Towards a brain-computer interface for dexterous control of a multi-fingered prosthetic hand

Recent advances in Brain-Computer Interfaces (BCI) have enabled direct neural control of robotic and prosthetic devices. However, it remains unknown whether cortical signals can be decoded in real-time to replicate dexterous movements of individual fingers and the wrist. In this study, single unit activity from 115 task-related neurons in the primary motor cortex (M1) of a trained rhesus monkey were recorded, as it performed individuated movements of the fingers and wrist of the right hand. Virtual multi-unit ensembles, or voxels, were created by randomly selecting contiguous subpopulations of these neurons. Non-linear hierarchical filters using Artificial Neural Networks (ANNs) were designed to asynchronously decode the activity from multiple virtual ensembles, in real-time. The decoded output was then used to actuate individual fingers of a robotic hand. An average real-time decoding accuracy of greater than 95% was achieved with all neurons from randomly placed voxels containing 48 neurons, and up to 80% with as few as 25 neurons. These results suggest that dexterous control of individual digits and wrist of a prosthetic hand can be achieved by real-time decoding of neuronal ensembles from the M1 hand area in primates.