Detecting the symptoms of Parkinson's disease with non-standard video.
Joseph Mifsud, Kyle R Embry, Rebecca Macaluso and 4 others
PMID 38702705WHAT IT FOUND
A machine learning model using smartphone videos of a finger-to-nose task correctly detected Parkinson's symptoms in 91% of cases.
However, it failed to detect symptoms from hand rotation videos, and the system still requires manual identification of the patient in the frame.
Key findings
01The finger-to-nose task classifier achieved high accuracy, correctly detecting symptoms in 91% of videos, while the hand rotation task classifier performed poorly, detecting symptoms in only 25% to 35% of videos.
02The model relied heavily on shoulder and elbow movements, with 54% of selected features derived from the shoulder joint and 31% from the elbow.
03The method is not currently scalable for home use because multiple people often appear in the frame, requiring manual annotation to identify the patient.
STILL TO COME
How it was doneWhat they foundWhat it means for PTsWhat it means for OTs
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 sample size was small (28 participants). The videos required manual annotation to identify the patient when other people were in the frame, making it impractical for unassisted home use. The clinical rating scale used (SMA) is not widely recognized, and there was no inter-rater reliability check. The model's performance varied significantly by task, with poor results for hand movements due to video quality issues. The study did not compare the video analysis to ground truth motion capture data.
Declared interests
The study was supported by the Michael J. Fox Foundation for Parkinson's Research.
The easy way to misread this
Do not assume this tool is ready for clinical use. The high accuracy for finger-to-nose tasks did not translate to hand rotation tasks, and the system currently requires a human to manually identify the patient in the video, preventing true automation.