Automatic and explainable assessment for Parkinson's disease by video-based human motion understanding.
Mengqi Liu, Yixing Ye, Haolun Li and 14 others
PMID 41964015WHAT IT FOUND
A phone-video system scored 16 Parkinson motor items from patient videos, with symptom accuracies of 90.1% to 97.2%.
When clinicians saw its movement measurements, their average item accuracy rose from 78.7% to 85.3%.
Key findings
01The system estimated ratings for all 16 vision-based MDS-UPDRS motor items, with accuracies of 97.2%, 90.1%, 96.6%, and 96.2% for masked face, bradykinesia, postural instability, and tremor.
02When 11 clinicians re-rated videos after seeing the system's objective indicators, their average accuracy across the 16 items increased from 78.7% to 85.3%.
03Tremor showed the largest clinician gain, with average accuracy increasing from 80.2% to 96.2% and all clinicians adopting identical ratings based on system-provided tremor amplitude.
STILL TO COME
How it was doneWhat they foundWhat it means for PTs
Read the rest of this summary
You get three full summaries a month, free, and we do not ask for a card. Search, the TL;DRs and your library stay unlimited either way.
What it does not show
The study measured how well the system matched a gold standard and how clinicians changed their ratings; it did not follow patients or test whether the system changed treatment decisions, falls, mobility, or quality of life. The gold standard for most items was the majority vote of five neurologists, so the system was compared with expert consensus rather than an external objective standard. The clinician-support experiment included 11 clinicians and 616 videos, and the tremor result may reflect strong reliance on the system because clinicians adopted identical ratings after seeing its amplitude values. Bradykinesia remained the weakest area: the system's accuracy was 90.1%, and the paper says clinicians still could not separate adjacent subjective ratings well when shown motion features. The videos came from hospital-based recordings of patients aged 47 to 71 years in China, and the paper does not report how the system performs on patient-recorded home videos or in other populations.
Declared interests
Funded by the Science and Technology Development Fund, Macau SAR; the National Nature Science Foundation of China; the Natural Science Foundation of Zhejiang Province; the National Natural Science Foundation of China; and the National Key Laboratory of Space Intelligent Control. The supplied text does not list competing interests.
The easy way to misread this
Do not conclude that this system should replace clinician assessment or improve Parkinson's care. The study measured agreement with ratings and a short-term change in clinician scoring, not patient outcomes, and clinicians' tremor agreement became identical partly because they adopted the system's amplitude values.
Summarised by AI from the full paper, without a clinician reviewing it. Check it against the source before it changes what you do. Read it on PubMed →