Knowledge Vault 3/53 - G.TEC BCI & Neurotechnology Spring School 2024 - Day 5
Non-invasive neuro-interfaces for interacting with robotics exoskeletons
Jose Maria Azorin, Universidad Miguel Hernández de Elche, Alicante (ES)
<Resume Image >

Concept Graph & Resume using Claude 3 Opus | Chat GPT4 | Llama 3:

graph LR classDef exoskeletons fill:#f9d4d4, font-weight:bold, font-size:14px; classDef rehabilitation fill:#d4f9d4, font-weight:bold, font-size:14px; classDef bci fill:#d4d4f9, font-weight:bold, font-size:14px; classDef results fill:#f9f9d4, font-weight:bold, font-size:14px; classDef future fill:#f9d4f9, font-weight:bold, font-size:14px; A[Jose Maria Azorin] --> B[BCIs control robotic exoskeletons,
aid disabled walking. 1] A --> C[Stroke, spinal cord injuries
impair walking. 2] C --> D[Traditional rehab: bottom-up,
therapists move legs. 3] B --> E[Exoskeletons apply forces,
still bottom-up approach. 4] B --> F[BCIs bypass damage, decode
brain to control exoskeleton. 5] F --> G[WALK project developed robust
BCI exoskeleton control. 6] G --> H[WALK partnered with hospitals,
University of Houston. 7] F --> I[WALK used motor imagery,
attention for walking intent. 8] G --> J[Results: BCI enabled exoskeleton walking. 9] J --> K[Challenges: EEG differences, conservative
control causing stops. 10] G --> L[VR training reduced exoskeleton
training time. 11] L --> M[VR group had better
motion accuracy. 12] G --> N[Usability tests: patients satisfied,
sub-maximal exertion. 13] A --> O[Regate project: BCI and
spinal cord stimulation. 14] O --> P[Regate: asynchronous BCI,
only 'walk'/'stop' thoughts. 15] O --> Q[Deep learning trained static,
motion models. 16] Q --> R[Fine-tuning all layers achieved
54% static/motion accuracy. 17] O --> S[Closed-loop: reliable start,
varied stop accuracy. 18] S --> T[Unintended stops important,
best ratio 27%. 19] O --> U[Error potential detection may
improve BCI algorithm. 20] G --> V[Videos showed successful BCI,
live patient demo. 21] A --> W[Conclusions: Patients control exoskeletons
via EEG BCIs. 22] W --> X[Next: clinical trials to
validate BCI rehab benefits. 23] A --> Y[Presenter acknowledged team's
hard work. 24] A --> Z[Impressive progress using non-invasive
BCIs for paralysis. 25] Z --> AA[Further development could improve
mobility, quality of life. 26] class B,E exoskeletons; class C,D,N,X,Z,AA rehabilitation; class F,G,I,O,P,Q,T,U,V,W bci; class J,K,L,M,R,S results; class H,Y future;


1.- Jose Maria Azorin discusses brain-computer interfaces (BCIs) for controlling robotic exoskeletons to help people with disabilities walk.

2.- Stroke and spinal cord injuries cause walking difficulties. Traditional rehab uses a bottom-up approach with therapists moving patients' legs.

3.- Exoskeletons with actuators can apply forces to help patients walk, but still use a bottom-up rehabilitation approach.

4.- BCIs can bypass damaged pathways, decoding brain signals to control exoskeletons. This combines bottom-up and top-down approaches for stronger neuroplasticity.

5.- The WALK project (2019-2020) developed a robust BCI to control an exoskeleton with basic commands like walk, stop, turn, speed up/down.

6.- WALK collaborated with hospitals and University of Houston. Open-loop training adjusted static and motion models. Closed-loop tests evaluated real-time control.

7.- WALK used motor imagery and attention level to detect walking intent. 27 EEG electrodes recorded signals, processed to control the exoskeleton.

8.- Results showed the person could use the BCI to walk with the exoskeleton. Accuracy improved over multiple sessions.

9.- Challenges included accounting for EEG differences between static and walking states, and conservative control sometimes stopping the exoskeleton unintentionally.

10.- A follow-up project introduced virtual reality training to reduce exoskeleton training time without negatively impacting closed-loop control performance.

11.- The VR training group achieved similar static accuracy and better motion accuracy compared to the non-VR control group.

12.- Usability tests with spinal cord injury patients found they were satisfied with the BCI-exoskeleton performance. Exertion increased but remained sub-maximal.

13.- The Regate project (2022) is developing a new rehab approach combining EEG-based BCIs and spinal cord stimulation.

14.- Regate uses an asynchronous BCI where the user only needs to think "walk" to start and "stop" to stop, not continuously imagine walking.

15.- Static and motion models were trained using deep learning (EEGNet) on data from initial sessions, then fine-tuned in later sessions.

16.- Fine-tuning all layers of the deep learning models achieved the best accuracy - 54% for static and motion.

17.- In closed-loop tests, users could reliably start walking with the exoskeleton. Stopping accuracy varied between users (35-90%).

18.- Avoiding unintended stops is important. The best unintended stop ratio was 27%, but some users had higher ratios.

19.- Current work aims to improve results by incorporating error potential detection in the BCI algorithm.

20.- Videos demonstrated successful BCI operation, including a live demo with a patient in front of the press, despite distractions.

21.- Key conclusions: Patients can control exoskeletons using EEG-based BCIs. This is likely beneficial, but clinical trials are still needed.

22.- Next steps are conducting clinical trials with many patients over extended periods to validate the rehab benefits of the BCI approach.

23.- The presenter acknowledged the hard work of his lab team members who are advancing the research he presented.

24.- Overall, the research demonstrates impressive progress in using non-invasive BCIs to restore walking in individuals with paralysis or motor impairments.

25.- Further development of this technology could provide a promising new rehabilitation option to improve patients' mobility and quality of life.

Knowledge Vault built byDavid Vivancos 2024