Abstract Background and aims Eye movement abnormalities are common after stroke and may reflect both neural damage and behavioral impairment. Here, we investigated whether variability in eye movements extracted from resting-state fMRI can serve as a longitudinal marker of stroke severity and recovery, and whether it relates to functional brain connectivity deviations over time. Methods We analyzed 109 stroke patients with resting-state fMRI acquired at acute (2 weeks, n=109), subacute (3 months, n=91), and chronic (1 year, n=85) stages. Oculomotor variance was quantified as the variance the of a multivariate pattern analysis within eye regions. Clinical severity was assessed using the NIH Stroke Scale (NIHSS). Functional brain deviation was computed by deriving functional connectivity gradients via diffusion embedding, constructing a reference gradient space from healthy controls' data, and calculating regional Euclidean distances from this reference for each patient. Longitudinal associations between oculomotor variance and either NIHSS or brain connectivity deviations were examined using linear mixed-effects models with subject-specific random intercepts and slopes. Results Oculomotor variance was significantly associated with NIHSS scores in the acute phase (ρ=0.24, p=0.011). Longitudinal mixed-effects modeling revealed a significant interaction between oculomotor variance and time, indicating that higher oculomotor variance was associated with a steeper improvement in clinical severity over time. When applied to functional brain deviation, oculomotor variance–time interactions showed trends in a few regions (p0.05, uncorrected), primarily within higher-order visual areas. Conclusions Oculomotor dynamics capture individual-specific recovery trajectories after stroke. They provide a promising bridge between neural disruption and behavioral impairment, with potential utility for personalized prognosis and longitudinal monitoring. Conflict of interest Cemal Koba: nothing to disclose. Jan Argasinski: nothing to disclose.
Koba et al. (Fri,) studied this question.