NLR, MLP, SVM, and LDA: a comparative analysis on EMG data from people with trans-radial amputation.
Alberto Dellacasa Bellingegni, Emanuele Gruppioni, Giorgio Colazzo and 4 others
PMID 28807038WHAT IT FOUND
For people with below-elbow amputation imagining five hand gestures, a raw-signal pattern-recognition method matched the standard method on the gesture-matching task and on a score that balances gesture matching with computer memory.
It was offline, not real-time prosthesis use.
Key findings
01The nonlinear logistic classifier trained on raw muscle signals performed similarly to the standard time-domain classifier on gesture matching and on the combined performance-memory score.
02SVM had the highest computational burden and a lower combined performance-memory score than the nonlinear logistic classifier when training data were sampled above 5 Hz.
03The study used steady-state phantom-gesture signals offline and did not test a real-time embedded prosthesis.
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 authors said the sample was not sufficient to generalize to all kinds of trans-radial amputation. All participants were already experienced in myoelectric control, so results may not apply to new prosthesis users. The analysis was offline and based on steady-state phantom-gesture signals, not real-time prosthetic use. No real-time embedded prosthesis testing or patient function outcome was reported. The final classifiers were trained at different sampling rates, so the comparison is technical rather than clinical.
Declared interests
Funded by the National Institute for Insurance against Accident at Work (INAIL) and the European Commission (H2020-ICT-2014-1, ICT-22-2014). No author conflicts of interest are stated.
The easy way to misread this
Do not read the gesture scores as evidence that this improves everyday prosthetic hand use. The study compared classification algorithms offline using steady-state signals, and it did not test a real-time embedded prosthesis.