Brain-Computer Interface (BCI) and Internet of Things (IoT) systems have recently been amalgamated to create BCIoT. Most of the early applications have focused on the healthcare sector, and more recently, in education, virtual reality, smart homes, and smart vehicles, amongst others. While there are many transversal developing stages that can be satisfied by a single system, no common enabling technology or standards exist. These challenges are address in the proposed platform, Brain-eNet. This technology was developed considering the constraints-space defined by BCIoT real-time mobile applications. This is expected to enable the development of BCIoT systems by providing modular hardware and software resources. Two instances of this platform implementation are provided, a motor intent detection for rehabilitation and an emotion recognition system.

Scalp electroencephalography (EEG) is a neural source signal that is extensively used in neuroengineering due to its non-invasive nature and ease of collection. However, a drawback to the use of EEG is the prevalence of physiological artifacts generated by eye movements and eye blinks that contaminate the brain signals. Previously, we have proposed and validated an H∞-based Adaptive Noise Cancellation (ANC) technique for the real-time identification, learning and removal of eye blinks, eye motions, amplitude drifts and recording biases from EEG simultaneously. However, the standard electroocu-lography (EOG) electrode configuration requires four electrodes for EOG measurement, which limits its applicability for reduced-channel mobile applications, such as brain-computer interfaces (BCI). Here, we assess multiple configurations with varying number of EOG electrodes and compare the ANC effectiveness of these configurations to the ideal four-electrode configuration. From an analysis of the root mean squared error (RMSE) and differences in signal to noise ratios (SNR) between the ideal four-electrode case and the alternative configurations, it is reported that several three-electrode alternative configurations were effective in essentially replicating the ability to remove EOG artifacts in an experimental cohort of ten healthy subjects. For nine subjects, it was shown that only two to three EOG electrodes were needed to achieve similar performance as compared to the four-electrode case. This study demonstrates that the typical four-electrode configuration for EOG recordings for adaptive noise cancellation of ocular artifacts may not be necessary; by using the proposed new EOG configurations it is possible to improve electrode allocation efficiency for EOG measurements in mobile EEG applications.

Decoding taste from brain activity is of paramount importance for understanding brain responses to gustatory stimuli with applications in the food industry, neuro-marketing, objective evaluation of consumer's preferences, and addiction prevention/treatment. However, progress in this field has been thwarted by challenging data acquisition and the multimodal aspects of gustatory events, which make traditional decoding based on event-related potentials (ERP) very difficult. Additionally, emerging approaches such as Deep Learning come short to give insights into decoding taste because their need of big amounts of data. In this paper, we propose a processing and analytical pipeline based on Multivariate Pattern Analysis (MPA) and Global Field Power (GFP) to analyze gustatory evoked potentials recorded with high-density scalp electroen-cephalography (EEG). Based on this analysis we show the temporal evolution of neural codes for identification of water-vs-wine and wine-vs-wine dyads.

Sixty percent of elderly hand movements involve grasping, which is unarguably why grasp restoration is a major component of upper-limb rehabilitation therapy. Neuromuscular electrical stimulation is effective in assisting grasping, but challenges around patient engagement and control, as well as poor movement regulation due to fatigue and muscle nonlinearity continue to hinder its adoption for clinical applications. In this study, we integrate an electroencephalography-based brain–machine interface (BMI) with closed-loop neuromuscular stimulation to restore grasping and evaluate its performance using an isometric force tracking task. After three sessions, it was concluded that the normalized tracking error during closed-loop stimulation using a state-space feedback controller (25 ± 15%), was significantly smaller than conventional open-loop stimulation (31 ± 24%), (F(748.03, 1) = 23.22,p< 0.001). Also, the impaired study participants were able to achieve a BMI classification accuracy of 65 ± 10% while able-bodied participants achieved 57 ± 18% accuracy, which suggests the proposed closed-loop system is more capable of engaging patients for rehabilitation. These findings demonstrate the multisession performance of model-based feedback-controlled stimulation, without requiring frequent reconfiguration.

