Retinal Biomarkers of Folate and Vitamin B12 Metabolic Dysfunction: A Framework for Machine Learning-Assisted Detection of Cerebral Folate Deficiency.

Disruptions in folate (vitamin B9) and vitamin B12 metabolism, including nutritional deficiency, folate receptor alpha (FRα) autoantibodies, and MTHFR/DHFR polymorphisms, impair one-carbon metabolism and produce measurable retinal structural, microvascular, and functional changes, offering a non-invasive window into systemic and cerebral metabolic dysfunction, particularly cerebral folate deficiency (CFD). This review synthesizes peer-reviewed evidence on retinal alterations linked to folate/B12 deficiency, hyperhomocysteinemia, FRα autoantibody syndromes, and MTHFR/DHFR variants, alongside artificial intelligence (AI) and machine learning (ML) approaches applied to retinal imaging for metabolic, anemic, and nutritional optic neuropathy detection. Three convergent phenotypes emerge: structural changes (retinal nerve fiber layer and ganglion cell complex thinning, optic disc pallor, chorioretinal atrophy), microvascular abnormalities (reduced vessel density, foveal avascular zone enlargement, capillary dropout), and functional deficits (centrocecal scotoma, dyschromatopsia, reduced contrast sensitivity); they arise from homocysteine-mediated endothelial toxicity, mitochondrial impairment, and eNOS uncoupling. Existing AI/ML models for anemia and optic neuropathy establish technical feasibility but do not target folate-specific phenotypes. We propose a dedicated multimodal ML framework integrating structural, perfusion, functional, and biochemical/genetic data as a research agenda for automated, non-invasive CFD detection. Given the established folate-autism spectrum disorder (ASD) risk association reported in FRAA-positive cohorts, such a framework, once validated, could support prenatal and neonatal screening in FRAA-positive or genetically high-risk pregnancies, enabling earlier leucovorin treatment and reducing neurodevelopmental risk.