Sampling representational plasticity of simple imagined movements across days enables long-term neuroprosthetic control

Cell, Journal Year: 2025, Volume and Issue: 188(5), P. 1208 - 1225.e32

Published: March 1, 2025

The nervous system needs to balance the stability of neural representations with plasticity. It is unclear what representational simple well-rehearsed actions is, particularly in humans, and their adaptability new contexts. Using an electrocorticography brain-computer interface (BCI) tetraplegic participants, we found that low-dimensional manifold relative distances for a repertoire imagined movements were remarkably stable. manifold's absolute location, however, demonstrated constrained day-to-day drift. Strikingly, statistics, especially variance, could be flexibly regulated increase during BCI control without somatotopic changes. Discernability strengthened practice was BCI-specific, demonstrating contextual specificity. Sampling plasticity drift across days subsequently uncovered meta-representational structure generalizable decision boundaries repertoire; this allowed long-term neuroprosthetic robotic arm hand reaching grasping. Our study offers insights into mesoscale statistics also enable complex control.

Language: Английский

---

Artificial neural network for brain-machine interface consistently produces more naturalistic finger movements than linear methods

bioRxiv (Cold Spring Harbor Laboratory), Journal Year: 2024, Volume and Issue: unknown

Published: March 5, 2024

Abstract Brain-machine interfaces (BMI) aim to restore function persons living with spinal cord injuries by ‘decoding’ neural signals into behavior. Recently, nonlinear BMI decoders have outperformed previous state-of-the-art linear decoders, but few studies investigated what specific improvements these approaches provide. In this study, we compare how temporally convolved feedforward networks (tcFNNs) and predict individuated finger movements in open closed-loop settings. We show that generate more naturalistic movements, producing distributions of velocities 85.3% closer true hand control than decoders. Addressing concerns may come inconsistent solutions, find regularization techniques improve the consistency tcFNN convergence 194.6%, along improving average performance, training speed. Finally, can leverage data from multiple task variations generalization. The results study methods produce potential for generalizing over less constrained tasks. Teaser A network decoder produces consistent shows real-world generalization through variations.

Language: Английский

---

Identifying Distinct Neural Features between the Initial and Corrective Phases of Precise Reaching Using AutoLFADS

Journal of Neuroscience, Journal Year: 2024, Volume and Issue: 44(20), P. e1224232024 - e1224232024

Published: March 27, 2024

Many initial movements require subsequent corrective movements, but how the motor cortex transitions to make corrections and similar encoding is unclear. In our study, we explored brain's signals both during a precision reaching task. We recorded large population of neurons from two male rhesus macaques across multiple sessions examine neural firing rates not only also movements. AutoLFADS, an autoencoder-based deep-learning model, was applied provide clearer picture neurons’ activity on individual sessions. Decoding reach velocity generalized poorly submovements. Unlike it challenging predict using traditional linear methods in single, global space. identified several locations space where submovements originated after reaches, signifying different than baseline before To improve movement decoding, demonstrate that state-dependent decoder incorporating at initiation correction improved performance, highlighting diverse features summary, show differences between encodes specific combinations position. These findings are inconsistent with assumptions correlations kinematic independent, emphasizing often fall short describing these processes for online

Language: Английский

---

Neural subspaces of imagined movements in parietal cortex remain stable over several years in humans

Journal of Neural Engineering, Journal Year: 2024, Volume and Issue: 21(4), P. 046059 - 046059

Published: Aug. 1, 2024

Abstract Objective. A crucial goal in brain–machine interfacing is the long-term stability of neural decoding performance, ideally without regular retraining. Long-term has only been previously demonstrated non-human primate experiments and primary sensorimotor cortices. Here we extend previous methods to determine humans by identifying aligning low-dimensional structures data. Approach. Over a period 1106 871 d respectively, two participants completed an imagined center-out reaching task. The longitudinal accuracy between all day pairs was assessed latent subspace alignment using principal components analysis canonical correlations multi-unit intracortical recordings different brain regions (Brodmann Area 5, Anterior Intraparietal junction postcentral intraparietal sulcus). Main results. We show stable representation activity subspaces from higher-order association areas humans. Significance. These results can be practically applied significantly expand longevity generalizability brain–computer interfaces. Clinical Trials NCT01849822, NCT01958086, NCT01964261

Language: Английский

---

Conserved structures of neural activity in sensorimotor cortex of freely moving rats allow cross-subject decoding

Nature Communications, Journal Year: 2022, Volume and Issue: 13(1)

Published: Dec. 2, 2022

Our knowledge about neuronal activity in the sensorimotor cortex relies primarily on stereotyped movements that are strictly controlled experimental settings. It remains unclear how results can be carried over to less constrained behavior like of freely moving subjects. Toward this goal, we developed a self-paced behavioral paradigm encouraged rats engage different movement types. We employed bilateral electrophysiological recordings across entire and simultaneous paw tracking. These techniques revealed coupling neurons with lateralization an anterior-posterior gradient from premotor primary sensory cortex. The structure population patterns was conserved animals despite severe under-sampling total number variations electrode positions individuals. demonstrated cross-subject cross-session generalization decoding task through alignments low-dimensional neural manifolds, providing evidence code.

