Most brain-computer interface (BCI) studies that use noninvasive EEG are based on detecting patterns in EEG that are known to correspond to various mental imagery, such as imagined hand and foot movements. To detect a wider variety of mental tasks, methods must be employed that find more complex patterns across multiple electrodes and over time. This talk will summarize the use of Short-time Principal Components Analysis to represent such patterns. Principal Components Analysis (PCA) is often used to project high-dimensional signals to lower dimensional subspaces defined by basis vectors that maximize the variance of the projected signals. Data containing variations of relatively short duration and small magnitude, such as those seen in EEG signals, may not be captured by PCA when applied to time series of long duration. Instead, PCA is applied independently to short segments of data and the basis vectors themselves are used as features for classification. In addition, time embedding the EEG by augmenting each sample with previous samples prior to PCA results in a representation that captures EEG variations in space and time. The resulting features are classified into categories corresponding to which mental task a subject is performing in a brain-computer interface (BCI) paradigm. Approximately 80% of test samples are correctly classified as one of five mental tasks. In addition, an on-line artifact removal method is demonstrated and an inexpensive hardware and software system for BCI research is described.
Responsive (closed-loop) electrical stimulation of the brain to treat medically intractable epilepsy is currently being tested in clinical trials. Responsive stimulation offers the potential advantages of reducing exposure to the acute and chronic toxicity of anticonvulsant medications and providing a reversible alternative to resective surgery through the parsimonious stimulation of the epileptic focus as the seizure begins. Preliminary trials demonstrate the feasibility and safety of the current implantable device. Simple algorithms for seizure detection can be effectively implemented. Chronic responsive stimulation of neocortical and allocortical structures is well tolerated, but current experience indicates the need to develop and refine stimulation protocols. In addition, implantable devices with recording and reporting capability give us the ability to monitor and characterize the electrical activity of the brain in people participating in activities of daily life over months and years. This novel temporal window offers us a new perspective on epileptic brain activity.
Neural interfaces are poised to revolutionize our ability to restore lost function to people with neurologic disease or injury. Over the past decade, technologies to record the individual and simultaneous activities of dozens to hundreds of cortical neurons have yielded new understandings of cortical function in movement, vision, cognition, and memory. This preclinical research, generally performed with healthy, neurologically intact non-human primate subjects, has demonstrated that direct, closed-loop, neural control of virtual and physical devices can be achieved. Recently, this exciting research has been translated into initial pilot clinical trials examining the feasibility of persons with tetraplegia controlling a computer cursor simply by imagining movement of their own hand. With encouraging initial results, our research group is examining the use of intracortically-based neuronal ensemble activity toward restoring the communication, mobility, and independence of people with paralysis.
The sudden and apparently unpredictable nature of seizures is one of the most disabling aspects of epilepsy. Although there are numerous studies exploring basic neuronal mechanisms that are likely to be associated with seizures, to date no definite information is available as to how or when or why a seizure occurs in humans. A reliable identification of seizure precursors from the EEG of epilepsy patients would allow the development of new therapeutic possibilities, which could improve the quality of life of epilepsy dramatically. The last three decades have witnessed a rapid increase in the development of new EEG analysis techniques that appear to be capable of defining seizure precursors. Since the 1970s studies on seizure prediction have advanced from preliminary descriptions of pre-seizure phenomena and proof of principle studies via controlled studies to studies on continuous multi-day recordings. I will present an overview of the field of seizure prediction, its history, accomplishments, recent controversies, and potential for future development, also with regard to real-time processing of multichannel EEG recordings.
One of the greatest challenges for developing antiepileptic devices is understanding epileptic networks, seizure generation, and their spatial and temporal characteristics. The scale of these processes, which is at present not well defined, has important implications for sensors, effectors to deliver therapy, and developing more definitive treatments for epilepsy based upon mechanistic hypotheses. In this talk I will touch on major challenges related to mapping functional and dysfunctional networks in human brain, with an emphasis on epilepsy and seizure generation. I will discuss the current state of the art in clinical practice, and potential directions for future human devices.
11,000 individuals suffer spinal cord injuries each year. Roughly half suffer at least partial paralysis of all four limbs. Unlike stroke, spinal cord injury is most prevalent among young people, with decades of remaining life. Progress has been made using functional electrical stimulation (FES) to activate forearm muscles to restore grasp. However, providing the means to control the many degrees of freedom needed for dexterous manipulation remains an important limitation. Existing systems are restricted to several pre-programmed grasp patterns controlled using residual voluntary movement of the proximal limb. We have previously shown that linear combinations of neural firing rate signals can be used to predict muscle EMG signals during reaching movements. Weimer cascade, non-linear systems offered rather moderate additional improvement. I will describe experiments to develop a brain controlled FES system capable of restoring voluntary hand movements in paralyzed subjects. We induced temporary paralysis in monkeys by blocking the median and ulnar nerves at the elbow. Following the paralysis, these monkeys were able roughly to double their maximum voluntary wrist flexion force with the brain-controlled FES system. Furthermore, they were able to grade the force accurately, in order to match a cursor to targets at different force levels, at speeds of one-half to two-thirds normal. These results provide an important proof of concept, demonstrating the feasibility of cortically controlled FES prostheses. Such systems could offer significant advantages to paralyzed patients with injuries in the mid-cervical spinal cord, and even greater benefits to high-cervical spinal cord injured patients with paralysis of the entire upper limb.
