Estimation of speed and direction of arm movements from M1 activity using a nonlinear neural decoder

Abstract

The current neural decoding algorithms for brain-machine interfaces (BMIs) have largely focused on predicting the velocity of arm movements from neuronal ensemble activity. Yet, mounting evidence indicates that velocity is encoded separately in motor cortical activity. In this regard, we aimed to decode separate speed and direction information independently using a machine learning algorithm based on long short-Term memory (LSTM). The performance of the proposed decoder was compared with the traditional decodres using velocity Kalman filter and the velocity LSTM. The proposed decoder showed better angular prediction than the other decoders. Also, the reconstruction hand trajectories with the proposed decoder acquired the targets more often. Movement time of the reconstructed trajectories by the proposed decoder was shorter than the others. Our results suggest advantages of decoding speed and direction independently using a nonlinear model such as LSTM for intracortical BMIs.