Language: Английский

---

Neural Data Transformer 2: Multi-context Pretraining for Neural Spiking Activity

bioRxiv (Cold Spring Harbor Laboratory), Journal Year: 2023, Volume and Issue: unknown

Published: Sept. 22, 2023

Abstract The neural population spiking activity recorded by intracortical brain-computer interfaces (iBCIs) contain rich structure. Current models of such are largely prepared for individual experimental contexts, restricting data volume to that collectable within a single session and limiting the effectiveness deep networks (DNNs). purported challenge in aggregating is pervasiveness context-dependent shifts distributions. However, large scale unsupervised pretraining nature spans heterogeneous data, has proven be fundamental recipe successful representation learning across learning. We thus develop Neural Data Transformer 2 (NDT2), spatiotemporal activity, demonstrate can leverage motor BCI datasets span sessions, subjects, tasks. NDT2 enables rapid adaptation novel contexts downstream decoding tasks opens path deployment pretrained DNNs iBCI control. Code: https://github.com/joel99/context_general_bci

Language: Английский

---

Measuring instability in chronic human intracortical neural recordings towards stable, long-term brain-computer interfaces

Communications Biology, Journal Year: 2024, Volume and Issue: 7(1)

Published: Oct. 21, 2024

Language: Английский

---

Using adversarial networks to extend brain computer interface decoding accuracy over time

bioRxiv (Cold Spring Harbor Laboratory), Journal Year: 2022, Volume and Issue: unknown

Published: Aug. 26, 2022

Abstract Existing intracortical brain computer interfaces (iBCIs) transform neural activity into control signals capable of restoring movement to persons with paralysis. However, the accuracy “decoder” at heart iBCI typically degrades over time due turnover recorded neurons. To compensate, decoders can be recalibrated, but this requires user spend extra and effort provide necessary data, then learn new dynamics. As neurons change, one think underlying intent signal being expressed in changing coordinates. If a mapping computed between different coordinate systems, it may possible stabilize original decoder’s from behavior without recalibration. We previously proposed method based on Generalized Adversarial Networks (GANs), called “Adversarial Domain Adaptation Network” (ADAN), which aligns distributions latent within low-dimensional manifolds. ADAN was tested only very limited dataset. Here we propose Cycle-Consistent (Cycle-GAN), full-dimensional recordings. both Cycle-GAN data multiple monkeys behaviors compared them linear Procrustes Alignment axes provided by Factor Analysis (PAF). Both GAN-based methods outperformed PAF. (like PAF) are unsupervised require little making practical real life. Overall, had best performance easier train more robust than ADAN, ideal for stabilizing systems time. Significance Statement The inherent instabilities acquired microelectrode arrays cause an interface (iBCI) decoder drop time, as must essentially representing ever-changing system. Here, address problem using Generative (GANs) align these coordinates compare their success another, recently that uses alignment. Our fully unsupervised, trained quickly, remarkably data. These should give users access unchanging dynamics, need periodic supervised

Language: Английский

---

From monkeys to humans: observation-based EMG brain–computer interface decoders for humans with paralysis

Journal of Neural Engineering, Journal Year: 2023, Volume and Issue: 20(5), P. 056040 - 056040

Published: Oct. 1, 2023

. Intracortical brain-computer interfaces (iBCIs) aim to enable individuals with paralysis control the movement of virtual limbs and robotic arms. Because patients' prevents training a direct neural activity limb decoder, most iBCIs rely on 'observation-based' decoding in which patient watches moving cursor while mentally envisioning making movement. However, this reliance observed target motion for decoder development precludes its application prediction unobservable motor output like muscle activity. Here, we ask whether recordings from surrogate individual performing same as iBCI can be used an decoder.

Language: Английский

---

Increasing Robustness of Intracortical Brain-Computer Interfaces for Recording Condition Changes via Data Augmentation

Computer Methods and Programs in Biomedicine, Journal Year: 2024, Volume and Issue: 251, P. 108208 - 108208

Published: May 3, 2024

Language: Английский

---