Multimodal data-driven eye-movement subtypes and their cerebral glucose metabolic patterns in Parkinson's disease

Previously reported Parkinson's disease (PD) subtyping schemes often show limited stability and cross-cohort generalizability.

To derive data-driven oculomotor subtypes in PD using multi-task eye-movement assessment and to characterize their cerebral glucose metabolic patterns.

We administered a non-invasive multi-task eye-movement battery to 122 patients with PD and 69 healthy controls. Multidimensional oculomotor features were analyzed using unsupervised k-means clustering to identify PD subtypes. In a PD subset undergoing ¹⁸F-fluorodeoxyglucose positron emission tomography (FDG-PET; n = 30), regional cerebral glucose metabolism was quantified to compare metabolic profiles between subtypes.

Clustering identified two PD subtypes: an oculomotor-efficient subtype (PD-E) and an oculomotor-inefficient subtype (PD-I). The subtypes differed across multiple oculomotor parameters, with antisaccade (AS) metrics showing the most prominent divergence. Compared with PD-I, PD-E showed higher FDG uptake in frontotemporal cortices. Metabolic differences were directionally concordant with groupwise patterns in cognitive measures.

Integrating eye-movement digital phenotypes with FDG-PET metabolism may provide complementary information for cognitive-domain profiling and assessment in PD. Longitudinal studies and independent cohort validation are needed to confirm stability and clinical translatability.