Parkinson's: 3D cameras measure movement without sensors, in study

A neurologist examining a Parkinson's patient asks for a few simple gestures: a clenched fist, forearms turned in and out, a walk across the room. Each movement receives a mark from 0, normal, to 4, severe impairment. The scale, called the MDS-UPDRS (Movement Disorder Society-Unified Parkinson's Disease Rating Scale), lets clinicians on any continent use the same grid. Its weakness is the grid itself: a broad range of movements is squeezed into five categories, and one doctor's 'minor impairment' is not always another's.

At Duke University, a team from the Dunn lab set out to replace that judgment with measurement. Kyungdo Kim, a PhD student in biomedical engineering, worked with neurologist Kyle Mitchell and assistant professor Timothy Dunn on TULIP, short for Three-dimensional Understanding and Learning of Impairments in Parkinson's disease. Synchronized cameras film the person from several angles, with no markers and no sensors to wear, so the movement is natural. Machine learning rebuilds each motion as a three-dimensional pose, which yields figures such as degrees of rotation or centimeters of movement. Kim gives a concrete example: instead of 'your gait score is 1', the clinician could say that a change in leg angle from 45 to 60 degrees shows the gait has improved and that medication may be helping.

The team applied TULIP to three standard exams. For hand movements, it measured repeated opening and closing; for the forearms, inner and outer rotation; for walking, whole-body movement and gait. Combining the three gave a fuller picture than any single test. The measurements automatically separated people with Parkinson's from people without it, and estimated severity and disease duration. Movement amplitude and speed became more pronounced as the disease lasted longer. The authors are careful about what this means: Dunn notes that there is no widely accepted laboratory or imaging biomarker for Parkinson's, and that the study remains a research result that the team will test on larger and more varied patient groups.

For patients, the immediate gain would be a finer record of how their movement changes between visits, measured rather than estimated. For clinical trials, where a drug's effect can be small and hard to detect, a more sensitive measure could make it easier to see whether a treatment works. The study does not give a date for any clinical use, and the team presents TULIP as a complement to existing assessments. Dunn adds that the approach could also suit other disorders that affect movement, such as Alzheimer's disease, stroke and other neurological injuries, but the paper focuses on Parkinson's.