The automated system identified cardinal positions in PERGs as turning points in local 3rd-order polynomials fitted in the -3dB bandwidth [0.5 ... 45]Hz using the webpage tool: http://clinengnhs.liv.ac.uk/esp_perg_1.htm Confidence limits were estimated from bootstrap resampling. We compared the human and automated systems for PERG detection and for the cardinal positions of P50 and N95.
The automated system detected 28 PERG signals above the noise level (p0.5, p.05); four of these had notable contamination with slow-wave EEG. Trial-by trial review of these data required approximately 6.5 hours while automated data processing required <4 minutes, excluding programming.
An automated computer system for PERG analysis, using bandwidth filtering and local polynomials to localize peaks offers rapid, objective detection and cursor positioning based on the statistical properties of the data. The bootstrap technique provides confidence intervals. The system provides data similar to that of expert clinical interpretation.
This PDF is available to Subscribers Only