Encoder-decoder optimization for brain-computer interfaces.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Computer Science

Background:

• Neuroprosthetic brain-computer interfaces (BCIs) decode neural activity for effector control.
• Both user learning and decoder adaptation can improve BCI performance.
• Co-adaptation involves simultaneous user learning and decoder adaptation to neural patterns.

Purpose of the Study:

• To provide a mathematical framework for co-adaptation in BCIs.
• To relate co-adaptation to the joint optimization of user encoding models and decoder parameters.
• To demonstrate an approach for optimizing decoders and user learning for enhanced BCI performance.

Main Methods:

• Developed a mathematical framework for co-adaptation.
• Defined co-adaptation as joint optimization of user's control scheme and decoder parameters.
• Used numerical methods to derive an optimized decoder.
• Validated the approach using a model BCI system with an online prosthesis simulator and a human-in-the-loop psychophysics setup.

Main Results:

• Co-adaptation does not outperform an optimal fixed decoder coupled with optimal user learning under specific framework assumptions.
• Users can learn control schemes (encoders) matched to fixed, optimal decoders.
• The proposed approach demonstrated expected performance advantages in simulations.

Conclusions:

• Optimal user learning with a fixed, optimized decoder can achieve performance comparable to co-adaptation.
• The study validates that users can adapt their control strategies to optimal decoders.
• The developed methods offer a pathway to enhance BCI system performance through optimized decoder selection and user training.