Assessing movement disorders such as Parkinson's disease means observing how somebody actually moves, across many kinds of movement. Digital assessment promises to make that objective and repeatable, but pipelines combining several sensing modalities across diverse motor tasks have not been well characterised.

This proof-of-concept study records synchronised EEG and inertial measurement data from six participants across ten motor activities, 52 recording pairs in total, and evaluates an EEGNet plus Transformer model. The scale is honest about what it is: a first careful look rather than a clinical claim.

The interesting phrase is task-dependent complementarity. The two modalities are not uniformly good or bad, they are good at different things depending on the movement, so which sensor to trust changes with the task. That argues against a single fixed fusion recipe and toward one that knows what the person is currently being asked to do.