Machine learning for automating subjective clinical assessment of gait impairment in people with acquired brain injury - a comparison of an image extraction and classification system to expert scoring.
Ashleigh Mobbs, Michelle Kahn, Gavin Williams and 3 others
PMID 39039594WHAT IT FOUND
A computer model scored arm movement abnormality during walking as consistently as three experienced physiotherapists.
It correctly identified impaired versus unimpaired movement in 84% of cases, matching human judgment without the usual disagreement between different clinicians.
Key findings
01The machine learning model achieved 84% accuracy in distinguishing impaired from unimpaired arm movement when compared to the median score of three expert assessors.
02Agreement between the model and individual expert assessors was comparable to the agreement seen between the assessors themselves.
03The model tended to slightly under-predict severity scores, but this error was small and did not significantly affect overall accuracy.
STILL TO COME
How it was doneWhat they foundWhat it means for PTsWhat it means for OTsWhat it means for SLPs
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 used a small sample size, which required complex statistical methods to validate the results. The model was trained and tested in a controlled laboratory environment with fixed camera positions and lighting, so its performance in a busy clinic or home setting is unknown. The 'gold standard' for comparison was the median score of three humans, not an objective instrument, so the model is only as good as the average human observer. The model slightly under-predicted severity, which might mean it could miss subtle improvements or deteriorations if not calibrated carefully.
Declared interests
Funding was provided by the Royal Automobile Club of Victoria and the Physiotherapy Research Fund. No other conflicts of interest were declared.
The easy way to misread this
Do not assume this technology is ready for clinical use. The model was tested in a controlled environment with specific camera setups and lighting, and its accuracy on the detailed severity scale was only moderate (60.4%). It has not been proven to work in real-world clinic or home settings.