Jose Carmena - US grants

Information theory has been used as an organizing principle in neuroscience for several decades. Estimates of the mutual information between signals acquired in neurophysiological experiments are believed to yield insights into the structure of the underlying signal processing architectures. The experimental side of the field has undergone an impressive paradigmatic shift over the last few years, including simultaneous recordings from neuron populations and brain-machine interfaces (BMIs). However, on the theoretical side, only minor modifications to the theory were made. This project aims at transforming the theoretical backdrop of systems neuroscience in general and neural ensemble electrophysiology in particular. To this end, it levies three recent and emerging theoretical advances in the network sciences: (1) the tremendous recent progress concerning the behavior and information-theoretic performance limits of communication networks, fueled by the resounding success of wireless communication; (2) we have recently developed a new information theory of joint computation/communication, showing that for networks with interference, joint computation/communication can be executed much more efficiently - an exponential energy savings as a function of the number of agents involved in the computation - than separate computation and communication (as is classically done in communication systems). Our hypothesis is that such a significant gap must have repercussions in neural systems, and we develop new information measures to look for places where indeed joint computation/communication could be a crucial component of the neural signal processing. Finally, (3), we exploit a new information-theoretic end-to-end perspective on communication systems. All of these methods are applicable only thanks to new emerging data sets pertaining to chronic, simultaneous recordings from large populations of neurons across multiple brain regions. The further acquisition and development of these data sets is therefore an integral part of this research project.

This project introduces a set of new information measures that are truly multivariate, and thus a perfect fit for the emerging multi-neuron data sets, and particularly for BMI. This involves both natural extensions of existing information and redundancy measures as well as a new computational information measure that assesses not just the quality of information transfer, but also the computational potential offered by a population of neurons. The project aims to develop a new set of theories, techniques, and methods for dealing with and understanding multi-neuron recordings across spatiotemporal dimensions. If successful, this will provide ideas for new experiments and sharpen our understanding of the connections between communication and computation.

0954243
Carmena

Brain-machine interface technology has the potential of improving the quality of life for millions of patients suffering from paralysis due to lesions of the central nervous system or other neurological disorders. Our previous findings showed that monkeys can learn to reach and grasp virtual objects by controlling a robot arm through a brain-machine interface using visual feedback, even in the absence of overt arm movements. This proposal aims at bringing the field one step closer towards the ultimate control of neuroprosthetic devices through an effortless recall of a motor memory in a manner that mimics the natural process of skill acquisition and motor control. This goal will be pursued through the following specific aims: 1) To investigate the formation and stabilization of a prosthetic motor memory; 2) To investigate the neuron-behavior relationship for prosthetic function.

This project will establish a scientific basis for understanding how the brain controls movement of disembodied devices, and will drive the development of the next generation of neural prosthetics that will restore motor function in millions of neurologically impaired patients. The strong educational component of this proposal relies on brain-machine interfaces as an ideal platform for interdisciplinary education in science and engineering. The proposed efforts aim to address the nations current talent shortage of science and engineering majors which could have a severe negative impact on economic growth. This will be pursued through a mentoring program aimed at increasing the number of underrepresented and women students entering careers in engineering. The program will also involve undergraduate and graduate students, academics, and the industrial community in brain-machine interface research through teaching, collaborations, data sharing, workshops and tutorials.

This interdisciplinary research proposal brings together leaders in neurophysiology and brain-machine interfaces (BMI), control systems, and exoskeleton design to significantly advance our understanding of fundamental principles in the neural control of movement in scenarios that involve physical interactions with the world. Furthermore, this work will transform neuroprosthetic systems to improve the quality of life for a large number of neurological patients. The central question that motivates this proposal is: Does the brain use motor programs to help it control a highly redundant multi-degree of freedom (DOF) biomechanical plant such as the arm? To answer this question, this project will conduct a series of experiments that require a combination of three major innovations at the experimental (BMI), theoretical (hybrid control), and technical (exoskeleton design) level. This proposal aims to synthesize all three innovations into a new experimental paradigm unifying brain, biomechanics, and behavior. Specifically, visually-cued motor plans in the motor cortices of macaque monkeys will be: 1) read by a BMI, 2) interpreted by a hybrid controller and a musculoskeletal model, and 3) translated into appropriate movements and stiffness in a multi-DOF upper-limb exoskeleton.

