Pattern Electroretinography Signal Reconstruction in Rare Eye Diseases using Wavelet Transform.

Pattern Electroretinography (PERG) used in visual electrophysiology can assess macular cone and retinal ganglion cell function. However, its small amplitude and poor SNR limit clinical application, especially in rare eye diseases due to small sample sizes and low recording amplitudes. To better understand these limitations, this study implements the Discrete Wavelet Transform (DWT) to reconstruct signals of normal and rare pathologies. The goal is to remove unnecessary noise, assessing which wavelets are better for reconstructing PERG signals, and providing the first exploration of frequency bands that may be containing or removing pathologic content. After excluding recordings not meeting inclusion criteria, 358 PERG recordings in both normal and pathologic subjects were obtained from an open-source dataset on physio.net. Two different wavelets, Haar and Daubechies 2 (db2), were used for extracting biologically relevant frequency components. Reconstruction accuracy was quantified in terms of correlation coefficients r. Db2 wavelet yielded superior performance and gave high r (>0.95) consistently for most conditions. Across 110 normal control recordings, db2 detail and approximation levels 6 together yielded a r = 0.99, while other pathologies like Congenital Stationary Night Blindness reconstructed with only detail level 5 and approximation level 6 yielded an r = 0.91. This wavelet-based methodology retains diagnostic information and reveals subtle signal variations by isolating disease-relevant signal components while minimizing noise. These results reinforce DWT as a helpful tool to enhance PERG analysis in rare eye diseases by overcoming data scarcity for early diagnosis. Future work will involve integrating these methods into learning models where select decomposition levels will be leveraged to enhance prognostication.