Room: Conference room 6

Abstracts

Brain-computer interface and sensorimotor oscillations: novel perspectives and methods

Abstract

Brain-computer interfaces (BCI) are systems capable of presenting feedback to a user based only on the modulation of their brain signals. I will present results of immediate brain plasticity after one hour of BCI performance from a completely inexperienced user with two different BCI systems, one based on event-related potentials and the other one based on the modulation of sensorimotor rhythms. Our analyses show changes in gray-matter of specific brain regions following a single session of BCI usage. More importantly, both BCI approaches resulted in plastic changes within the respective brain regions that were the most activated according to the BCI task: sensorimotor region after the motor imagery based BCI session and occipital/parietal areas following the event-related potential BCI use. I will also present a new method to maximize the coherence of neural signals from different sources, that can be used to, for example, better estimate corticomuscular coherence and show results comparing this novel method with other mass-bivariate approaches. All this research has been performed collaboratively between the Department of Neurology at MPI for Human Cognitive and Brain Sciences, Leipzig, the Machine Learning Group of the TU-Berlin and the Dept. of Informatics, Statistics and Mathematics from the Public University of Navarre.

A Platform for Large-scale Neuroscientific Studies

Abstract

Neurophysiological research is expensive and complicated. The resulting necessity to conduct studies in a laboratory limits ecological validity, scalability, and generalisability of findings. We introduce MYND: a platform for at-home participation in large-scale neuroscientific studies. Our goal is to establish user-experience design as a paradigm in neuroscientific research to overcome the limits of current studies and to improve ecological validity. MYND provides a smartphone application with a simple user interface that guides subjects through experiment selection, hardware fitting, recording, and upload. This interface enables subjects to self-administer multi-day studies that include both neurophysiological recording sessions and questionnaires. Thirty-two subjects recorded a total of 5846 neurophysiological trials over seven days at home. Results indicate that the application can be used for studies outside of a laboratory, without the need for external guidance. MYND shows that the intersection between neuroscience and human-computer interaction could help to break with the limitations of neuroscientific research by enabling neurophysiological recordings with large populations in a realistic context.

Human-in-the-Loop Machine Learning for closed-loop human augmentation

Abstract

Learning inference and control problems is highly efficient in settings where simulation is available that provide efficient access to the model to be learned. However, in many human-in-the-Loop settings, such as Brain-Computer-Interfaces, the challenge is that the individual human decision maker we want to learn from interaction cannot be substituted by simulation. This is general problem to human-in-the-Loop settings where we want to learn interacting with a specific individual. To address these challenges we develop Closed-loop autoregressive (such as GP autoregressors) and hierarchical learning approaches that we combine with novel neurobehavioral interfaces for human intention decoding.

Machine learning for Multiscale Diagnostic Bioelectronics: From nerve signals to millions of vital signs

Abstract

With the advent of better peripheral neural interfaces, non-invasive physiological sensors and rapid digitization of large amounts of healthcare data, new opportunities arise for creating machine learning algorithms that will generate a more holistic view of patient health. These algorithms will power the next generation bioelectronic medicine devices to enable early diagnosis, disease severity assessment and personalization and adaptability of a therapy. We will present work from the Neural and Data Science lab that combines machine learning with neural and physiological signal processing, using data from multiple sources and scales. Using vagus nerve recordings, we focus on how the peripheral nervous system senses the state and affects the function of the immune and metabolic systems. We want to use this knowledge to develop devices that are able to diagnose and treat various diseases and conditions before symptoms appear. For the same reasons, we are also using multimodal noninvasive physiological signals to infer autonomic nervous system function and imbalance. Finally, we are leveraging the access to large amounts of healthcare data through the Northwell health system, building machine learning algorithms that can predict patient decompensation, using millions of vital signs.

Can we help the brain rewire itself? Hints from the functions and mechanisms acting during sleep

Abstract

Direct or indirect stimulation of brain activity through brain machine interfaces holds the promise of triggering the enduring network modifications necessary to functional recovery. Possibly daunting at first, this aim appears more reachable with regard to the dramatic plastic changes occurring in healthy brains on a daily basis. In light of recent results, I will discuss the putative mechanisms exploited by the mammalian brain to control plasticity during sleep. I will then elaborate on how such mechanisms, originally promoting long-term memory and homeostasis, can be possibly leveraged for recovery.