Intellectual Merit

At a societal level, this proposal seeks to create a significant advancement in neuroprosthetic systems to improve the quality of life of patients suffering from paralysis due to lesions of the central nervous system or other neurological disorders. BMIs will make a great impact in the quality of life for neurological patients by providing reliable performance when interacting with real objects and in real-world scenarios. This proposal also informs motor systems neuroscience by proposing a novel framework to study how neural ensembles can learn to control a multi-DOF exoskeleton by volitional modulation of neural activity in real-world tasks. It provides a critical link between neural events and real-world dynamics through a novel hierarchical distributed control scheme for hybrid systems identification and control that captures the continuous time evolution of the arm/exoskeleton, as well as the dynamically changing sequence of tasks. No motor task of this complexity has ever been demonstrated in a BMI system. The potential impact of the proposed work is immense. If successful, this work will transform our understanding about how the brain controls movement, and will introduce a paradigm shift in the development of the next generation of neural prosthetics that will restore motor function in millions of neurologically impaired patients ? a development which may very well impact other domains such as human-machine interaction and innovative user interfaces.

Broader Impact
The dissemination of data and research findings from this project will be done through a project website. Our codes will be open source, and the methods, algorithms and results will be available through publications in the fields of neuroscience, control and robotics. Through the Center for Information Technology Research in the Interest of Society (CITRIS) at Berkeley, research results from this program will be immediately accessible and distributed to a large engineering community, in particular other interdisciplinary research groups. Also, the findings in the proposed research will be outreached to K-12 students and their parents through various activities at UC Berkeley and Lawrence Hall of Science (LHS), such as CAL-day. In addition, research findings will be disseminated through workshops at the main annual conferences in neuroscience, robotics, control, and biomedical engineering. The educational component of this proposal relies on BMI as a platform for interdisciplinary education in science and engineering. New graduate-level courses will be introduced and existing courses will be enhanced based on results of this project. The proposed efforts will open doors to student rotations between neuroscientists, control theorists and roboticists. In addition to postdocs and graduate students, undergraduate students will participate in all aspects of the project: modeling, analysis, simulation, prototype development and experimentation. Special effort will be placed on the recruitment of individuals from underrepresented groups including women, thereby building on the strong record the PIs have in this area.

Proposal No:1551239, EAGER-Neural dust stimulation for closed loop neuromodulation

One of the most important challenges that remains in neuroengineering is the development and demonstration of a clinically viable neural interface which can both record from and stimulate many individual neurons and lasts a lifetime. These chronic or long-term neural interfaces are of increasing interest for both central (CNS) and peripheral nervous system (PNS) interventions. Creating lasting, durable, untethered interfaces raises a variety of issues, ranging from the nature of the physical substrate (avoiding the biotic and abiotic effects that presumably lead to performance degradation at the electrode-tissue interface, the density and spatial coverage of the sensing sites), the type of signals measured, and the computation and communication capabilities (how much signal processing on-chip data to transmit wirelessly) under the power budget of the whole system. This proposal seeks to extend our recently published Neural Dust platform to allow for stimulation of nerves via neural dust motes. We believe this to be an aggressive vision which would open the door to a vast array of interventions, including untethered neural recording of human nerves and neurons, untethered stimulation of these processes and record-and-stimulate closed loop systems. Such a vision will require a number of fundamental technological innovations that will have impact across domains including basic neuroscience, clinical interventions of neurological disorders, and prosthetics. For example, the ability to precisely monitor and modulate peripheral nerve activity with a minimally invasive medical device would enable a wide-range of therapeutic opportunities. This closed-loop neuromodulation cannot be done with existing technologies because they suffer from one of two major drawbacks: lack of spatial resolution or high degree of invasiveness.

We recently proposed an ultra-miniature as well as extremely compliant system that could enable massive scaling in the number of recordings from the brain or the peripheral nervous system, providing a path towards truly chronic BMI. At the core of this vision is a platform for powering, receiving and transmitting information from inside a peripheral nerve to outside the body using aggressive, state-of-the-art circuit design and the recent demonstration of ultrasonic, piezocrystal "neural dust? motes. The work envisioned in this proposal will leverage recent application specific integrated circuit (ASIC) technology to build stimulating motes that can address individual neurons (or peripheral fibers) and will demonstrate untethered stimulation of nerve fibers, paving the way to closed-loop record-and-stim technology using neural dust. This is a very aggressive, high risk direction which leverages existing neural dust developments with a very high potential payoff (as it enables untethered closed-loop neuromodulation systems). Our long term vision is a system capable of recording and stimulation in closed-loop.

