Hierarchical Dynamical Model for Multiple Cortical Neural Decoding.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Motor brain-machine interfaces (BMIs) translate brain activity into movement commands for prosthetics.
• The medial prefrontal cortex (mPFC) and primary motor cortex (M1) are crucial for reward-guided motor learning during BMI use.
• Existing Kalman decoding methods struggle with complex brain states and noisy, non-Gaussian neural data.

Purpose of the Study:

• To develop an improved BMI decoding algorithm by incorporating hierarchical brain structure and robust noise handling.
• To enhance movement decoding performance during the learning phase of BMI control.

Main Methods:

• Proposed a hierarchical model representing evolving brain states across multiple cortical areas (mPFC and M1).
• Integrated correntropy theory to address heavy-tailed, non-Gaussian noise in neural recordings.
• Tested the algorithm on in vivo recordings from rats learning a lever-pressing task.

Main Results:

• The hierarchical model demonstrated superior movement decoding performance compared to the classic Kalman filter.
• The algorithm effectively integrated past failed trial information from multisite recordings.
• Correntropy criterion successfully mitigated the impact of noisy, heavy-tailed neural data.

Conclusions:

• A hierarchical approach considering mPFC and M1 interactions improves BMI decoding during motor learning.
• The proposed method offers enhanced robustness against noise in neural recordings.
• This framework has the potential to advance BMI effectiveness for prosthetic control.