Native components analysis of the spectral electroretinogram

The electroretinogram (ERG) is a mass electrical response from all electrically activated components of the retina, recorded with the goal of identifying the individual contributions of relevant components for the purposes of electrodiagnosis of eye diseases and other systemic medical conditions. The primary hypothesis being tested was that the ERGs across the spectrum in the mesopic range of intensities could be fully accounted with a duplex (two-component) model of linear combinations of rod- and cone-pathway responses. Full-field square-wave ERGs were measured with the RETeval device at 2 Hz for 7 spectral bands: red, yellow, green, cyan, blue, magenta, and white, in increasing steps of 0.5 log units from 3 to 300 phot cd/m², totaling 35 conditions for each eye of three neurotypical participants. A novel three-stage process termed Native Components Analysis (NCA), designed to overcome the distributive and orthogonality disadvantages of conventional linear component analysis, was implemented to identify the components contributing to the On-response of the overall ERG. The first step was select the ERG waveforms representative of each region of the response matrix. They were thus designated in terms of a) high and low intensities and b) the narrowband red, green and blue spectral regions. These 6 waveforms were taken as the native component candidates for an optimized fit to the full dataset. The second step was to determine the fit of these ERG components so-defined to the overall set of recorded ERG On-responses from each eye – a 140-parameter fit to the 10,500-parameter dataset. This approach was then compared with the standard approach of orthogonal Principal Components Analysis (PCA) to provide comparable compression. Over 6 datasets from the two eyes of three participants, the fit of the first 4 factors of the novel NCA approach accounted for 95.0 % of the overall variance in the data, compared with 97.5 % for the standard PCA approach. Adding components beyond the best 4 provided no significant improvement in the fits. For the individual datasets, the fit of the PCA accounted for 95.4 – 99.1 % of the variance, while the fit of the representative ERGs of the NCA approach accounted for 89.6–98.1 % of the variance across the individual datasets, validating the strategy of using representative ERG responses as the analytic components. The NCA fits strongly disconfirm the duplex rod/cone model that the ERG is a combination of just two temporal components, showing that as many as four separate components are required to account for the variance in the 35 waveforms in the participant group, with consistent structure across the spectral datasets. These results validate the utility of the novel Native Components Analysis approach to functional response analysis of retinal signals.