Canalization of Retinal Electrophysiological Failure and Recovery:A Multi-Cohort State-Space Analysis of PERG and Full-Field ERG

Abstract

Objective: To test whether retinal electrophysiological dysfunction exhibits a low-dimensional canalization geometry: larger departures from a normative state are accompanied by smaller angular deviation from a shared failure direction, and whether recovery after an intervention is directionally organized rather than isotropic. Methods: Three openly available datasets were analyzed. First, transient PERG waveforms from PERG-IOBA were reconstructed from raw signals into amplitude-loss (A), waveform-geometry (G), and latency-distortion (T) coordinates. A cross-disease failure axis was estimated with diagnosis-wise cross-fitting. Second, the axis direction was frozen and transferred without outcome-driven reorientation to the independent LEOPs LA3 full-field ERG dataset, where analogous A/G/T coordinates were defined from control-referenced amplitude, waveform geometry, and implicit time. Third, an independent silicone-oil removal (SOR) cohort with paired pre/post full-field ERGs was analyzed in an A-G state space to test recovery geometry. Results: In PERG-IOBA, 170 participants across 12 disease groups showed a strong inverse association between failure radius and angle to the cross-fitted axis (Spearman rho=-0.6194; within-diagnosis permutation p=0.000050). The association remained negative after every single-diagnosis deletion and all 66 two-diagnosis deletions; diagnosis-cluster bootstrap 95% CI was -0.709 to -0.405. In LEOPs LA3 (217 participants), the fixed PERG-derived direction retained an inverse radius-angle association (rho=-0.3283; group-stratified permutation p=0.000020; bootstrap 95% CI -0.442 to -0.205), although a stringent random-axis comparison was borderline (p=0.0504). In the paired SOR cohort (11 eyes), combined A-G abnormality radius decreased after SOR in 8/11 cases (Wilcoxon one-sided p=0.0122; 100,000 sign-flip p=0.00949), and recovery vectors were directionally concentrated relative to an isotropic null (resultant length=0.667; p=0.00490). Conclusions: These analyses identify a reproducible cross-sectional PERG failure-canalization pattern, a cross-modality ERG signal aligned with the frozen PERG direction, and independent paired evidence for non-random recovery geometry. The findings support a testable state-space framework for retinal electrophysiological failure and recovery, but do not establish a universal biological law or a longitudinal PERG trajectory. Independent prospective cohorts are required.Keywords: pattern electroretinogram; electroretinogram; retinal electrophysiology; state space; canalization; waveform geometry; recovery