Connecting brains and machines for neurorehabilitation: from in vivo studies to human applications

Abstract

Neuroprostheses are devices that interface with the nervous system and supplement or substitute functionality in the patient's body. In the collective imagination, neuroprotheses are primarily used to restore sensory (e.g. acoustic or visual neuroprostheses) or motor capabilities (e.g. artificial limbs), but in the recent years, new devices to be applied directly at the brain level are taking place. The idea is to use them to treat the neuronal injury in the brain, where the damage is actually located, and to promote brain plasticity in order to speed up the recovery process. Their implementation requires the knowledge of how the neural dynamics is affected by the lesion and whether and how electrical stimulation can reshape or restore the original behavior, both on a short and a long term perspective. To respond to these and related questions, more than 10 years ago researchers have explored the possibility to create ‘neurohybrid’ systems at the interface between neuroscience and robotics, thus providing an excellent test bed for modulation of neuronal tissue and forming the basis of future bi-directional Brain Machine Interfaces and Prostheses. Within this framework, I will present recent results related to closed-loop paradigms for brain repair in vivo, focusing on how different types of stimulation (open vs closed-loop activity-dependent stimulation) can affect the neural activity. Further, I will describe our current work on human rehabilitation: we make use of neuromodulation of brain cortex in adjunction to the traditional motor or cognitive treatment, and we observe the neural correlates of rehabilitation intervention through high-density electroencephalography (hdEEG). Finally, we will briefly present exoskeletons and prostheses developed in our department for humans, and will address how, in the future, the introduction of neural signals into the control policy of such devices might affect rehabilitation.

Neural Interfaces Clinical applications: focus on stroke rehabilitation

Abstract

Rehabilitation studies leveraging brain-machine interface (BMI) technology have demonstrated clinical improvements with systems linking decoded brain activity with peripheral feedback through the movement of an orthosis. In these systems, the decoded movement intention from brain signals is translated into movement of the peripheral orthosis, giving patients feedback about their brain activity. This allows them first to control their neural activity patterns and adjust them to a given movement and second leverages the endogenous sensorimotor learning system to link these brain signals and the activation of the peripheral nervous system. So far, studies have been limited to linking patterns from non-invasively acquired brain signals to single movements such as only grasping or only reaching, because of the low number of movements that can be reliably decoded on a single-trial basis using non-invasive means. Thus, one way to improve upon current state-of-the-art is to use a neural recording modality that allows for accurate, single-trial decoding of a greater number of movements, and deliver feedback of these decoded movement intention signals through an orthosis with multiple degrees of freedom, allowing functional movements. I will present a novel brain-machine interface rehabilitation system for hemiplegic chronic stroke patients that leverages a temporarily implanted intracortical multi-electrode array, surface EMG and a seven degree-of-freedom arm and hand orthosis. Furthermore, I will present the usability of the system, behavioral and clinical results in a severely paralyzed chronic stroke patient implanted for more than 17 months.

Sensory Neuroprosthetics: Methods to assess functional reorganization and predict the outcome

Abstract

Sensory Substitution Devices (SSD) aim to restore a lost sensory modality using another that remains intact; this is possible due to neuroplasticity, the brain’s ability to adapt to a changing environment. Different SSD have been developed for visual, auditory, vestibular and peripheral sensation restoration but the brain changes behind are still unknown. Neuroimage data analysis offers a great opportunity to better understand these brain changes and be able to develop more personalized rehabilitation strategies. While most of the studies in the literature focus on comparing the brain’s activation pattern during pre- and post-treatment task, functional connectomics are better suited to study how the networks of the brain have been reorganized and to identify the networks that need to be stimulated in order to obtain a better outcome. I will present the results obtained after 6 months of visual rehabilitation of blind children using a SSD that transform visual information to tactile stimuli using a small hand-sized elastomeric device. I will focus on the different methods used to evaluate the brain functional reorganization with neuroimage data to study functional differences in blind people compared to sighted individuals, analyze connectivity changes after training, and perform individualized prediction of rehabilitation outcome. This research has been performed collaboratively between Massachusetts General Hospital - Harvard Medical School, Complutense University of Madrid and Tecnalia.

Real-time, bidirectional interfacing with the nervous system: lessons from biology and engineering

Abstract

Bidirectional neural-computer interfaces record and decode neural activity, and modulate it in real-time by delivering neurostimulation back into the central or the peripheral nervous system, in a responsive and adaptive manner. Applications of such interfaces range from neuroplasticity, neurorehabilitation and delivery of sensory feedback to the cortex, to modulation of circuits involved in abnormal motor movement, chronic pain, etc. Such interfaces rely on semi-real-time operation of a closed-loop system and on reliable chronic interfacing with neural tissue, two principles that pose special challenges on both biological and engineering fronts. This presentation will discuss some of these challenges and ways to address them, in the context of systems that interface with the brain and with the peripheral nervous system.