Multi-finger movement representations in human premotor cortex exhibit pseudo-linear summation and a specific neural geometry, informing fine motor BCI decoding.1
This work provides neural geometry insights for dexterous neuroprosthetic hand control.11
Deep research (2026-07-20)
Source note: the exact bioRxiv URL could not be retrieved; findings below use the peer-reviewed version — Shah NP, Avansino D, Kamdar F, et al. “Pseudo-linear summation explains neural geometry of multi-finger movements in human premotor cortex,” Nat Commun. 2025 May 30;16(1):5008 (originally bioRxiv 2023.10.11.561982).23
Authors/institutions: Nishal P. Shah (Rice University; formerly Stanford), Donald Avansino, Foram Kamdar, Francis R. Willett, Jaimie M. Henderson (Stanford Neurosurgery, HHMI, Wu Tsai), Leigh R. Hochberg (Mass General/Harvard, BrainGate, Brown, Providence VA), Carlos Vargas-Irwin (Brown), Chethan Pandarinath (Emory/Georgia Tech), Krishna V. Shenoy (Stanford, HHMI); NIH-funded (R01 DC009899, DP2 NS127291, UH2 NS095548, U01 NS098968, U01 DC017844, R01 DC014034).2
Intracortical multi-electrode arrays (BrainGate-style, implanted in motor/premotor cortex) recorded from n=2 male participants with tetraplegia while they attempted single, pairwise, and higher-order finger movements.2
Neural activity for simultaneous multi-finger movements closely followed linear summation of the corresponding single-finger patterns, but with two systematic, reproducible deviations — hence “pseudo-linear” rather than pure linear.42
Deviation 1 (normalization): population activity magnitude did not grow proportionally with the number of simultaneously moving fingers — overall neural gain saturated rather than summing without bound.42
Deviation 2 (tuning-direction shift): the neural tuning direction of weakly represented fingers (e.g., middle finger) changed significantly when other, more strongly represented fingers moved simultaneously.24
These nonlinearities caused nonlinear decoding methods to outperform purely linear decoders for classifying/reconstructing simultaneous finger movements — direct evidence that pseudo-linear (not linear) models better capture premotor neural geometry.34
Exact single-vs-combined decoding accuracy percentages, trial counts, and dimensionality/variance-explained values were not reported in the retrieved abstract; full numeric tables are in the paper body — not reported.
Author lineage: builds directly on the BrainGate consortium’s intracortical BCI-for-paralysis work (Hochberg, Henderson, Shenoy labs) and prior Stanford population-geometry decoding.2
Clinical/translational implication: encoding finger combinations via a structured, near-linear-but-normalized code (rather than fully independent or fully linear) informs decoder design for dexterous multi-finger neuroprosthetic hands, suggesting nonlinear calibration layers are needed to decode natural hand postures rather than single-DOF classifiers.
Competing-interest disclosures note several co-authors consult for or hold IP licensed to BCI companies (Neuralink, Synchron, Blackrock Neurotech, Paradromics, Reach Neuro, Axoft, Precision Neuroscience), reflecting close industry-academia ties in the implanted-BCI field.3
Related work
Willett FR, et al. BrainGate high-performance handwriting/speech BCI decoding (Nature, 2021/2023) — same electrode/participant infrastructure lineage.
Vargas-Irwin CE, et al. multi-finger and hand-kinematic decoding from motor cortex (Brown University/Providence VA).
Abramovich Krasa B, et al. “Premotor cortex uses a compositional neural geometry…” bioRxiv, 2026.5
Neural Noise