What is the difference between a bci and a neuroprosthetic device

Thus, the present Research Topic aims to coalesce novel research investigations aimed at using data driven approaches in BCIs and AI for better understanding of cortical function, with the goal of developing closed-loop neuroprosthetic or clinical interventions for mitigating effects of neurological injury.

While broad, this Research Topic requires that potential authors make clear how submitted manuscripts 1 make use of computational or experimental data driven approaches, and 2 could directly lead to clinically relevant neuroprosthetic interventions. Important Note : All contributions to this Research Topic must be within the scope of the section and journal to which they are submitted, as defined in their mission statements.

Frontiers reserves the right to guide an out-of-scope manuscript to a more suitable section or journal at any stage of peer review. With their unique mixes of varied contributions from Original Research to Review Articles, Research Topics unify the most influential researchers, the latest key findings and historical advances in a hot research area!

Find out more on how to host your own Frontiers Research Topic or contribute to one as an author. The control of a robot or prosthetic device and the feedback contingency are of vital importance to enable neuro-motor-rehabilitation. Here, we developed and tested in healthy participants an on-line proprioceptive BCI, closing the loop between brain, movement and proprioception.

The difference between this system and previous studies [20] , [21] consists of the online feedback being proprioceptive feeling the hand moving and visual watching the hand moving during voluntary brain control as opposed to online visual feedback only i. However, with at least some afferent pathways intact, the sensory information to the brain produced by moving the paretic limb engages remaining motor areas in the vicinity of the lesion to control the BCI.

Since EEG has a limited spatial resolution it is difficult to separate activity from somatosensory cortex, premotor or motor cortex even using advanced spatial filtering methods [12] , [24]. From previous work we know that passive movement affects frequency bands in a similar way but somewhat weaker than active movement and motor imagery [25] , [26]. The afferent excitation of the sensorimotor brain through the robotic orthosis produces similar EEG frequency changes.

Only preliminary data are available regarding the use of proprioceptive on-line BCI [27] , [28]. In order to test these alternatives, we developed a sensorimotor rhythm based on-line proprioceptive BCI, linking brain oscillations with a robotic hand orthosis and investigated the effects of proprioception on BCI control. The participants performed five different tasks: 1 motor imagery without any feedback and no movement, 2 motor imagery with proprioceptive feedback of the BCI-dependent movement, 3 passive and 4 active movement without a BCI, and 5 rest, comparable with the major ingredients of rehabilitation therapies for movement disorders.

Participants were sitting in an upright position wearing a channel EEG cap. The experimental protocol was approved by the ethics committee of the University of Tubingen, Medical Faculty. Participants provide their written informed consent to participate in this study. Participants were asked to perform 5 different tasks following 5 randomly presented auditory cues the name of the task taped in advance or from a taped recording of the voice of one of the experimenters :.

The participants were separated in 3 different groups receiving 3 different feedback contingencies. Only during task2 participants used the EEG-based proprioceptive BCI to control the orthosis with opening and closing the hand motor imagery. The first group received contingent positive feedback moving the orthosis with SMR desynchronization in task 2: 9 Participants , the second received contingent negative feedback moving the orthosis with SMR synchronization in task 2: 8 Participants and the third received sham feedback the orthosis moved independently from brain activity but participants believed in their control: 7 Participants.

All auditory cues were normalized in pitch, length and volume. In task 1 and task 2 participants were asked to perform kinaesthetic motor imagery, i. During task 1 no feedback was presented to the participants in contrast to task 2, in which direct visual feedback of the hand, moving and proprioceptive feedback while the hand was moved by the brain-driven orthosis was provided Fig. The participants performed 4 different training sessions at 4 different days completing 10 runs of 25 trials each.

The participants had no prior BCI experience. A Timing of an experimental trial. Each trial starts with a baseline of 3 seconds followed by an auditory instruction period.

B BCI. Participant wearing the EEG channels cap seated with the hand attached to the orthosis showing the components used during all tasks C Close look at the orthosis with the fingers attached. D Schematic of the channels and shaded in grey the 61 channels used during the experiments. In Table 1 a demographic description of the participants is presented together with the oscillation type sensorimotor rhythm synchronization S or desynchronization DS used in each group to compute performance for the five different motor tasks.

BCI performance except for task2 was computed off-line. Only 61 EEG channels over the motor areas on both hemispheres were used recording from pre-motor, motor and parietal areas Fig. Data were sampled at Hz and transferred to a PC for storage and real-time signal processing using the BCI platform www. The motor drove a Bowden cable via cogwheel and cograil. A finger holder was mounted on the other side of each Bowden cable Fig. Close to this finger holder an optical position sensor was mounted to detect the finger position independent of the bowden cable tolerance and elasticity.

Strain gauges were placed on the Bowden cables near the fingers to detect the finger force in order to regulate the motor force to zero no friction for trials with active movement. A closed and an open finger position were predefined individually for each volunteer depending on their hand and finger size.

The BCI system determined the orthosis position and velocity and the device transmitted its actual position and velocity to the host computer upon request.

Once the BCI system sends a position and a velocity command, the orthosis would then initiate a movement to the given position with the given velocity. Movement stopped when either the current position was identical to the position command sent by the BCI system as set in the most recent position command , or when the velocity command was set to zero by BCI system. The direction of the movement was determined by the difference between current and desired position.

