Visual Acuity Prediction Based on Transient Sweep Visual Evoked Potentials and the Effect of Image Degradation
Purpose: To investigate the accuracy of a new transient sweep visual evoked potential (SVEP) technology to estimate optotype visual acuity (VA) under various viewing conditions and to compare the standard model for SVEP VA estimation (linear regression) with innovative analysis models. Methods: VERIS™ 6.4 software (Electro-Diagnostic Imaging, Inc.) was used for transient SVEP recordings. Several technical adjustments were required to resolve a flash artefact inherent to LCD monitors. Thirteen participants (ages: 21-59; 7 males; mean VA -.11 logMAR) underwent an orthoptic examination, pattern visual evoked potentials, and SVEP testing under normal conditions and with image degradation by lenses (+0.50 to +3.00) and/or Bangerter foils (BF) (8 foils: VA ~0.1 to light perception). SVEP response curves were obtained by two techniques: subjective wave marking (Expert) and objective Wavelet Analysis (WA) (MATLAB®, Natick MA). The Haar wavelet was used to decompose the raw waves into the summation of high-frequency wavelet coefficients and to create automated SVEP response curves. Analysis models were used to determine optotype VA from SVEP response curves in a population-based data set: 1) linear regression; 2) amplitude ratios; 3) peak position; 4) peak amplitude; 5) points >1.6 x noise level; 6) area under the curve. These models and their possible combinations were applied to individual data to test for clinical application of the technique. Predictive relationships were determined by correlation and repeatability was assessed using Bland-Altman plots Results: Although good overall correlation was seen between Expert and WA (lens data: 0.92; BF data: 0.95), WA showed precedence for noise assessment, reduced data variability and repeatability. WA was eventually used of all data analysis. Population- averaged (“group”) data revealed that SVEP results underestimated optotype VA when no or minimal image degradation was used and overestimated optotype VA with marked image degradation. Of all 6 analysis models, Model 2 (ratio between low and peak spatial frequency stimuli) and Model 5 (number of data points above 1.6 x the noise level) showed the most consistency for both image degradation conditions (lens and BF) and highest correlation coefficients (all above 0.91). Individual data showed more variability with lower correlation coefficients for all models (range: 0.33 to 0.49) but improved correlation coefficients with all Models 1-6 combined (0.62). However, Bland-Altman plots showed substantial variance with midrange acuities (15-25 c/d). Binary combinations optimized the correlation coefficient to 0.65 with model combinations of 1,2,4 and 1,2,4,5. Receiver operating curve for individual data revealed good sensitivity/specificity for high and low levels of ETDRS acuity with area under the curve values of 0.90 and 0.94 respectively with poor values for intermediate acuity levels (~ 0.57).Subsequent review of raw data revealed high and low normative amplitude groups with correlation coefficient of 0.80 and 0.63 respectively using model combination 1-6. Conclusion: Automated WA shows precedence over the Expert technique and provides objectivity for SVEP data that inherently shows substantial variability. Correlations between SVEP predicted acuity and optotype acuity can be optimized by using new analysis models and model combinations. Normative SVEP response curves appear to fall into normative high and low peak amplitude groups with much better correlation between SVEP and optotype acuity with normative high amplitudes.