Purpose: To evaluate discrete wavelet transform (DWT) features as quantitative biomarkers of macular cone function from pattern electroretinography (PERG) in macular-predominant inherited retinal diseases (mpIRDs).
Methods: In total, 486 PERG recordings from 123 participants were obtained from the PERG-Institute of Applied Ophthalmobiology open-access data set and analyzed. Twenty mother wavelets were screened with an energy-to-entropy ratio criterion; six (haar, sym2, sym4, db4, coif1, fk4) were retained for feature generation. After feature cleaning and correlation pruning, a final set of 141 features was obtained and averaged per participant to avoid visit bias. Group separation was assessed with nonparametric statistics. Inverse-DWT signal reconstruction was performed with the sym2 wavelet to algorithmically determine time-frequency indices needed to preserve N35, P50, and N95 peaks. The smallest set of indices that achieved this was retained.
Results: Sym2-D6-2 (38-75 ms, 13-27 Hz) emerged as the top discriminative feature (res = 0.644, common-language effect size = 0.875) and correlated strongly with the clinical macular cone marker |P50-N35| (rcorr = 0.95) across 67 normal participants (262 recordings). Compared with |P50-N35|, the same index showed tighter, nonoverlapping group distributions, a higher diagnostic area under the curve (0.875 vs. 0.835), and a larger effect size (res = 0.644 vs. 0.576).
Conclusions: DWT-derived time-frequency features, particularly sym2-D6-2, provide robust, multidimensional biomarkers of macular cone function. These quantitative endpoints hold promise for monitoring disease progression and evaluating therapeutics in mpIRDs.
Translational relevance: Sym2-D6-2 provides an objective metric of macular cone function that could serve as a quantitative endpoint in mpIRD trials.