Co-adaptive Kalman filtering in a naïve rat cortical control task

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Current prosthetic control relies on mapping brain activity to measured limb movements, which is not feasible for individuals with severe mobility impairments.
• Developing non-invasive or less physically demanding methods for prosthetic control is crucial for broader clinical application.

Purpose of the Study:

• To explore an alternative approach for controlling prosthetic devices using adaptive decoding filters that adjust to neural activity patterns.
• To investigate if subjects can learn prosthesis control through neural modulation and adaptive filtering without requiring direct measurement of limb movements.

Main Methods:

• Utilized a rat model to train subjects in prosthesis control using an adaptive decoding filter.
• Employed a modified Kalman filter that "co-adapts" to neural modulation patterns during expected prosthetic device movement.
• Provided task-relevant feedback to guide the animal in optimizing its response strategy for reward maximization.

Main Results:

• One subject demonstrated consistent performance above chance levels after only 2 days (4 sessions) of training.
• The adaptive decoding filter successfully adjusted to neural modulation patterns, enabling prosthesis control.
• The methodology integrated learning strategies with online filter adaptation for effective control.

Conclusions:

• Adaptive decoding filters offer a viable alternative for prosthetic control, particularly for individuals unable to perform measured movements.
• This approach, combining learning and adaptive filtering, shows promise for developing intuitive and effective neuroprosthetic systems.
• The study highlights the potential for closed-loop systems that facilitate learning and optimize control strategies based on neural feedback.