Artificial neural networks for HD-sEMG-based hand position estimation: addressing inter- and intra-subject variability.
Giovanni Rolandino, Leonardo Lion, Taian Vieira and 3 others
PMID 41656222WHAT IT FOUND
Training the system with random electrode shifts and skin changes prevents accuracy loss when users put it on differently across sessions.
Adding other people's data did not help if the user's own data was present. A brief personal calibration remains necessary.
Key findings
01Without training on electrode repositioning, accuracy dropped significantly when testing on a new session, with mean distance error rising from 31.5 mm to 65.1 mm.
02Including electrode repositioning during training mitigated this drop, keeping extra-session error at 42.9 mm, which was significantly better than the no-repositioning baseline.
03Adding data from other subjects did not improve performance when any data from the target user was included in training; it only helped if no user-specific data was available.
STILL TO COME
How it was doneWhat they found
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 repositioning analysis used only 4 participants, limiting the precision of the effect estimates. All participants were able-bodied, right-handed volunteers; results may not apply to people with limb loss or neuromuscular conditions. Performance was assessed offline; real-time closed-loop control with functional tasks was not tested. The two datasets had different session intervals (less than 30 minutes for DS2.a vs. hours to days for DS1), which may have influenced the magnitude of performance drops. The study did not control the specific magnitude or direction of electrode shifts, only that they occurred.
The easy way to misread this
Do not assume that a large pool of other users' data can replace the need for personal calibration. The study showed that adding other subjects' data actually degraded performance if any of the target user's own data was included in the training set.
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 →