Methods for estimating neural step sequences in neural prosthetic applications

The prospect of helping disabled patients by translating neural activity from the brain into control signals for prosthetic devices is currently being realized. Initial proof-of-concept systems have demonstrated the need for faster and more accurate estimation algorithms, without requiring unrealistically many neurons. To address this need, we recently reported the plan-movement maximum likelihood (PMML) algorithm. It combines plan activity, specifying reach end-point, with movement activity, specifying instantaneous direction and speed of the arm movement, to yield more accurate estimates with fewer neurons. This approach could greatly benefit from an improved ability to track the switching of plan activity, which precedes movement onset, so that a more accurate plan estimate can be incorporated into movement decoding. In this paper, we propose a modified point-process filter, employing an adaptive parameter, that is capable of more accurately tracking constant plan periods and step changes than conventional methods. We also suggest that this algorithm is more attractive than an alternate maximum likelihood step tracking scheme. Ultimately, the adaptive algorithm is well-suited for use with the PMML algorithm, or for directly controlling prosthetic devices with plan activity, and should improve neural prosthetic system performance.