Covariate shift adaptation in EMG pattern recognition for prosthetic device control

Abstract

Ensuring robustness of myocontrol algorithms for prosthetic devices is an important challenge. Robustness needs to be maintained under nonstationarities, e.g. due to electrode shifts after donning and doffing, sweating, additional weight or varying arm positions. Such nonstationary behavior changes the signal distributions - a scenario often referred to as covariate shift. This circumstance causes a significant decrease in classification accuracy in daily life applications. Re-training is possible but it is time consuming since it requires a large number of trials. In this paper, we propose to adapt the EMG classifier by a small calibration set only, which is able to capture the relevant aspects of the nonstationarities, but requires re-training data of only very short duration. We tested this strategy on signals acquired across 5 days in able-bodied individuals. The results showed that an estimator that shrinks the training model parameters towards the calibration set parameters significantly increased the classifier performance across different testing days. Even when using only one trial per class as re-training data for each day, the classification accuracy remained > 92% over five days. These results indicate that the proposed methodology can be a practical means for improving robustness in pattern recognition methods for myocontrol.

Concurrent Adaptation of Human and Machine Improves Simultaneous and Proportional Myoelectric Control

“…Most of them focused on the classification-based approach [19]- [22] and employed adaptation with the goal to compensate for nonstationarities [23]. In classification, unsupervised adaptation is possible by using the classifier-output as adaptation target and considering only classifier decisions with high confidence.…”

Improving robustness against electrode shift of sEMG based hand gesture recognition using online semi-supervised learning

“…sEMG is surface action potentials, and is sensitive to electrode repositioning. Electrode position change may be due to donning and doffing of prosthesis, or muscle fatigue in day-to-day [11]. Electrode shift has negative effect on the performance of a myoelectric control system.…”

“…Although these methods achieve satisfying results in hand gesture recognition, collecting labelled samples from users is inconvenience. A study [11] devised the re-training process which requires less required labeled samples. However, although the duration of data collection is shorten, the setup procedure may still be complicated.…”

“…The movement of electrode is usually caused by donning and doffing of prosthesis, and muscle fatigue [4], [8]. Many studies showed that accuracy of the recognition significantly [8][9][10][11] drops due to electrode shift since the distribution of collected signals changes. Although re-training the system [11], [12] can maintain satisfying performance, a training process requires labelled samples.…”

Model and experiments to optimize co-adaptation in a simplified myoelectric control system

“…The activity of remaining muscles might still be regressed against movements intended by the amputees while producing muscle contractions, but injuries could be such that the resulting EMG-movement relationship might be too deteriorated to be used to control movement. And, if the activity of the remaining muscles gathered while amputees are simulating movements, are sufficient to elicit a decent initial myoelectric control, signal nonstationarities due for instance to electrode shifts, sweat or varying postures, remain a major difficulty to deal with (Vidovic et al, 2014).…”