Case Report: Representational geometry of multi-digit coordination in the macaque grasping network for intracortical BCIs

Introduction Intracortical brain-computer interfaces (BCIs) hold considerable promise for restoring dexterous hand function in individuals with severe paralysis or limb loss. However, reliable control of coordinated multi-digit movements remains challenging because the cortical representation of finger coordination is not fully understood. Whether this representation is shared across the cortical areas that contribute to grasping, or instead varies from one area to another, remains unclear. Case report We report the surgical implantation and neural recordings from a male rhesus macaque implanted with six floating microelectrode arrays (192 channels) across three nodes of the cortical grasping network: the anterior intraparietal area (AIP), ventral premotor cortex (area F5), and primary motor cortex (M1). The subject was trained to perform individual and combined finger flexions using a microswitch-based manipulandum. To our knowledge, this is the first study to record intracortical population activity simultaneously from these three cortical areas during coordinated multi-digit movements. Observations Population activity in AIP, F5, and M1 diverged into distinct, movement-specific trajectories. The number of independent activity patterns needed to capture most of these movement-related differences was larger in F5 than in AIP or M1. Alignment and reconstruction analyses additionally showed descriptive area-specific patterns in how closely combined-movement activity resembled a simple combination of the two single-finger activity patterns it comprises. This resemblance was generally limited across a range of dimensionalities. Error-trial analyses showed complementary within-area temporal patterns: early AIP activity preferentially reflected the instructed movement, whereas M1 activity near erroneous presses reflected the movement produced. Conclusion In this animal, condition-dependent dimensionality was significantly higher in F5 than in AIP or M1, while alignment and reconstruction analyses showed descriptive area-specific patterns in how closely combined-finger activity resembled a simple combination of single-finger activity. For brain-computer interfaces, these observations raise the hypothesis that whether single-finger training generalizes to combined presses may depend on which cortical area a decoder reads from, not only on how it is built. Shared low-dimensional structure could still support a small, easy-to-calibrate signal set in some areas, while area-specific structure could limit generalization in others. These possibilities require direct comparisons of decoder performance and calibration strategies across areas, as well as replication beyond this single subject.