Detecting abrupt change in neuronal tuning via adaptive point process estimation.

Neuronal tuning property such as preferred direction and modulation depth could change gradually or abruptly in brain machine interface (BMI). The decoding performance will decay in static algorithms where dynamic neuronal tuning property is regarded as stationary. Many adaptive algorithms have been proposed to update the time-varying decoding parameter with main consideration on the decoding performance, but seldom focus on exploring how individual neuronal tuning property changes physiologically. We propose a novel adaptive algorithm based on sequential Monte Carlo point process estimation to capture the abrupt change of neuronal modulation depth and preferred direction. At each time point, the tuning parameter is assumed as static with a large probability and searched within a local area. Meanwhile, the abrupt change is thought to occur with a small probability and explored within a global range. This algorithm is tested on synthetic neural data and compared with a static point process algorithm. The results show that our adaptive algorithm succeeds in detecting the abrupt change in neuronal tuning, which contributes to a better reconstruction of kinematics.