Wearable robotic devices are being designed to assist the elderly population and other patients with locomotion disabilities. However, wearable robotics increases the risk from falling. Neuroimaging studies have provided evidence for the involvement of frontocentral and parietal cortices in postural control and this opens up the possibility of using decoders for early detection of balance loss by using electroencephalography (EEG). This study investigates the presence of commonly identified components of the perturbation evoked responses (PEP) when a person is in an exoskeleton. We also evaluated the feasibility of using single-trial EEG to predict the loss of balance using a convolution neural network. Overall, the model achieved a mean 5-fold cross-validation test accuracy of 75.2 % across six subjects with 50% as the chance level. We employed a gradient class activation map-based visualization technique for interpreting the decisions of the CNN and demonstrated that the network learns from PEP components present in these single trials. The high localization ability of Grad-CAM demonstrated here, opens up the possibilities for deploying CNN for ERP/PEP analysis while emphasizing on model interpretability.

The use of a brain-machine interface (BMI) in combination with powered exoskeletons can assist patients with lower limb disabilities to walk again. These neurorobotic systems are commonly based on motor imagery, but their performance may suffer from lack of user engagement in the task or from cognitive load due to multi-tasking. The present paper shows a novel algorithm based on the gamma spectral band, using the Stockwell transform and a set of smoothing filters, to assess the quality of and improve the decoding of motor imagery during the use of a BMI-Rex exoskeleton system. The results computed in a pseudo-online scenario reveal a high accuracy with a very low false positive ratio.

In this communication, a translational roadmap for a noninvasive Brain Machine Interface (BMI) system for rehabilitation is presented. This multi-faceted project addresses important engineering, clinical, end user and regulatory challenges. The goal is to improve the feasibility of at-home neurorehabilitation for patients with chronic stroke by providing a low-cost, portable, form fitting, reliable, and easy-to-use system. The proposed BMI system is also designed to enable direct communication between the end-user and clinician, allowing for continuous patient specific rehabilitation optimization.

Powered exoskeletons for gait rehabilitation and mobility assistance are currently available for the adult population and hold great promise for children with mobility limiting conditions. Described here is the development and key features of a modular, lightweight and customizable powered exoskeleton for assist-as-needed overground walking and gait rehabilitation. The pediatric lower-extremity gait system (PLEGS) exoskeleton contains bilaterally active hip, knee and ankle joints and assist-as-needed shared control for young children with lower-limb disabilities such as those present in the Cerebral Palsy, Spina Bifida and Spinal Cord Injured populations. The system is comprised of six joint control modules, one at each hip, knee and ankle joint. The joint control module, features an actuator and motor driver, microcontroller, torque sensor to enable assist-as-needed control, inertial measurement unit and system monitoring sensors. Bench-testing results for the proposed joint control module are also presented and discussed.

Neural decoding of human locomotion, including automated gait intention detection and continuous decoding of lower limb joint angles, has been of great interest in the field of Brain Machine Interface (BMI). However, neural decoding of gait in developing children has yet to be demonstrated. In this study, we collected physiological data (electroencephalography (EEG), electromyography (EMG)), and kinematic data from children performing different locomotion tasks. We also developed a state space estimation model to decode lower limb joint angles from scalp EEG. Fluctuations in the amplitude of slow cortical potentials of EEG in the delta band (0.1 - 3 Hz) were used for prediction. The decoding accuracies (Pearson's r values) were promising (Hip: 0.71; Knee: 0.59; Ankle: 0.51). Our results demonstrate the feasibility of neural decoding of children walking and have implications for the development of a real-time closed-loop BMI system for the control of a pediatric exoskeleton.

