The brain-machine interface (BMI) is an exciting new class of device in the field of biomedical engineering that shows great promise for the rehabilitation of persons with paralysis by recording neural signals and translating them into movement of objects such as prosthetics and computer cursors. This study aims to present a brief history of the devices including the three main methods of recording neural signals as well as some of the functions possible with BMIs and their basic design. It will also provide insight into some of the technical challenges currently preventing BMIs from widespread use for rehabilitative therapy including, but not limited to, signal degradation and a lack of design consensus. This study will also give examples of exciting new methods that are being considered for integration into the BMI world such as functional electrical stimulation and optogenetics as well as providing some examples of currently available commercial BMIs that are on the market. The study will conclude with a discussion of what needs to be done in order for BMIs to eventually enable paralyzed persons to live independently. A hypothetical scenario is depicted that highlights some of the factors that will need to be considered in order to allow a paralyzed person to fully rely on their BMI. Finally, a discussion of the ethical implications of BMIs are presented including how BMIs should be implemented with children as there is currently no research on that subject. Pediatric adoption of cochlear implants is used as an example of a similar technology that has already been widely accepted for public use despite lingering ethical concerns.

<p>Brain-machine interfaces (BMIs) offer the potential to assist millions of people worldwide suffering from immobility due to loss of limbs, paralysis, and neurodegenerative diseases. BMIs function by decoding neural activity from intact cortical brain regions in order to control external devices in real-time. While there has been exciting progress in the field over the past 15 years, the vast majority of the work has focused on restoring of motor function of a single limb. In the work presented in this thesis, I first investigate the expanded role of primary sensory (S1) and motor (M1) cortex during reaching movements. By varying target size during reaching movements, I discovered the cortical correlates of the speed-accuracy tradeoff known as Fitts' law. Similarly, I analyzed cortical motor processing during tasks where the motor plan is quickly reprogrammed. In each study, I found that parameters relevant to the reach, such as target size or alternative movement plans, could be extracted by neural decoders in addition to simple kinematic parameters such as velocity and position. As such, future BMI functionality could expand to account for relevant sensory information and reliably decode intended reach trajectories, even amidst transiently considered alternatives.</p><p> The second portion of my thesis work was the successful development of the first bimanual brain-machine interface. To reach this goal, I expanded the neural recordings system to enable bilateral, multi-site recordings from approximately 500 neurons simultaneously. In addition, I upgraded the experiment to feature a realistic virtual reality end effector, customized primate chair, and eye tracking system. Thirdly, I modified the tuning function of the unscented Kalman filter (UKF) to conjointly represent both arms in a single 4D model. As a result of widespread cortical plasticity in M1, S1, supplementary motor area (SMA), and posterior parietal cortex (PPC), the bimanual BMI enabled rhesus monkeys to simultaneously control two virtual limbs without any movement of their own body. I demonstrate the efficacy of the bimanual BMI in both a subject with prior task training using joysticks and a subject naïve to the task altogether, which simulates a common clinical scenario. The neural decoding algorithm was selected as a result of a methodical comparison between various neural decoders and decoder settings. I lastly introduce a two-stage switching model with a classify step and predict step which was designed and tested to generalize decoding strategies to include both unimanual and bimanual movements.</p> / Dissertation

No description available.

Balancing subject learning and decoder adaptation is central to increasing brain machine interface (BMI) performance. We addressed these complementary aspects in two studies: (1) a learning study, in which mice modulated “beta” band activity to control a 1D auditory cursor, and (2) an adaptive decoding study, in which a simple recurrent artificial neural network (RNN) decoded intended saccade targets of monkeys. In the learning study, three mice successfully increased beta band power following trial initiations, and specifically increased beta burst durations from 157 ms to 182 ms, likely contributing to performance. Though the task did not explicitly require specific movements, all three mice appeared to modulate beta activity via active motor control and had consistent vibrissal motor cortex multiunit activity and local field potential relationships with contralateral whisker pad electromyograms. The increased burst durations may therefore by a direct result of increased motor activity. These findings suggest that only a subset of beta rhythm phenomenology can be volitionally modulated (e.g. the tonic “hold” beta), therefore limiting the possible set of successful beta neuromodulation strategies. In the adaptive decoding study, RNNs decoded delay period activity in oculomotor and working memory regions while monkeys performed a delayed saccade task. Adaptive decoding sessions began with brain-controlled trials using pre-trained RNN models, in contrast to static decoding sessions in which 300-500 initial eye-controlled training trials were performed. Closed loop RNN decoding performance was lower than predicted by offline simulations. More consistent delay period activity and saccade paths across trials were associated with higher decoding performance. Despite the advantage of consistency, one monkey’s delay period activity patterns changed over the first week of adaptive decoding, and the other monkey’s saccades were more erratic during adaptive decoding than during static decoding sessions. It is possible that the altered session paradigm eliminating eye-controlled training trials led to either frustration or exploratory learning, causing the neural and behavioral changes. Considering neural control and decoder adaptation of BMIs in these studies, future work should improve the “two-learner” subject-decoder system by better modeling the interaction between underlying brain states (and possibly their modulation) and the neural signatures representing desired outcomes.

