End-to-End Hand Kinematic Decoding from LFPs Using Temporal Convolutional Network
Nur Ahmadi, Timothy G. Constandinou, Christos-Savvas Bouganis · 2019
In recent years, local field potentials (LFPs) have emerged as a promising alternative input signal for brain-machine interfaces (BMIs). Several studies have demonstrated that LFP-based BMIs could provide long-term recording stability and, at the same time, comparable decoding performance to their spike counterparts. However, despite the compelling results, most LFP-based BMIs still make use of hand-crafted features which is a time-consuming process and can be suboptimal. In this paper, we propose an end-to-end system approach based on temporal convolutional network (TCN) to automatically extract features and decode kinematics of hand movements directly from raw LFP signals. We benchmark its decoding performance against traditional approaches incorporating long short-term memory (LSTM) decoders driven by hand-crafted LFP features. Experimental results demonstrate significant performance improvement of the proposed approach compared to the traditional approaches, demonstrating the suitability and the potential of TCN-based end-to-end systems in providing stable and high decoding performance LFP-based BMIs.