Improved Brain-Computer Interface Decoding for Activities of Daily Life

ABSTRACT The development of brain-controlled prosthetic arms promises to provide independence to people with paralysis. To date, however, Brain-Computer Interfaces (BCIs) have not conferred on users the ability to use the prosthesis to carry out activities of daily living (ADLs) with adequate reliability and flexibility. This inability can be traced back to at least three shortcomings. First, while we naturally closely coordinate arm and hand movements, current BCI users reach and grasp sequentially, in large part due to the way BCI decoders are built. Second, existing decoders use the component of the neuronal activity that has a direct and immediate relationship with motor output to infer motor intent. While this approach has been successful even for control of an anthropomorphic robotic arm and hand, it does not harness all the behaviorally relevant M1 activity. Indeed, activity that has a direct and immediate relationship with behavior – the so-called output-potent activity – constitutes only a small fraction of the total M1 activity. The remaining neuronal activity – so-called output-null activity – plays a role in generating the output-potent activity but is overlooked by standard decoding approaches. Third, while robotic hands have become increasingly sophisticated and anthropomorphic, no existing prototype approaches the functionality of a human hand, either in terms of actuation or sensorization. The goal of the proposed project is to address each of the aforementioned limitations by building more biomimetic decoders – that allow for coordinated arm and hand movements and more effectively harness M1 activity – and by challenging them in a flexible and realistic virtual reality platform. First, we will build decoding approaches that support coordinated movements of the arm and hand. To this end, we will train decoders while subjects reach to and grasp objects that differ in shape, size, and orientation, forcing significant hand orienting and pre-shaping during reaching. Second, we will further elaborate these decoders so that they leverage both output-potent and output-null activity. To this end, we will leverage recent insights into M1 dynamics and their relationship to behavior to build decoders that harness all the behaviorally relevant activity in M1. Finally, we will test novel decoders in VR by having subjects perform standard tests of arm and hand function as well as tasks that mimic complex activities of daily living and develop performance metrics for these VR scenarios. We are well positioned to achieve these objectives as part of a multi-site clinical trial on BCI with 3 subjects implanted across two locations, with existing funding for two more subjects.