Activity Recognition in Individuals Walking With Assistive Devices: The Benefits of Device-Specific Models.
Luca Lonini, Aakash Gupta, Susan Deems-Dluhy and 3 others
PMID 28798008WHAT IT FOUND
For patients using a new knee-ankle-foot orthosis, activity recognition works better when the model is trained on that device.
Healthy-person models performed poorly, and stair climbing remained hard to detect.
Key findings
01A global model trained on healthy people had a median balanced accuracy of 53%, while a global model trained on the new orthosis reached 61%.
02The best personal model used each patient's own data from the new orthosis and reached a median balanced accuracy of 76%.
03Even in the best model, stair ascent recall was 43.1% and stair descent recall was 48.0%.
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
Only 11 patients were studied, so model performance may change with a larger or different patient group. 2 patients could not ascend or descend stairs, so stair data were underrepresented. Participants performed a structured set of activities in a lab under supervision, not in natural environments. The study used a single waist-mounted accelerometer, so sensor choice and placement may affect accuracy. The study did not test combined global and personal approaches or transfer learning methods.
Declared interests
The authors declared no conflicts of interest.
The easy way to misread this
Do not conclude that wearable activity recognition is ready to monitor patients with a new orthosis. A model trained on healthy people had a median balanced accuracy of 53%, and even the best model recognized stair ascent at 43.1% and stair descent at 48.0%.