Data were collected from 48 able-bodied human participants on four separate lab visits each using the protocol for the real-time, human-in-the-loop, invasive brain-computer interface model called the jaBCI (published model details at DOI 10.1088/1741-2552/ac97c3). Briefly, a subject wears a CyberGlove III that monitors the relative excursions of 19 different hand and finger joint angles. The same pre-trained artificial neural network used in the jaBCI validation study computes a set of emulated neural firing rates based on the preceeding 100ms of joint angle input data. To align with contemporary firing rate integration bin widths used in iBCI studies, we set the emulator to emit neural firing rates every 50 ms. At each time bin the subject's neural decoder uses the emulated neural activity to compute the velocity of the computer cursor and the graphical display is updated accordingly. The joint angle tracking, neural emulation, and decoding all run in real-time. To mimic the phenomenon of neuron turnover in iBCI studies we used four separate neural readout modules (one for each visit) that emulated 71, 45, 70, and 82 neurons, thus requiring subjects to recalibrate their decoders at the start of each visit. These data are collected as subjects attempt to pilot the computer cursor to displayed targets in the center-out task described in the accompanying publication. Therefore, the three low-level timeseries signals we record during closed-loop decoder use are 1) the subject's finger kinematics (in glove sensor units, 19 dimensional), 2) the corrisponding emulated neural firing rates (in average action potentials per second, c dimensional where c is the number of neurons in the given visit), 3) the state of the decoded computer cursor being controlled (a 4 dimensional vector of x position, y position, x velocity, and y velocity in arbitrary workspace units). These low level data are used to determine all subsequent behavioral outcomes, such as how long the subject took to reach the target or whether the subject hit the displayed target.

We designed the discrete direction selection (DDS) decoder for intracortical brain computer interface (iBCI) cursor control and showed that it outperformed currently used decoders in a human-operated real-time iBCI simulator and in monkey iBCI use. Unlike virtually all existing decoders that map between neural activity and continuous velocity commands, DDS uses neural activity to select among a small menu of preset cursor velocities. We compared closed-loop cursor control across four visits by each of 48 naïve, able-bodied human subjects using either DDS or one of three common continuous velocity decoders: direct regression with assist (an affine map from neural activity to cursor velocity), ReFIT, and the velocity Kalman Filter. DDS outperformed all three by a substantial margin. Subsequently, a monkey using an iBCI also had substantially better performance with DDS than with the Wiener filter decoder (direct regression decoder that includes time history). Discretizing the decoded velocity with DDS effectively traded high resolution velocity commands for less tortuous and lower noise trajectories, highlighting the potential benefits of simplifying online iBCI control.

Funding provided by: National Institute of Neurological Disorders and StrokeROR ID: https://ror.org/01s5ya894Award Number: R01NS109257