Stacked recurrent neural network for decoding of reaching movement using local field potentials and single-unit spikes

Decoding intended movement trajectory from neural activity is crucial for developing neuroprosthetic devices. In this study, we propose a processing framework to combine different information from two types of neural activities: action potentials (spikes) and local field potentials (LFPs). For this purpose, we proposed a stacked generalization approach based on recurrent neural network to enhance decoding accuracy of movement kinematics. We examined decoding performance of the proposed stacked recurrent neural network (SRNN) on decoding of reaching movement using intracortical datasets. The results show that the stacked generalization approach can enhance the decoding performance and can be used as an information fusion tool for multi-modal neuroprosthetic devices.