Classification complexity in myoelectric pattern recognition.
Niclas Nilsson, Bo Håkansson, Max Ortiz-Catalan
PMID 28693533WHAT IT FOUND
Two methods for judging how hard it is to separate muscle-signal movement classes tracked how reliably a control system could classify those movements better than two other methods.
The test used healthy people, not patients.
Key findings
01Modified Mahalanobis was the distance definition that correlated most closely with average results of LDA and SVM classification accuracy, and it was used in the remaining results.
02Feature sets chosen by nearest neighbor separability were significantly higher in classification accuracy than reference sets in all but one comparison, and feature sets chosen by modified Mahalanobis were significantly higher in nine out of 12 comparisons.
03Purity and repeatability index had low correlation with classification accuracy and were dropped from later experiments.
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 study used healthy subjects only, so it does not show performance in people with stroke, amputation, phantom limb pain, or other clinical conditions. The outcomes were algorithm classification accuracy and motion-test completion time, not patient function, pain, prosthesis use, or rehabilitation benefit. The real-time test was a short pre-recorded movement task, not everyday use of a prosthetic or rehabilitation device. The data sets were small, with 20 subjects in one set and 17 subjects in the other. The paper evaluates engineering methods for choosing movements and features, not a treatment delivered by therapists.
Declared interests
The work was funded by Stiftelsen Promobilia, VINNOVA, and Innovations-Kontor Väst. The supplied text lists only these funders.
The easy way to misread this
Do not treat this as proof that myoelectric pattern recognition improves patient care. The algorithms were tested on healthy people performing recorded movements, and the outcomes were classification accuracy or task completion time, not clinical function.