Understanding and Addressing Variability Through Recalibration in a Brain–Computer Interface for Intelligent Vehicle Control Using Long-Term Implanted Human Neural Data: A Case Study

This work involved human subject in its research. Approval of all ethical and experimental procedures and protocols was granted by the FDA and IRB (NCT01997125). Even when trained on sufficient neural data, the performance of a brain-computer interface (BCI) is susceptible to error due to both systematic and random noise. Error from both sources may increase over time with changes in the neural signal or recording process. In practice, systematic error could be addressed through recalibration of model parameters, where the frequency of recalibration is based on the rate of performance loss. Random prediction error could be addressed by incorporating more observations into each prediction. However, recalibration incurs computational overhead and time delay; increasing the number of observations per prediction increases the latency from observation to prediction. The needs of the BCI application will determine whether the necessary frequency of recalibration or increased observations per prediction are valid configurations. This study evaluates recalibration as a mechanism to address accumulating systematic error in the unique context of vehicle driving, using neural data recorded previously in a driving simulation environment. A prior study demonstrated vehicle control but required some modifications to the paradigm to address variability, such as increasing the amount of data used to make each inference. In this study, offline analysis revealed substantial variation in the neural data recorded during the original experiments. Adding recalibration using labels assigned by the route navigation system improved classification accuracy from 55% to over 90% and reduced the latency of the classifier output. These findings demonstrate the potential of recalibration in BCI as a mitigation for neural variability, with application to vehicle control by individuals with tetraplegia.