Restoring Dexterous Hand Function with Artificial Neural Network-Based Brain-Computer Interfaces

点击翻译按钮获取中文摘要

Project Summary/Abstract Intracortical brain-computer interfaces (iBCIs) are promising solutions for restoring function to people with paralysis with orders of magnitude greater performance than their non-invasive analogs. Present iBCIs monitor the user’s brain signals and use a decoding algorithm to map the measurements from the brain directly to external variables, such as computer cursor velocity. People with tetraplegia have indicated that restoring hand function is their highest priority, however, current hand-focused iBCIs are unable to match the capabilities of the native human hand in terms of the number of independently-controlled fingers, the quality of the control, and the simultaneous use of the fingers with movements of the arm. These limitations hinder the widespread clinical deployment of hand-focused iBCIs. Recent studies have shown that much of the activity in motor cortex does not directly correspond to movement variables (like finger angles), but instead serves an internal, computational role to reliably generate motor outputs. Neural population dynamics, which are rules that govern the evolution of neural population activity over time, can be used to more accurately parse movement- and computation-related activity in motor cortex. Dynamics-based decoders first model the dynamics driving recorded neural activity, then use a decoder to map the estimated dynamics to movement. Dynamics-based decoding has already improved iBCI performance for predicting the arm movements of monkeys by 36%, but it remains unknown how well dynamics-based decoding can predict the movements of human fingers. The objective of this proposal is to restore dexterous finger control with an iBCI in people with paralysis. The central hypothesis is that dynamics-based decoders will bridge the gap in capabilities between hand-focused iBCIs and able-bodied hand function. The rationale for the proposed research is that the performance improvements introduced by dynamics-based decoders will translate from predicting arm movements to predicting finger movements. The hypothesis will be tested with people with upper extremity paralysis through the following two specific aims: 1) increasing the number of independently-controlled fingers of a robotic hand without sacrificing control quality, and 2) maintaining performance of controlling dexterous finger movements while simultaneously controlling movements of the entire robotic arm. The dynamics-based decoders will use state-of-the-art artificial neural networks (ANNs)-based dynamics models to achieve the best estimate of the underlying dynamics paired with ANN-based dynamics decoders to translate the estimated dynamics into movement. The dynamics-based decoders will be compared against direct decoders that have been traditionally used in human iBCIs. This work may be the first step toward providing people with paralysis a general-purpose iBCI-controlled robotic arm to assist them with independently completing activities of daily living at home.