Artificial Intelligence Assisted Whole Waveform Electrooculography as a Functional Biomarker Profile of Retinal Pigment Epithelium Health

The retinal pigment epithelium (RPE) is a highly specialised polarised monolayer essential for photoreceptor survival and outer retinal homeostasis. In current clinical electrophysiology, RPE function is assessed using the electrooculogram (EOG), most commonly interpreted through the Arden ratio, a scalar metric derived from the dark-trough and light-peak amplitudes of the 30-minute recording. This conventional approach reduces a temporally rich electrophysiological time series to two values, discarding the majority of the physiological information embedded in the waveform. This manuscript proposes a conceptual framework for AI-assisted whole-waveform analysis of the clinical EOG to extract multidimensional functional information about RPE physiology. The central idea is that the EOG waveform represents a dynamic molecular profile of RPE function, in which distinct temporal features correspond to distinct ion transport processes and cellular mechanisms within the RPE monolayer. Within this framework, the full EOG time series is interpreted as an RPE Functional Profile — a multi-axis electrophysiological signature reflecting the integrated activity of multiple molecular systems, including Na⁺/K⁺-ATPase–driven standing potential regulation, bestrophin-1-mediated chloride conductance, intracellular Ca²⁺ cycling, apical chloride transport associated with the fast oscillation, and metabolic stability of the RPE ion transport machinery. The manuscript introduces several conceptual advances: • Waveform reconceptualisation — the EOG signal is treated as an information-dense temporal waveform rather than a source of two amplitude values.• Mechanistic feature mapping — physiologically interpretable waveform features (kinetics, latency, oscillatory behaviour, signal complexity, and stability metrics) are mapped to candidate molecular mechanisms within the RPE.• AI-assisted feature extraction — modern deep-learning architectures operating on full electrophysiological time-series data are proposed as analytical tools capable of extracting high-dimensional waveform features inaccessible to conventional analysis. The framework is explicitly presented as a hypothesis-generating theoretical model rather than a validated diagnostic tool. Several proposed capabilities — including early detection of RPE dysfunction prior to amplitude changes — remain predictions requiring empirical validation through prospective clinical datasets. The goal of this work is to establish the conceptual and computational foundation for transforming electrooculography from a single-ratio screening test into a multidimensional functional assay of RPE physiology, enabling richer electrophysiological phenotyping of retinal diseases and improved interpretation of an already widely available clinical test. Version 2 Note Version 2 incorporates multiple scientific corrections and clarifications relative to Version 1 (DOI: 10.5281/zenodo.18918403), including: • consistent apical localisation of Na⁺/K⁺-ATPase in the RPE• clarification of the relationship between applied electric field experiments and the native transepithelial potential• more precise framing of the light-peak signalling cascade as a proposed model rather than an established pathway• appropriately hedged interpretation of the fast oscillation and CFTR-related chloride transport• clarification of the ATP1B1 instability hypothesis relative to established transcriptomic findings• improved hypothesis framing and document layout Keywords retinal pigment epithelium; RPE; electrooculography; EOG; Arden ratio; RPE Functional Profile; electrophysiology; artificial intelligence; deep learning; time-series analysis; transepithelial potential; BEST1; macular degeneration; retinal electrophysiology; Arden ratio; LP:DT ratio