Adaptive Spike-Based Kalman Filter for Neural Decoding in Brain-Machine Interface

Brain-Machine Interface (BMI) assists paralyzed patients in restoring motor functions by controlling neuropros-thetic devices through intentions captured from brain activities. However, traditional spike-based neural signal decoding algorithms typically employ spike counting to estimate firing rates and rely on static functional relationships between neural activi-ties and movements, overlooking the facts that all information in a spike sequence is contained within the spike intervals and neural tuning dynamically varies over time. To address these challenges, this study introduces a novel decoding algorithm that combines spike-based Kalman filter and sequential Monte Carlo point process methods. The algorithm directly utilizes spike signals as observations through spike-based Kalman filter, effectively reducing the loss of information caused by traditional statistical methods in neural information processing. The sequential Monte Carlo point process method dynamically updates the parameters to accommodate the changes of tuning function caused by neural degradation. In simulation experiments, our algorithm demonstrates superior performance compared to the traditional spike-based filters, especially under conditions involving dynamic changes in neural tuning, thus providing a robust solution for long-term reliable BMI applications.