Recent advancements in powered lower limb prostheses have led to the development of neural-machine interfaces for natural control during bipedal locomotion. In particular, electromyography (EMG) patterns recorded from the amputated limb can be leveraged to infer the intended gait pattern of the user. However, the optimal control strategy for translating the EMG patterns to kinematic space remains a challenge. In this study, six able bodied subjects were instrumented for mobile brain-body imaging and asked to walk on a multi-terrain gait course. A non-linear extension of the Kalman filter was used to predict knee and ankle joint kinematics from lower limb muscle activation patterns during overground locomotion. Specifically, muscles of the anterior and posterior thigh were used to predict both the knee and ankle joint position. The results revealed that muscles in the thigh can be used to predict the position of the knee and ankle with high accuracy. The highest mean r-value obtained for each of the six subjects was 0.92, 0.77, 0.38, 0.39, 0.63, and 0.77, with corresponding SNR values of 10.8 dB, 6.7 dB, 5.9 dB, 2.8 dB, 9.1 dB, and 9.2 dB, for each subject, respectively. This study is the first to demonstrate that continuous EMG can be used to predict the joint kinematics of the knee and ankle during overground locomotion. This approach may provide improvements during closed-loop control of a powered lower limb prosthesis when compared to other pattern-recognition based methods.

This study investigates the neural features of locomotion mode transitions (i.e., level-ground walking to stair ascent) from non-invasive electroencephalography (EEG) signals. A systematic EEG processing method was implemented to reduce artifacts. Source localization using independent component analysis and k-mean clustering algorithm revealed the involvement of four clusters in the brain (Left and Right Occipital Lobe, Posterior Parietal Cortex, and Motor Cortex) during the walking tasks. Our results show significant differences in spectral power in the Occipital cluster between level-ground (LW) and stair (SA) walking. Additionally, significant increases in spectral power were detected up to 1.4 second before the critical transition time (LW to SA). The findings have implications for developing noninvasive lower-limb neuroprostheses that predict, rather than respond to, the user gait intentions. This work is a further step toward the development of a multimodal Neural-machine Interface (NMI) that fuses EEG and electromyography (EMG) signals for intuitive and flexible control of power prosthetic legs.

Recent advances in brain-computer interface (BCI) technologies have shown the feasibility of neural decoding for both users' gait intent and continuous kinematics. However, the cortical adaptation and the dynamics of cortical involvement in human upright walking with a closed-loop BCI in virtual environment (VE) have yet to be demonstrated. To address explore this possibility, we designed a closed-loop BCI to allow users to control a virtual avatar to walk using their encephalography (EEG). Delta band EEG (0.1-3 Hz) was used as the main feature in prediction. Our results demonstrate the feasibility of using a closed-loop BCI to learn to control a walking avatar. The average decoding accuracies (Pearson's r values) across all subjects increased from (Hip: 0.18 ± 0.31; Knee: 0.23 ± 0.33; Ankle: 0.14 ± 0.22) on Day 1 to (Hip: 0.40 ± 0.24; Knee: 0.55 ± 0.20; Ankle: 0.29 ± 0.22) on Day 8. Source localization revealed significant differences in cortical network activity between walking with and without closed-loop BCI control of the virtual avatar. This current study demonstrates the feasibility of using a closed-loop EEG-based BCI-VR to trigger cortical adaptation, promoting cortical involvement, and monitoring cortical activity from non-invasive EEG. Our system may be relevant for neurological gait rehabilitation as a clinical tool for post-stroke physical training and clinical assessment.

Advanced powered lower-limb prosthetic devices require an intuitive and flexible user control interface to work in a dynamic environment. This study investigated the feasibility of inferring muscle activation patterns (electromyography, EMG, envelope) from non-invasive electroencephalography (EEG) signals. Six healthy individuals participated in this study; the subjects were instructed to walk at a comfortable speed across various terrains (e.g. level-ground, up/down slope, up/down stair walking). An unscented kalman filter (UKF) was used to predict the EMG envelope from fluctuations in the amplitude of slow cortical potentials of EEG in the delta band (0.1-3 Hz). The highest decoding accuracy obtained was an r-value (Pearson's correlation r-value) of 0.57 in the medial gastrocnemius of a single subject. In the same subject, the mean r-value across all the muscle groups exceeded 0.4. The mean accuracy across all subjects and muscle group corresponded to an r-value of 0.236. As for the Signal to Noise Ratio (SNR), 79.3% of the obtained results were more than 0 dB with mean performance SNR of 0.8 (max: 2.8 to min: -1.7). The highest accuracy was obtained using a lag of 50ms with a window length (tap) of 500ms. In conclusion, this is the first study to show offline continuous decoding of the EMG envelope during over-ground walking on multiple terrains. The results show the feasibility of such neural decoding. This method could be coupled with EMG-based terrain prediction techniques to further improve the neural control interface with powered lower-limb prostheses.

