Estimating Neural Modulation via Adaptive Point Process Method in Brain-machine Interface.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Brain-machine interfaces (BMIs) translate neural activity into device commands.
• Neurons exhibit changing activity patterns (neural tuning parameters) during BMI use.
• Existing adaptive algorithms struggle with abrupt parameter changes.

Purpose of the Study:

• To develop a robust method for estimating time-varying neural tuning parameters in BMIs.
• To address the limitations of local search in detecting abrupt neural plasticity.
• To improve the stability and performance of BMI decoding.

Main Methods:

• Proposed a global search strategy using adaptive point process estimation.
• Introduced a neural modulation parameter (range [0,1]) representing kinematic-neural similarity.
• Decoupled the preferred hyper-tuning direction using gradient descent.
• Applied the method to real-world BMI data, including a switch from manual to brain control.

Main Results:

• The proposed global search method demonstrated superior tracking of neural hyper-tuning parameters compared to local search.
• Effectively detected abrupt changes in neural tuning parameters during mode switching.
• Validated through the Kolmogorov-Smirnov (KS) statistical test.

Conclusions:

• The novel adaptive point process estimation with global search offers improved performance for BMI neural decoding.
• This method enhances the ability to track dynamic neural changes, crucial for stable BMI operation.
• Provides a more reliable approach to understanding and adapting to neural plasticity in BMIs.