Development and Application of Computational Tools for Multi-Electrode Electroretinography

The spatial distribution of activity at the retina determines the spatial distribution of electroretinogram (ERG) potentials at the cornea. Here, a three-dimensional surface spline method is evaluated for interpolating corneal potentials between measurement points in multi-electrode electroretinography (meERG) data sets. A user interface was programmed to utilize this interpolation method to compare meERG data sets. The order of the spline was optimized for application with corneal potential maps. The effect of meERG electrode number and array arrangement on interpolation accuracy was investigated. 25-channel meERG responses were obtained from rat eyes before and after treatment to create local lesions. A 3rd order surface spline was used to interpolate meERG values resulting in smooth color-coded maps of corneal potentials. Potential maps were normalized using standard score values. Pre- and post-treatment responses were characterized by spatial standard deviation and by difference-from-normal plots. User interfaces were programmed in MATLAB to allow for both meERG signal processing and data analysis to generate interpolated feature maps of the cornea for comparison between individual or groups of eyes. Interpolation error analysis was performed on both meERG data sets, as well as test voltage inputs, to characterize and optimize the value for the spline order, m. The number and arrangement of electrodes in the meERG lens array were varied, and the interpolation error determined for each variation to determine minimum acceptable number of channels (channel yield) and also determine which pattern groups of electrodes eliminated resulted in least amount of interpolation error. The spatial standard deviation for eyes with local lesions were significantly higher than for healthy eyes. The 3rd order spline resulted in well-behaved corneal potential maps that maintained low error rate. Post-normalization, responses could be combined within experimental groups, and individual eyes with lesions were clearly distinguished from the healthy-eye mean response. The user interface allowed for streamlined analysis of multiple large meERG data sets efficiently. An optimum (lowest error) value of spline order, m = 2, was arrived at through interpolation error analysis for six different spline order values m = 1, 2, 3, 4, 5, 6. Channel elimination error analysis found that up to 64% of channels of the 25-channel array could be eliminated and still result in detectable spatial differences in the corneal potential signal. Eliminating groups of channels in patterns that were distributed evenly resulted in lower interpolation error than for eliminated patterns that were in clusters of adjacent electrodes in the lens array. This work demonstrates solutions to key challenges in the recording and analysis of meERG responses: visualization, normalization, channel loss, interpolation error minimization, identification of abnormal responses, channel yield, and electrode array pattern effects. Continued development of the meERG technique is relevant to research and clinical applications, especially where local detection of dysfunction (early progressive disease) or local therapeutic effect (subretinal injection) is of interest.