University of Twente Student Theses
Automatic EMG envelope detection for leg muscles using deep learning from a multi-electrode embedded garment
Rispens, O.C. (2024) Automatic EMG envelope detection for leg muscles using deep learning from a multi-electrode embedded garment.
|
PDF
2MB |
| Abstract: | Surface electromyography (sEMG) can be used to determine the activity of a muscle. However, applying electrodes in the right positions is time-consuming and requires extensive anatomical knowledge. A multi-electrode embedded garment was created by Simonetti et al. for the lower leg, to speed up electrode placement. Muscle-specific EMG envelopes were extracted using a non-negative matrix factorization (NNMF)- based clustering method for lower leg muscles. In the present study, the garment and the NNMF-based clustering method were adapted to the upper leg. Additionally, two new approaches were created to extract muscle-specific EMG envelopes: one using a convolutional neural network (CNN), and the other an encoder-decoder network (EDN). Validation was performed using two datasets: for the upper leg EMG recordings of 8 subjects were made, and for the lower leg three subjects. Before recordings, two electrodes of the garment for five lower or upper leg muscles were manually selected based on the SENIAM guidelines. The identified EMG envelopes by manual selection, NNMF-based clustering, the CNN, and the EDN were gait-cycle averaged. The performance of NNMF-based clustering and the CNN and EDN were compared by calculating the R 2 value between each and manual selection. The median overall R 2 value of both CNN (0.9) and EDN (0.9) for the lower leg was significantly larger (p<0.01) than that of NNMF (0.8), indicating that the EMG envelopes created by CNN and EDN more closely resembled those made through manual selection. For the upper leg, no significant results could be reported due to the small number of subjects, but the NNMF-based method had more R 2 values in the ’very weak’ (<0.2) and ’weak’ (0.2-0.39) categories and less in the ’very strong’ (>0.8) than CNN and EDN. NNMF-based clustering, CNN, and EDN generally found muscle-specific EMG envelopes that resembled those found by manual selection. To conclude, the newly created CNN and EDN-based methods outperformed the previously created NNMF-based methods and all methods found satisfactory EMG envelope results. These methods could play a great role in accelerating the electrode placement process. |
| Item Type: | Essay (Master) |
| Faculty: | ET: Engineering Technology |
| Programme: | Biomedical Engineering MSc (66226) |
| Link to this item: | https://purl.utwente.nl/essays/102809 |
| Export this item as: | BibTeX EndNote HTML Citation Reference Manager |
Show download statistics for this publication
Show download statistics for this publicationRepository Staff Only: item control page