As a physical connection between orthosis and host computer, a RS serial connection was used at a speed of bps. The BCI two class classifier motor imagery versus baseline sent an output every 40 ms and five consecutive outputs for the same class were needed in order to send the orthosis a no-move zero velocity value or a move positive velocity command. This time filter was installed to avoid false positives and false negatives. This randomization of the output was identical to the averaged time participants from the contingent positive group achieved to move the orthosis during task 2.

The features to be used by the BCI platform were defined through a visual inspection of the R-squared values [29] obtained when comparing EEG activity during rest versus intention to move hand open and close.

The result was normalized zero mean, unit variance with respect to the inter-trial interval period of each training run.

We defined this final outcome as BCI output. Due to the weights used i. In the online application, a center-surround local spatial filtering approach, in which a radial difference-of-Gaussians function was used to weight the electrodes at each spatial location, was applied to the EEG activity from each electrode. The spatial filtered EEG was modeled as an autoregressive AR process [30] of order 16 over a normalized sliding temporal window of ms shifting every 40 ms and power spectral density of the AR-model for each electrode was computed to calculate the mean SMR-band power in each chosen frequency bin.

The BCI software maintained a history of the mean sensorimotor rhythm amplitude estimate from each trial and assigned this to a distribution representing observations for the two classes rest or motor intention. The classification threshold, defined as the zero mean distance to the two distributions, was adaptive to account for changes in the shapes of these distributions over the course of training.

For EEG off-line analysis we performed a time-frequency analysis using a 1. The event related spectrum perturbation was then calculated using Morlet transforms [31] with 3 cycles at lowest frequencies and The EMG data were filtered using a high pass filter at 10 Hz, bipolarized, rectified and visually inspected. Trials presenting muscle activity during the resting task or absence of activity during active opening and closing of the hand were excluded from the event related spectrum perturbation analysis.

One EEG-screening was performed the day before the first training session and was used as a calibration session to identify the best features electrodes and frequency bins to be used by the BCI classifier. In this screening session the participants were randomly presented with visual and auditory cues corresponding to 3 different tasks indicating to either relax task 1 , actively open and close the right task 2 or the left hand task 3.

After a 5 s period performing the tasks a rest cue was presented indicating to stop. The inter-trial-interval time was randomized between 5 and 7 s. The participants underwent 4 to 5 runs of 25 trials.

The features to be used by the BCI platform were defined through a visual inspection of the R-square [29] values obtained when comparing EEG activity during rest versus intention to move hand open and close. The power in the electrodes and frequency bins with highest R-square values were identified as customized sensorimotor rhythm SMR features and used as input for the classifier.

The group matching was performed based on age, handedness and the R-squared values obtained comparing the distribution of data during the screening session rest versus hand motor imagery. After the screening, a cursor control training was performed at the end of the same session, to familiarize the participants with the BCI.

For the cursor control training session participants controlled the velocity in the Y axis of a cursor moving from left to right on the screen at a constant speed trying to reach a target presented at the right side of the screen. The participants performed 4 runs containing 12 trials. The participants arrived at 4 consecutive days to perform one session every day. In every session the participants were presented with the 5 different tasks described before.

We analyzed how the BCI output changes during the different tasks and investigated the effect of the feedback contingency on BCI control. In addition to the online classification translated into orthosis movements task 2 , we simulated the performance the participants would have obtained if the orthosis would have moved during every motor task in an online setup.

For example, how would the brain activity elicited during passive movement have been classified, if the classifier set up for motor imagery same electrodes and frequency bins would have been used to move the orthosis.

Furthermore, several performance measures indicating different aspects of the SMR modulation were calculated off-line for all the tasks:. These performance measures were calculated simulating an online scenario before and after EEG and EMG artefact removal to explore the influence of data contamination and the importance of implementing on-line artefact removal filters.

We assumed that the proprioceptive feedback is felt by the user as number of times they can make the orthosis switch from not moving to moving number of orthosis moving onsets , how fast they can start moving the orthosis onset latency , percent of time the orthosis is moving during the trial and maximum consecutive time they moved the orthosis.

We expected to observe learning effects in the contingent feedback groups negative and positive during the two tasks involving motor imagery with and without feedback. For all performance measures there were none or only few violations of the Levene-tests. Since the number of participants was less than 10 in each group and the number of performed tests was 20, slight violations were ignored and the error variances were assumed to be homogeneous.

Neuroprosthetics is an area of neuroscience concerned with neural prostheses — using artificial devices to replace the function of impaired nervous systems or sensory organs. The most widely used neuroprosthetic device is the cochlear implant, which was implanted in approximately , people worldwide as of There are also several neuroprosthetic devices that aim to restore vision, including retinal implants, although this article only discusses implants directly into the brain.

The differences between BCIs and neuroprosthetics are mostly in the ways the terms are used: neuroprosthetics typically connect the nervous system, to a device, whereas the term "BCIs" usually connect the brain or nervous system with a computer system.

Practical neuroprosthetics can be linked to any part of the nervous system, for example peripheral nerves, while the term "BCI" usually designates a narrower class of systems which interface with the central nervous system.