ResLAT: A Hybrid Neural Network for Continuous Hand Movement Decoding in Intracortical Brain-Computer Interfaces

Intracortical brain-computer interfaces (iBCIs) hold significant promise for restoring motor function to individuals with paralysis by translating neural activity into control signals for prosthetic devices. Despite substantial advancements, the movement performance of current prosthetic systems still falls far short of the flexibility and coordination of natural limbs. One of the current limiting factors is the decoding algorithm that efficiently and accurately converts neural signals into multi-degree-of-freedom, continuous motion control commands. To break through this bottleneck, we proposed and validated a hybrid neural network architecture combining Residual Networks, Long Short-Term Memory and multi-head self-attention mechanism (ResLAT) for decoding continuous motion trajectories of fingers in Crab-eating macaque. This decoder achieved an average decoding accuracy of 75.8% (48 days, about 7000 trials), and this work advances Intracortical brain-computer interfaces decoding and provides a new algorithmic foundation for more natural intuitive neural-prosthetic control systems and.