abstract: Brain-machine interfaces (BMIs) were first imagined as a technology that would allow subjects to have direct communication with prosthetics and external devices (e.g. control over a computer cursor or robotic arm movement). Operation of these devices was not automatic, and subjects needed calibration and training in order to master this control. In short, learning became a key component in controlling these systems. As a result, BMIs have become ideal tools to probe and explore brain activity, since they allow the isolation of neural inputs and systematic altering of the relationships between the neural signals and output. I have used BMIs to explore the process of brain adaptability in a motor-like task. To this end, I trained non-human primates to control a 3D cursor and adapt to two different perturbations: a visuomotor rotation, uniform across the neural ensemble, and a decorrelation task, which non-uniformly altered the relationship between the activity of particular neurons in an ensemble and movement output. I measured individual and population level changes in the neural ensemble as subjects honed their skills over the span of several days. I found some similarities in the adaptation process elicited by these two tasks. On one hand, individual neurons displayed tuning changes across the entire ensemble after task adaptation: most neurons displayed transient changes in their preferred directions, and most neuron pairs showed changes in their cross-correlations during the learning process. On the other hand, I also measured population level adaptation in the neural ensemble: the underlying neural manifolds that control these neural signals also had dynamic changes during adaptation. I have found that the neural circuits seem to apply an exploratory strategy when adapting to new tasks. Our results suggest that information and trajectories in the neural space increase after initially introducing the perturbations, and before the subject settles into workable solutions. These results provide new insights into both the underlying population level processes in motor learning, and the changes in neural coding which are necessary for subjects to learn to control neuroprosthetics. Understanding of these mechanisms can help us create better control algorithms, and design training paradigms that will take advantage of these processes. / Dissertation/Thesis / Doctoral Dissertation Bioengineering 2015

Population-level neurocontrol has been advanced predominately through the miniaturization of hardware, such as MEMS-based electrodes. However, miniaturization alone may not be viable as a method for single-neuron resolution control within large ensembles, as it is typically infeasible to create electrode densities approaching 1:1 ratios with the neurons whose control is desired. That is, even advanced neural interfaces will likely remain underactuated, in that there will be fewer inputs (electrodes) within a given area than there are outputs (neurons). A complementary “software” approach could allow individual electrodes to independently control multiple neurons simultaneously, to improve performance beyond naïve hardware limits. An underactuated control schema, demonstrated in theoretical analysis and simulation (Ching & Ritt, 2013), uses stimulus strength-duration tradeoffs to activate a target neuron while leaving non-targets inactive. Here I experimentally test this schema in vivo, by independently controlling pairs of cortical neurons receiving common optogenetic input, in anesthetized mice. With this approach, neurons could be specifically and independently controlled following a short (~3 min) identification procedure. However, drift in neural responsiveness limited the performance over time. I developed an adaptive control procedure that fits stochastic Integrate and Fire (IAF) models to blocks of neural recordings, based on the deviation of expected from observed spiking, and selects optimal stimulation parameters from the updated models for subsequent blocks. I find the adaptive approach can maintain control over long time periods (>20 minutes) in about 30% of tested candidate neuron pairs. Because stimulation distorts the observation of neural activity, I further analyzed the influence of various forms of spike sorting corruption, and proposed methods to compensate for their effects on neural control systems. Overall, these results demonstrate the feasibility of underactuated neurocontrol for in vivo applications as a method for increasing the controllable population of high density neural interfaces.

