Multiscale decoding for reliable brain-machine interface performance over time.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Invasive neural recordings for brain-machine interfaces (BMIs) can degrade over time, impacting electrode performance.
• Signal degradation in BMIs reduces control accuracy and limits clinical viability.
• Local field potentials (LFPs) offer a more robust signal modality compared to spikes, potentially improving BMI longevity.

Purpose of the Study:

• To develop and validate a multiscale decoder capable of integrating spike and LFP data for improved BMI performance.
• To assess the decoder's ability to model distinct statistical properties and time-scales of spike and LFP activity.
• To evaluate the decoder's effectiveness in decoding upper-arm joint angles during a complex motor task.

Main Methods:

• Developed a multiscale decoder to simultaneously model discrete spike data and continuous LFP data.
• Implemented multi-time-scale processing (milliseconds for spikes, tens of milliseconds for LFPs).
• Validated the decoder using motor cortical spike/LFP recordings from a non-human primate performing a 3D reach-to-grasp task, decoding 7 upper-arm joint angles.

Main Results:

• The multiscale decoder improved decoding accuracy by incorporating LFP information alongside spike data.
• Enhanced decoding performance was achieved while maintaining the millisecond time-scale relevant for spike activity.
• Significant improvements were observed even with a limited number of LFP channels, indicating the method's robustness.

Conclusions:

• Multiscale decoders integrating spike and LFP data can enhance the accuracy and reliability of neural decoding for BMIs.
• This approach offers a promising strategy to improve the longevity and clinical applicability of brain-machine interfaces.
• The robust performance with few LFP channels suggests practical advantages for future BMI development.