Work done in collaboration with Eric A. Pohlmeyer, Emily R. Oby, Eric J. Perreault, Sara A. Solla, Kevin L. Kilgore, and Robert F. Kirsch.
Brain computer interface (BCI) technology has classically focused on two signal acquisition modalities for control: multi, single-unit activity (MSU) and electroencephalography (EEG). While MSU activity provides arguably the best multi-dimensional signal for BCI control, obtaining long-term stability of single unit recordings has proven difficult at best due to glial encapsulation issues. EEG, on the other hand, is a non-invasive technique where relatively large electrodes are placed on the surface of the scalp to record ensemble activity emanating from the underlying cortex. Given the large separation between the cortical surface and the recording electrodes as well as the inhomogeneous conductivity of the dura, skull, and skin; a rather large area of cortex needs to be synchronously active to be "electrically visible" on the scalp (~6 cm^2). Training such large cortical networks for BCI control of a few degrees-of-freedom can take month(s) to learn. On the other hand, Local Field potentials taken from either within brain parenchyma or on the surface of the brain (ECoG) are typically generated from much smaller neural ensembles (1-2 mm^2). Our recent results in non-human primates suggest that LFP spectral power in the 100-200 Hz range is well correlated with single unit activity while the lower frequency rhythms correlate with EEG. This suggests that LFPs can be used to obtain many of the same control features of both MSU and EEG recordings simultaneously and may be a more pragmatic solution to clinical BCI technology.
Patients with amyotrophic lateral sclerosis form a target group for BCI applications. In this talk studies will be reviewed that tested the feasibility of BCIs based on sensorimotor rhythms (SMR)(N=4 patients) and the P300 evoked potential (N=6 patients). Results indicate that the both BCIs can be used by patients with amyotrophic lateral sclerosis, but that the P300-based BCI allows the patients to communicate from the first training, whereas patients have to go through time-consuming training sessions to use a SMR-based BCI. Furthermore, the mood of the patient and the motivation he has to work with the BCI (assessed before BCI training sessions) are related fluctuations in performance. These psychological parameters predict performance, which is very useful for the impementation of BCIs at the homes of people.
A last topic of this talk will be the development of non-visual BCIs. Many ALS patient develop vision impairments when they reach the final stage of the disease (complete locked-in state; CLIS), but most BCI research focuses on visual BCI applications. This might explain why no CLIS patient has been able to use a BCI for communication. We will discuss results from a study with an auditory P300-based BCI and an auditory SMR-based BCI and show that in this case, the auditory SMR-based BCI might be the better approach. Furthermore, we discuss why it may be so difficult for CLIS patients to learn how to use a BCI.
Fundamental to understanding how our world is represented in our brain is the ability to observe the collective activity of ensembles of neurons acting in concert, and to correlate this activity with an observed behavior. Recent advances in the fabrication of implantable high-density microelectrode arrays have triggered a surge in research aimed at interfacing the brain at the microscale. Despite the large volumes of physiological and behavioral data obtained, our understanding of the central nervous system remains elusive. There are significant biological and engineering challenges to the ability to fully observe, analyze and decode large scale neural data with an implanted device.
In this talk, I will demonstrate how some of these challenges can be overcome using a blend of statistical signal processing, machine learning and computational neuroscience methods. In particular, I will show how we can conceptually design a Multiscale intra-Cortical Neural Interface System (MiCNIS) that extracts all the hypothesized constituents of the neural code in real time within the resource-constrained environment of an implanted system. I will also demonstrate how we can utilize these methods to characterize and quantify cortical plasticity that underlies our ability to learn and memorize. Specifically, I will show how we can infer the topology of cortical circuits involved in encoding task-dependent behavioral parameters. I will conclude with a brief discussion on the potential of these methods to enhance our understanding of the brain encoding mechanism underlying single cell and population activity and how they can be useful to improve neural decoding in future generations of Brain Machine Interface systems.
Closed-loop neural interfaces are one of the most exciting emerging technologies to impact biomedical research, human health, and rehabilitation. By combining engineering and neurophysiologic knowledge into bio-interactive brain-machine interfaces (BMI), a new generation of medical devices is being developed to functionally link large ensembles of neurons in the central nervous system (CNS) directly with man-made systems. This talk will present overview of the design and analysis of decoding methodologies for closed-loop BMI systems. Included are techniques for neurophysiologic feature detection and methods for determining representation in multiscale signals. Within the context of these traditional decoding approaches, a new method of continuous decoding based on reinforcement learning (RL) will be presented. New principles of neural control will be developed as they are learned through experience and interaction with the environment. It will be shown how a theory and method of co-adaptive shaping using RL to achieve brain control of a prosthetic enables the development of complex tasks while reducing the "learning curve" for patients using a BMI.