This research program aims to elucidate the role of the ipsilateral hemisphere in motor control, entailing parallel studies of upper limb movement in monkey and human. The outcome of the proposed work has the potential to guide the development of cortically-controlled neuroprosthetic systems for patients with neurological disorders by recording from humans and monkeys at different spatial levels of functional neural organization. Previous work from our group has demonstrated that distributed activity in primate motor areas is reliably correlated with ipsilateral upper limb kinematics in monkey and human. This information can be decoded by applying linear methods to neural signals at a variety of temporal and spatial scales, and can be successfully incorporated into a closed-loop BMI system. However, the functional contribution of ipsilateral motor cortex to limb movement remains unclear. Is ipsilateral control limited to proximal muscles or are these signals also relevant for the control of distal muscles? How do ipsilateral representations change in different task contexts, especially when considering contexts in which the two hands are either used independently or in a coordinated manner? This research program will address these fundamental questions. We outline four key hypotheses: H1) Activity in motor cortex provides additional control signals for ipsilateral movements, independent of the activity dedicated to contralateral movements. H2) Movements of one limb transcallosally activate homologous circuits in the ipsilateral motor cortex, activity that might facilitate mirror-symmetric movement or provide a mechanism for generalized motor learning. H3) Activity in ipsilateral motor cortex reflects the simultaneous, bilateral preparation of unimanual movements. H4) Ipsilateral motor activity is related to the control of the contralateral hand, but is modulated by the degree of ipsilateral movements as part of a network required for bimanual coordination. To evaluate these hypotheses, this proposal is structured around two aims, each entailing a series of experiments. The specific aims in both monkey and man are: 1) to characterize the role of ipsilateral motor cortex during unimanual movement and 2) to characterize the role of ipsilateral motor cortex during bimanual movements. Parallel neurophysiological studies will be conducted in human and non-human primates to record and analyze neural activity at different levels of integration. State-of- the-art methods for complementary neural recordings (human electrocorticogram, monkey local field potential and single unit activity) and data analysis will be employed to address the four hypotheses.

SUMMARY We propose an ultra-miniature as well as extremely compliant system that enables massive scaling in the number of neural recordings from the brain while providing a path towards large-scale neural recordings and truly chronic brain-machine interfaces (BMI). This will be achieved via two fundamental technology innovations: 1) 10 ? 100 ?m scale, free-floating, independent sensor nodes, or neural dust, that detect and report local extracellular electrophysiological data, and 2) a subcranial mm-scale interrogators that establish power and communication links with the neural dust. The interrogator array is placed beneath the skull and below the dura mater, to avoid strong attenuation of ultrasound by bone and is powered by an external transceiver via EM energy transfer. Building on an initial theoretical treatment and in-vitro validation regarding the feasibility of power coupling and backscatter communication at these scales within the brain, and more recent in-vivo experimental data showing high-fidelity transmission of electromyogram (EMG) and electroneurogram (ENG) signals from anesthetized rats, this work will provide the first in vivo demonstration that this type of recording modality is possible in the central nervous system. In the process, we aim to map the fundamental system design trade-offs and ultimate size, power, and bandwidth scaling limits of neural recording systems at the cortical level, built from low-power electronics coupled with ultrasonic power delivery and backscatter communication. The use of distributed, ultrasonic backscattering systems to record high frequency (~kHz) neural activity would pave the way for both massive scaling in the number of neural recordings from the nervous system as well as truly chronic BMIs.

PROJECT SUMMARY This proposal aims to elucidate the computational and neural basis of neuroprosthetic skill learning by leveraging recent advances in the science and engineering of closed-loop brain-machine interfacing. The outcome of the proposed work has the potential to guide the development of the next generation of neurophysiologically-informed, cortically-controlled neuroprosthetic systems for patients with neurological disorders. State-of-the-art brain-machine interfaces (BMIs) leverage machine learning to rapidly calibrate to the neural activity of individuals, but performance also benefits from subjects learning to reliably produce desired neural activity patterns. The basic science and engineering principles of designing such a ?2-learner BMI? in which the brain and machine synergistically learn are not well understood. Hence, this proposal aims to investigate how the brain learns when the machine undergoes different degrees of learning, how different degrees of brain learning affect long-term BMI performance, robustness, and generalization, and how these principles can guide the design of a 2-learner BMI system which facilitates brain learning. The proposal is structured in three aims: 1) To study the impact of decoder adaptation on the development of neural encoding models underlying neuroprosthetic skill; 2) To Study how decoder adaptation and resultant neural encoding model influences BMI performance with perturbations (robustness) and BMI performance on unpracticed tasks (generalization); and 3) Design and validation of the next-generation Flexible 2-Learner Decoder architecture. The analyses and experiments proposed in these aims will leverage the fundamental knowledge gained about how the brain learns and acquires neuroprosthetic skills into the neurophysiologically-informed design of robust and high-performance closed-loop motor neuroprosthetics that generalize to new tasks.