Non-invasively multimodal approach for decoding the intent of movement impaired individuals and low dimensional control of movement based on muscle synergies
Author
Pedersen, Jacob Koch
Term
4. term
Publication year
2009
Pages
133
Abstract
This thesis explores a non-invasive, multimodal approach to decode movement intent in individuals with severe motor impairment and to simplify the control of electrically stimulated neuroprostheses (FES/FET). The approach combines gaze direction with surface EMG from proximal arm muscles and uses a multilayer perceptron (MLP) neural network to predict user intent and muscle activation, with a Kalman filter as a benchmark. Additionally, muscle synergies were extracted via non-negative matrix factorization to enable low-dimensional, more manageable control. In data from seven subjects, integrating surface EMG with simulated gaze direction clearly improved intent prediction compared with EMG alone, and the MLP outperformed the Kalman filter. In one subject with recorded gaze, continuous activation of nine distal arm muscles was predicted from deltoid EMG and gaze direction. Furthermore, activity across twelve muscles could be reconstructed from a small number of synergies, suggesting a simplified control strategy for FES/FET. Overall, the findings indicate that multimodal, non-invasive decoding and synergy-based control hold promise for enhancing neuroprosthesis operation, though the evidence is based on limited data and requires validation in larger samples and real-time applications.
Denne specialerapport undersøger en ikke-invasiv, multimodal strategi til at afkode bevægelsesintention hos personer med betydelig motorisk funktionsnedsættelse og at forenkle styring af neuroproteser baseret på elektrisk stimulation (FES/FET). Tilgangen kombinerer blikretning og overflade-EMG fra proksimale armmuskler og anvender et multilags perceptron (MLP) neuralt netværk til at forudsige brugerens intention og muskelaktivering, med Kalman-filteret som reference. Desuden blev muskel-synergier udtrukket med ikke-negativ matrixfaktorisering for at muliggøre lav-dimensionel, mere håndterbar kontrol. I data fra syv forsøgspersoner gav integration af overflade-EMG og simuleret blikretning en tydelig forbedring i intentionforudsigelse i forhold til EMG alene, og MLP’et overgik Kalman-filteret. Hos én forsøgsperson med registreret blikretning kunne aktiveringsniveauet i ni distale armmuskler forudsiges kontinuerligt ud fra deltoideus-EMG og blikretning. Endvidere kunne aktiviteten i tolv muskler rekonstrueres ud fra et lavt antal synergier, hvilket peger på en mere simpel styringsstrategi til FES/FET. Samlet peger resultaterne på, at multimodal, ikke-invasiv afkodning og synergi-baseret kontrol kan forbedre anvendelsen af neuroproteser, men konklusionerne er baseret på et begrænset datagrundlag og bør bekræftes i større studier og i realtidsapplikationer.
[This apstract has been generated with the help of AI directly from the project full text]