Graduate Thesis Or Dissertation
 

Adaptive Movement Intent Decoding for Intuitive Control of Neuroprostheses

Public Deposited

Downloadable Content

Download PDF
https://ir.library.oregonstate.edu/concern/graduate_thesis_or_dissertations/wp988s60w

Descriptions

Attribute NameValues
Creator
Abstract
  • Movement intent decoders, which interpret volitional movement intent from human bioelectric signals, can be incorporated into modern neuroprostheses to offer people living with limb loss or paralysis the potential to regain their lost motor control. Machine learning methods have become the research standard for continuous decoders with high degrees of freedom (DoFs), but these methods suffer from performance deterioration over time because the human body is a time-varying system, whereas these decoders are typically fixed after initial training. To mitigate this performance deterioration, this thesis presents a novel formulation and real-time framework for neural network-based online learning decoders, in which the decoder parameters adapt to changes in the user's bioelectric signals. This formulation incorporates label estimation for creating semi-supervised training labels during the decoder's normal operation, and online learning for updating the decoder parameters in a stable manner using those label estimates. The novel real-time adaptive decoding framework enables the adaptation to occur in real-time for each DoF independently. This thesis applied the adaptive decoding framework to the control of prosthetic hands using surface electromyograms. A short-term experiment on 10 intact-arm volunteer subjects evaluated the adaptive decoder's ability to quickly adapt to fatigue over 30 minutes, and a medium-term experiment recorded weekly datasets from 5 subjects over 6 weeks to determine the adaptive decoder's performance over a longer time span. In the short-term experiment, the adaptive decoder showed no significant performance degradation and significantly outperformed the non-adaptive decoder, which exhibited a 5% drop in success rate due to fatigue. In the medium-term experiment, the non-adaptive decoder dropped from 62% to 24% success rate over the 6 weeks, whereas the adaptive decoder showed a significantly slower rate of performance deterioration. Lastly, offline analyses showed that the DTW-based movement detection and the least-squares trajectory estimation methods developed in this thesis outperformed existing baselines by over 10% in terms of their label estimation error. In summary, the adaptive decoder presented in this work did not eliminate all performance deterioration, but it successfully slowed the rate of deterioration. Therefore, this adaptive framework could reduce the frequency of retraining for commercial prostheses, making such prostheses more usable and reducing the likelihood of their rejection.
License
Resource Type
Date Issued
Degree Level
Degree Name
Degree Field
Degree Grantor
Commencement Year
Advisor
Committee Member
Academic Affiliation
Rights Statement
Funding Statement (additional comments about funding)
  • This work was supported by the National Science Foundation through the Cyber-Human Systems Grants 1901492 and 1901236 and the Graduate Research Fellowship under Grant 1840998.
Publisher
Peer Reviewed
Language
Embargo reason
  • Pending Publication
Embargo date range
  • 2021-08-18 to 2022-09-18

Relationships

Parents:

This work has no parents.

In Collection:

Items

Thumbnail Title Date Uploaded Visibility Actions
file details GeorgeDamonA2021.pdf 2021-08-17 Public Download