Smartphone-Derived Ocular Motor Biomarkers Enable AI to Assess Neurodegeneration in Huntington’s Disease

Digital biomarkers derived from consumer devices offer new opportunities for remote neurological assessment. However, most AI-based approaches depend on large, disease-specific training datasets, limiting their applicability in rare disorders. Large language models (LLMs), trained on broad medical corpora, may enable clinically meaningful inference without disease-specific model training when provided with structured physiological inputs. In this prospective proof-of-concept study, individuals with genetically confirmed Huntington’s disease (HD) and age-matched healthy controls completed an ocular motor assessment using an in-house-developed smartphone application. Quantitative eye movement metrics were validated against expert neurologist ratings and subsequently provided to LLMs using a structured prompt. Models generated an AI-assigned HD probability score (HAIPS) based exclusively on ocular motor data. Twenty-six participants were included. Smartphone-derived metrics showed strong agreement with clinical ratings (Spearman ρ 0.76–0.95; all p < 0.001). HAIPS reliably discriminated individuals with HD from controls (AUC 0.879–0.944), with no significant differences across models. Among HD participants, higher HAIPS correlated with established motor and cognitive measures (Spearman ρ 0.74–0.86; all p < 0.01). These findings demonstrate that LLMs can generate clinically meaningful probabilistic assessments of HD from smartphone-derived ocular motor data without disease-specific training, highlighting a scalable framework for AI-supported assessment in neurodegenerative disorders.