The state-of-the-art in clinical epilepsy stimulation is faced with the challenge of specifying why some epileptic individuals respond to stimulation treatment and why others don't. To address this issue, we are investigating a chronic animal model of temporal lobe epilepsy to develop new treatments for the human condition. Here, we seek to advance the knowledge and technologies needed to produce more effective therapeutic stimulation paradigms by studying the onset of seizure from multiple neurophysiologic scales that are simultaneously recorded: single neuron, multiunit activity, and local fields. Our recent studies have shown that abnormal neuromodulation in pyramidal cells, interneurons, and population spikes occur in advance of impending seizures in the animal model.
Brain-computer interfaces (BCIs) convert brain signals into outputs that communicate a user's intent. BCIs can be used by people who are severely paralyzed to communicate and interact with their environment. However, practical applications of BCI technology are currently impeded by the limitations and requirements of the prevailing non-invasive and invasive methods.
Non-invasive BCIs use electroencephalographic activity (EEG) recorded from the scalp. While EEG-based BCIs support higher performance than often assumed, the acquisition of high levels of control typically requires extensive user training. Invasive BCIs use local activity from multiple neurons recorded within the brain. Signals recorded within cortex have higher fidelity and might support BCI systems that require less training than EEG-based systems. However, clinical implementations of intracortical BCIs are currently impeded mainly by the difficulties in maintaining stable long-term recordings.
Electrocorticographic (ECoG) recordings from the cortical surface could be a powerful and practical alternative to current non-invasive and invasive BCI recording methods. ECoG has higher spatial resolution than EEG, broader bandwidth, higher characteristic amplitude, and far less vulnerability to artifacts such as EMG or ambient noise. At the same time, because ECoG does not require penetration of the cortex, it is likely to have greater long-term stability and to produce less tissue damage and reaction than intracortical recordings.
This talk will review the current state of ECoG-based BCI systems. It will first describe the ECoG responses that correspond to actual or imagined limb movements. It will then describe the one- and two-dimensional BCI experiments supported by these tasks. Finally, it will show that it is possible to use ECoG to accurately decode specific parameters of hand movements, finger movements, and speech. In summary, these recent findings are strong evidence that ECoG could be a robust and practical alternative for clinical application of BCI technology.
Recent technological developments have made it possible to simultaneously record the spiking activity of increasingly large populations of cortical neurons using extracellular microelectrodes. In multiple cell recordings identifying the number of neurons and assigning each action potential to a particular source, commonly referred to as 'spike sorting', is a highly non-trivial problem. Density grid contour clustering provides a computationally efficient way of locating high-density regions of arbitrary shape in low-dimensional space. When applied to waveforms projected onto their first two principal components, the algorithm allows the extraction of templates that provide high-dimensional reference points that can be used to perform accurate spike sorting. A strategy based on template matching can locate these templates in waveform samples despite the influence of noise, spurious threshold crossing and waveform overlap. Tests with a large synthetic dataset incorporating realistic challenges faced during spike sorting (including overlapping and phase-shifted spikes) reveal that this strategy can consistently yield results with less than 6% false positives and false negatives (and less than 2% for high signal-to-noise ratios) at processing speeds exceeding those previously reported for similar algorithms by more than an order of magnitude.
A brain-computer interface, or BCI, creates a new non-muscular output channel for the brain. Instead of being executed by peripheral nerves and muscles, the use's wishes are conveyed by brain signals, and these brain signals do not depend for their generation on neuromuscular activity. BCIs can allow people who are severely paralyzed, or even "locked in," to use brain signals to write, communicate with others, control their environments, access the Internet, or operate neuroprostheses. Current BCIs determine the intent of the user from electrophysiological signals recorded non-invasively from the scalp (EEG) or invasively from the cortical surface (ECoG) or from within the brain (neuronal action potentials or local field potentials). These signals are translated in real-time into commands that operate a computer display or other device. Successful operation requires that the user encode commands in these signals and that the BCI derive the commands from the signals. Thus, the user and the BCI system need to adapt to each other both initially and continually so as to ensure stable performance. This dependence on the mutual adaptation of user to system and system to user is a fundamental principle of BCI operation. EEG-based systems use a variety of features. Human BCI experience to date consists largely of non-invasive EEG-based research. Several different kinds of EEG-based BCIs are now being explored in people. They are distinguished by the particular EEG features they use to derive the user's intent. Sensorimotor rhythm-based BCIs use 8-12 Hz (mu) and 18-26 Hz (beta) oscillations in the EEG recorded over sensorimotor cortices. P300-based BCIs use the P300 evoked potential, which appears in the EEG over central areas about 300 ms after a salient or attended stimulus. Present-day EEG-based BCI methods can support word-processing or other simple functions. They also appear able to control the multidimensional movements of a neuroprosthesis or a device such as a robotic arm. Currently, EEG-based BCIs are being evaluated for independent use for communication, word processing, and environmental control by people with severe disabilities in their own homes.