Author(s) :

Riddhi Bhashkar, Omar Tawakol.

Volume/Issue :

Abstract :

Brain-computer interfaces (BCIs) allow for decoding neural signals associated with intended movements and converting that to commands that can control external hardware, including prosthetics. This can help individuals with paralysis who lack the ability to voluntarily control their limb movements. This study aims to investigate decoding the direction of arm movement from single-neuron non-human primate recordings obtained during a center-out reaching task. The intended direction of movement was decoded from a population of directionally tuned neurons using two decoding algorithms: Long Short-Term Memory (LSTM) and another algorithm known as the Poisson State Space Model (POSSM). LSTM learns spike patterns directly from the data, while POSSM uses a framework specifically tuned to neuronal firing dynamics. The results demonstrated that the POSSM achieved relatively higher decoding accuracy when compared to LSTM, suggesting that the neural activity in the dataset follows patterns suited to the Poisson model’s assumptions. POSSM is a relatively novel algorithm, and this manuscript represents an early attempt to adopt it in a motor decoding task. The results suggest that it outperforms LSTM. These findings may inform the design of more accurate and efficient brain–computer interface systems and support future advancements in motor rehabilitation strategies.