Brain-machine interface (BMI) devices have unparalleled potential to restore functional movement capabilities to stroke, paralyzed and amputee patients. Although BMI systems have achieved success in a handful of investigative studies, translation of closed-loop neuroprosthetic devices from the laboratory to the market is challenged by gaps in the scientific data regarding long-term device reliability and safety, uncertainty in the regulatory, market and reimbursement pathways, lack of metrics for evaluating and quantifying performance in BMI systems, as well as patient-acceptance challenges that impede their fast and effective translation to the end user. This review focuses on the identification of engineering, clinical and user's BMI metrics for new and existing BMI applications.

Achieving dexterous volitional control of an upper-limb prosthetic device will require multimodal sensory feedback that goes beyond vision. Haptic display is well-positioned to provide this additional sensory information. Haptic display, however, includes a diverse set of modalities that encode information differently. We have begun to make a comparison between two of these modalities, force feedback spanning the elbow, and amplitude-modulated vibrotactile feedback, based on performance in a functional grasp and lift task. In randomly ordered trials, we assessed the performance of N=11 participants (8 able-bodied, 3 amputee) attempting to grasp and lift an object using an EMG controlled gripper under three feedback conditions (no feedback, vibrotactile feedback, and force feed-back), and two object weights that were undetectable by vision. Preliminary results indicate differences between able-bodied and amputee participants in coordination of grasp and lift forces. In addition, both force feedback and vibrotactile feedback contribute to significantly better task performance (fewer slips) and better adaptation following an unpredicted weight change. This suggests that the development and utilization of internal models for predictive control is more intuitive in the presence of haptic feedback.

Advanced prosthetic hands offer the promise of great dexterity; however, myoelectric control techniques, successful with low degree-of-freedom prosthetics, are often set aside by amputees due to the lack of important sensations of touch and effort experienced in the interaction between prosthetic hand and task. In this paper, we explore the efficacy of various modalities of feedback (visual, tactile, visual and tactile, and none) conveying proprioceptive information, specifically the error in joint angles between a desired and actual pose of a virtual prosthetic hand. Our analysis of performance in achieving and maintaining a desired prosthetic hand pose indicates a significant effect of feedback condition, with visual and visual+tactile outperforming tactile alone and a no-feedback condition. Further, the combination of tactile and visual feedback does not seem to have significant drawbacks over visual feedback alone. For tasks that rely on proprioception in the absence of visual feedback, or when attention must be focused elsewhere, we see a performance benefit to the inclusion of tactile cueing, with no lags in reaction times or requirements for increased effort measured by muscle activation.

The reach-to-grasp movement exhibits both parallel and serial combinations of tightly scheduled movement components. Among the movement components are a free movement phase involving transport and pre-shaping of the hand, and in-contact phases, involving grasping, lifting, and manipulation of objects with appropriate forces. Several factors are known to influence the shape of the hand (hand preshaping) during a reach-to-grasp movement, including the initial aperture of the fingers, object size, or movement speed. Moreover, the spatiotemporal specification of the parameters for hand pre-shaping may evolve gradually during the reach-to-gasp movement. In this paper we simulate three computational models of coordination between hand transport and hand preshaping under different initial finger aperture conditions to investigate the spatiotemporal dynamics of hand preshaping.

A multilayer perceptron trained with the backpropagation procedure is used to detect the fundamental frequency (F/sub 0/), or pitch, of a speech signal. The model does not require preprocessing of the signal and has good discriminatory capabilities. Preliminary results suggest that a multilayer perceptron can be trained to extract F/sub 0/ as well as the formants. In the preliminary experiments, the detection rate of F/sub 0/ was 100% for different numbers of hidden units. As the number of hidden units was increased, the generalization capabilities of the neural net decreased.>