<p dir="ltr">Sensory feedback is a critical component for controlling neuroprosthetic devices and brain-machine interfaces (BMIs). A lack of sensory pathways can result in slow, coarse movements when using either of these technologies and, in addition, the user is unable to fully interact with the environment around them. Spinal cord stimulation (SCS) has shown potential for restoring these pathways, but traditional stimulation patterns with constant parameters fail to reproduce the complex neural firing necessary for conveying sensory information. Recent studies have proposed various biomimetic stimulation patterns as a more effective means of evoking naturalistic neural activity and, in turn, communicating meaningful sensory information to the brain. Unlike conventional patterns, biomimetic waveforms vary in frequency, amplitude, or pulse-width over the duration of the stimulation. To better understand the role of these parameters in sensory perception, this thesis worked to investigate the effects of SCS patterns utilizing stochastic frequency modulation, linear frequency modulation, and linear amplitude modulation. By calculating sensory detection thresholds and just-noticeable differences, the null hypothesis for stochastically-varied frequency and linear amplitude modulation techniques was rejected.</p>

Brain Machine Interfaces (BMI) are a rapidly developing technology that raise unique ethical issues that demand review. They have demonstrated impressive restorative potential, particularly for individuals living with epilepsy, and those who are locked in. Although BMIs have the potential to provide significant benefit to millions of users, further advancement of the technology should proceed cautiously, according to the guidelines outlined in this paper. Failure to adhere to ethical guidelines could lead to severe privacy concerns, and would violate moral principles of beneficence, virtue ethics, care ethics, and utilitarianism. Despite the moral risks, BMIs hold promise for reshaping future healthcare delivery.

Brain Machine Interfaces (BMI) provide a communication pathway between the electrical conducting units of the brain (neurons) and external devices. BMI technology may provide improved neurological and physiological functions to patients suffering from disabilities due to damaged nervous systems. Unfortunately, microelectrodes used in Intracortical Neural Interfaces (INI), a subset of the BMI device family, have yet to demonstrate long-term in vivo performance due to material, mechanical and electrical failures. Many state-of-the-art INI devices are constructed using stacks of multiple materials, such as silicon (Si), titanium (Ti), platinum (Pt), parylene C, and polyimide. Not only must each material tolerate the biological environment without exacerbating the inflammatory response, each of the materials used must physically withstand the environment as well as interact well with each other. One approach to address abiotic mechanisms has been optimizing the materials required to fabricate the INI devices. Silicon Carbide (SiC) is a physically robust, hemo and biocompatible, and chemically inert semiconductor. An ‘all-SiC’, or monolithic SiC, device may be the disruptive technology needed in the BMI field to finally achieve long-term and wide-spread use of this technology in humans. The all-SiC device concept is where SiC serves as all device layers: the base (substrate), the conducting traces (electrodes), and the surface conformal insulating layer. The monolithic SiC neural probe is realized by forming high-quality pn junctions of heavily doped SiC on a layer of the opposite polarity. Heavily doped semiconductors display semi-metallic electrical performance, which allow for efficient electrical conduction in the electrode without the deleterious effects of metal ions interacting with the neural environment. The conformal insulator is realized using amorphous-SiC (a-SiC) which can be patterned to open windows to allow electrical conduction to occur between the electrode tips and the brain. Several generations of monolithic SiC devices have been fabricated, tested and are reported in this dissertation. The devices were fabricated utilizing two polytypes of SiC (4H-SiC and 3C-SiC). The monolithic SiC microelectrodes were fabricated utilizing techniques used in the fabrication of Si based microelectrodes. Monolithic SiC devices fabricated include planar single-ended MEAs (with varying lengths and varying active recording area), 60-channel MEAs for in-vitro studies, and 16-electrode Michigan style neural probes for in-vivo studies. Electrical testing of the pn junction demonstrated that the 4H-SiC device can block a forward bias voltage of up to 2.3V and displays reverse bias leakage below 1 nArms well past -20V. Current leakage between adjacent electrodes was ~7.5 nArms over a voltage range of -50V to +50V. Furthermore, electrochemical results show that the 4H-SiC microelectrodes interact with an electrochemical environment primarily through capacitive mechanisms and has an impedance comparable to gold electrodes. Electrode impedance ranged from 675±130 kΩ (GSA = 496 µm2) to 46.5±4.80 kΩ (GSA = 500K µm2). However, the 4H-SiC devices cannot deliver charge as efficiently as other conventionally used microelectrode materials, such as iridium oxide (IrOx), but a larger water window compensates for this since larger stimulation voltages are supported compared to IrOx. All studies and data collected thus far indicate that the monolithic SiC neural device can aid in the advancement of chronic INI use in clinical settings. The all-SiC devices rely on the integration of only robust and highly compatible SiC material, they may offer a promising solution to probe delamination and biological rejection associated with the use of multiple materials used in many current INI devices. Follow-on work is planned to prove this assertion via in vivo studies.