Electroencephalography (EEG) analysis has been an important tool in neuroscience with applications in neuroscience, neural engineering (e.g. Brain-computer interfaces, BCI's), and even commercial applications. Many of the analytical tools used in EEG studies have used machine learning to uncover relevant information for neural classification and neuroimaging. Recently, the availability of large EEG data sets and advances in machine learning have both led to the deployment of deep learning architectures, especially in the analysis of EEG signals and in understanding the information it may contain for brain functionality. The robust automatic classification of these signals is an important step towards making the use of EEG more practical in many applications and less reliant on trained professionals. Towards this goal, a systematic review of the literature on deep learning applications to EEG classification was performed to address the following critical questions: (1) Which EEG classification tasks have been explored with deep learning? (2) What input formulations have been used for training the deep networks? (3) Are there specific deep learning network structures suitable for specific types of tasks? A systematic literature review of EEG classification using deep learning was performed on Web of Science and PubMed databases, resulting in 90 identified studies. Those studies were analyzed based on type of task, EEG preprocessing methods, input type, and deep learning architecture. For EEG classification tasks, convolutional neural networks, recurrent neural networks, deep belief networks outperform stacked auto-encoders and multi-layer perceptron neural networks in classification accuracy. The tasks that used deep learning fell into five general groups: emotion recognition, motor imagery, mental workload, seizure detection, event related potential detection, and sleep scoring. For each type of task, we describe the specific input formulation, major characteristics, and end classifier recommendations found through this review. This review summarizes the current practices and performance outcomes in the use of deep learning for EEG classification. Practical suggestions on the selection of many hyperparameters are provided in the hope that they will promote or guide the deployment of deep learning to EEG datasets in future research.

It is generally thought that the signal-to-noise ratio, the bandwidth, and the information content of neural data acquired via noninvasive scalp electroencephalography (EEG) are insufficient to extract detailed information about natural, multijoint movements of the upper limb. Here, we challenge this assumption by continuously decoding three-dimensional (3D) hand velocity from neural data acquired from the scalp with 55-channel EEG during a 3D center-out reaching task. To preserve ecological validity, five subjects self-initiated reaches and self-selected targets. Eye movements were controlled so they would not confound the interpretation of the results. With only 34 sensors, the correlation between measured and reconstructed velocity profiles compared reasonably well to that reported by studies that decoded hand kinematics from neural activity acquired intracranially. We subsequently examined the individual contributions of EEG sensors to decoding to find substantial involvement of scalp areas over the sensorimotor cortex contralateral to the reaching hand. Using standardized low-resolution brain electromagnetic tomography (sLORETA), we identified distributed current density sources related to hand velocity in the contralateral precentral gyrus, postcentral gyrus, and inferior parietal lobule. Furthermore, we discovered that movement variability negatively correlated with decoding accuracy, a finding to consider during the development of brain–computer interface systems. Overall, the ability to continuously decode 3D hand velocity from EEG during natural, center-out reaching holds promise for the furtherance of noninvasive neuromotor prostheses for movement-impaired individuals.

Stroke significantly affects thousands of individuals annually, leading to considerable physical impairment and functional disability. Gait is one of the most important activities of daily living affected in stroke survivors. Recent technological developments in powered robotics exoskeletons can create powerful adjunctive tools for rehabilitation and potentially accelerate functional recovery. Here, we present the development and evaluation of a novel lower limb robotic exoskeleton, namely H2 (Technaid S.L., Spain), for gait rehabilitation in stroke survivors. H2 has six actuated joints and is designed to allow intensive overground gait training. An assistive gait control algorithm was developed to create a force field along a desired trajectory, only applying torque when patients deviate from the prescribed movement pattern. The device was evaluated in 3 hemiparetic stroke patients across 4 weeks of training per individual (approximately 12 sessions). The study was approved by the Institutional Review Board at the University of Houston. The main objective of this initial pre-clinical study was to evaluate the safety and usability of the exoskeleton. A Likert scale was used to measure patient’s perception about the easy of use of the device. Three stroke patients completed the study. The training was well tolerated and no adverse events occurred. Early findings demonstrate that H2 appears to be safe and easy to use in the participants of this study. The overground training environment employed as a means to enhance active patient engagement proved to be challenging and exciting for patients. These results are promising and encourage future rehabilitation training with a larger cohort of patients. The developed exoskeleton enables longitudinal overground training of walking in hemiparetic patients after stroke. The system is robust and safe when applied to assist a stroke patient performing an overground walking task. Such device opens the opportunity to study means to optimize a rehabilitation treatment that can be customized for individuals. Trial registration: This study was registered at ClinicalTrials.gov ( https://clinicaltrials.gov/show/NCT02114450 ).

Visuomotor adaptation to a gradual or sudden screen cursor rotation was investigated in healthy young and elderly subjects. Both age groups were equally divided into two subgroups; one subgroup was exposed to 11.25° step increments of visual feedback rotation, every 45 trials (up to a total of 90°), whereas a second subgroup was subjected to 90° rotation from the onset of exposure. Participants performed discrete, horizontal hand movements to virtual targets in four randomized directions. Targets appeared on a computer screen in front of them, and a board prevented vision of the hand at all times. Differential effects of aging on visuomotor adaptation were found, depending on the time course of the visual distortion. In both age groups, early exposure to the sudden visual feedback distortion resulted in typical spiral-like trajectories, which became straighter by late exposure. However, the final adaptation level was reduced in the aged group, although the aftereffects were similar. When subjects were exposed to the gradual distortion, no statistically significant differences in measures of adaptation with advancing age were found. In this case, both age groups appeared to adapt equally. However, after removal of the distortion, elderly subjects showed reduced aftereffects as compared with the young group. These findings suggest differential effects of aging on adaptation to gradual versus sudden visual feedback distortions, and may help to explain the conflicting results obtained in previous visuomotor adaptation studies.