Multiscale brain-machine interface decoders.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Brain-machine interfaces (BMI) traditionally use single neural signal scales (spikes or ECoG/LFP).
• Simultaneous multi-scale neural recordings (spikes, LFP, ECoG) present challenges in joint modeling and decoding due to differing data types and timescales.

Purpose of the Study:

• To develop a novel multiscale modeling and decoding framework for integrating diverse neural activity scales.
• To improve the accuracy and performance of BMI control by jointly decoding spikes, LFP, and ECoG signals.

Main Methods:

• Developed a multiscale decoder capable of processing discrete (spikes) and continuous (LFP/ECoG) neural data.
• Ensured the decoder operates at the fast millisecond timescale of spikes, while incorporating slower ECoG/LFP signals.
• The framework specializes into Kalman Filter (KF) or Point Process Filter (PPF) when only continuous or spike data is available, respectively.

Main Results:

• The multiscale decoder significantly enhances BMI control accuracy and error performance compared to decoding single signal types alone.
• Simulations demonstrated the decoder's effectiveness in closed-loop BMI scenarios.
• The decoder successfully integrates information across different neural signal scales and timescales.

Conclusions:

• The proposed multiscale modeling and decoding framework offers a significant advancement for BMI technology.
• This approach has the potential to unlock improved BMI control by leveraging simultaneous multiscale neural activity.
• Future BMI systems can benefit from this integrated decoding strategy.