Voice, finger-tapping and gait: AI spots early Parkinson's from a phone
A multimodal AI model that combines smartphone voice, finger-tapping and gait data flagged early-stage Parkinson's with strong accuracy.
Published in npj Parkinson's Disease, this study explored whether everyday smartphone signals could help identify Parkinson's earlier. Researchers combined three kinds of data — voice recordings, finger-tapping movement, and gait — into a single multimodal model built on a support vector machine.
The integrated model reached an accuracy of around 0.86 overall, and about 0.82 for detecting early-stage Parkinson's during the medication 'off' phase. Features such as arm swing, the proportion of time spent with tremor, and finger-tapping rhythm differed meaningfully between people with early Parkinson's and age-matched controls.
Because the inputs come from a device most people already own, the approach points toward accessible, low-cost screening that could support remote and at-home triage — the same goal that drives modern AI screening tools.
This is an in-house summary. Read the full reporting at the original source:
