EMG Based Movement Intent Detection

Stroke is the leading cause of adult disability worldwide. Functional effects after a stroke include loss of motor control, atrophy, spasticity, alteration in muscle activation pattern, muscle synergies and compromised sensory inputs. Stroke causes hemiplegia which affects the upper extremities in particular the hand. Sensory motor recovery after a stroke is possible through intensive movement training (Ward et al.2019). Robot-assisted devices are an effective way to do repetitive and labor-intensive therapies. Active participation of patients is crucial in such robot-assisted therapies. Patients begin to slack if the therapies are not challenging enough. Strengthening of cortical connections by temporal coupling between the voluntary effort and afferent feedback of assisted movement is evident from the theory of Hebbian learning (Ethier, C., et al 2015). An effective way to guide adaptive plasticity could be through closed loop-control of robot-assisted therapy using movement intent detected from residual movements of stroke patients. However, this method does not work for severely affected stroke patients with no residual movements. EEG is used to detect movement intention in such patients, which is used to drive assisted movements of the limb. In line with this, (Ramos et al. 2013) reported that robot-assisted therapy contingent on movement intent detected from EEG is slightly more beneficial.

But EEG- BCI has drawbacks such as poor signal-to-noise ratio (SNR), and difficulty in precisely identifying which body part the patient is intending to move. Also, difficulty in setting up the system makes the EEG-BCI system unsuitable for routine clinical use. EMG could be a viable alternative for EEG-BCI. EMG is a simple, compact, and robust system suitable for routine clinical use. In a previous study by (Balasubramanian et al.2018), 73% of severely affected stroke patients had residual EMG who did not have any residual movement. The drawback of the study was its small sample size and the relatively simple EMG detector employed in the study. The best detectors for low SNR EMG is currently unknown.

Objectives :

Current

1.Analysis of existing algorithms for low SNR-EMG detection.

• Different detectors from the literature are tested on EMG signals having SNR of 0 dB and -3 dB generated from three different signal models, Gaussian, Laplacian and Biophysical models. The statistical decision-based detector – approximate generalized likelihood ratio test detector, the simple threshold-based detector – modified Hodges, and the entropy-based detectors – Fuzzy and Sample entropy detectors perform best across all three models of EMG.

2.Clinical study to estimate the percentage of severely affected stroke patients with residual EMG. (n =100 )

Future

1.To evaluate the efficacy of EMG- triggered robot-assisted therapy in severely affected stroke patients.

Reference

[1] Balasubramanian S, Garcia E, Birbaumer N, Burdet E, Ramos A. Is EMG a viable alternative to BCI for detecting movement intention in severe stroke?. IEEE Trans Biomed Eng. 2018 Mar 21.