Journal of Neurorestoratology 2020, 8(1): 12–25

DOI 10.26599/JNR.2020.9040001 ISSN 2324-2426

REVIEW ARTICLE

State-of-the-art non-invasive brain–computer interface for neural

rehabilitation: A review
Miaomiao Zhuang1,5,§, Qingheng Wu2,5,§, Feng Wan3, Yong Hu4,5 ()

1 Xiangya School of Medicine, Central South University, Changsha 410013, Hunan, China
2 Department of Dentistry, Nanjing Medical University, Nanjing 211166, Jiangsu, China
3 Department of Electrical and Computer Engineering, Faculty of Science and Technology, University of Macau, Macau, China

4 Shenzhen Key Laboratory for Innovative Technology in Orthopaedic Trauma, The University of Hong Kong-Shenzhen Hospital,

Shenzhen 518053, China

5 Department of Orthopaedics and Traumatology, The University of Hong Kong, Pokfulam, Hong Kong, China
§
These authors contributed equally to this work.

ARTICLE INFO ABSTRACT

Received: 9 October, 2019 Brain–computer interface (BCI) is a novel communication method
Revised: 2 February, 2020 between brain and machine. It enables signals from the human brain to
Accepted: 5 February, 2020 influence or control external devices. Currently, much research interest
© The authors 2020. This is an is focused on the BCI-based neural rehabilitation of patients with motor
open access article under the CC
BY‐NC License and cognitive diseases. Over the decades, BCI has become an alternative
(http://creativecommons.org/li- treatment for motor and cognitive rehabilitation. Previous studies
censes/by‐nc/4.0/)
demonstrated the usefulness of BCI intervention in restoring motor
KEYWORDS function and recovery of the damaged brain. Electroencephalogram
brain–computer interface (EEG)-based BCI intervention could cast light on the mechanisms
(BCI); underlying neuroplasticity during upper limb recovery by providing
neural rehabilitation; feedback to the damaged brain. BCI could act as a useful tool to aid
stroke; patients with daily communication and basic movement in severe motor
spinal cord injury (SCI); loss cases like amyotrophic lateral sclerosis (ALS). Furthermore, recent
amyotrophic lateral sclerosis findings have reported the therapeutic efficacy of BCI in people suffering
(ALS); from other diseases with different levels of motor impairment such as
attention-deficit/hyperactivity
spastic cerebral palsy, neuropathic pain, etc. Besides motor functional
disorder (ADHD)
recovery, BCI also plays its role in improving the behavior of patients
with cognitive diseases like attention-deficit/hyperactivity disorder
(ADHD). The BCI-based neurofeedback training is focused on either
reducing the ratio of theta and beta rhythm, or enabling the patients to
regulate their own slow cortical potentials, and both have made progress
in increasing attention and alertness. With summary of several clinical
studies with strong evidence, we present cutting edge results from the
clinical application of BCI in motor and cognitive diseases, including
stroke, spinal cord injury, ALS, and ADHD.

Corresponding author: Yong Hu, E-mail: [email protected]

Journal of Neurorestoratology 13

1 Introduction children with neurodevelopmental problems, such

as autism [12] and attention-deficit/hyperactivity
Brain–computer interface (BCI) provides human disorder (ADHD) [13]. By reducing the ratio of
beings with an alternative pathway between the theta and beta rhythms (theta/beta ratio, TBR) in
brain and external environment. During the past electroencephalogram (EEG), BCI-based neural
two decades, BCI related research has become feedback training improves participants’ focused
extremely active, with articles in this field attention [14, 15]. Research that enables patients
emerging in quantity [1]. This novel technology to control slow cortical potential also improves
has aroused wide interest of scientists, clinical their cognitive performance [16].
doctors and the public for its multiple application. In the present review, first we explain the
Presently, people have managed to apply BCI principle of BCI and briefly introduce the frontier
to assist communication [2], provide control [3], technologies used in BCI system. Then, we discuss
restore motor functions [4] and even strengthen the prevailing clinical application of BCI in neural
brain abilities [5]. Especially, the therapeutic disease recovery, including motor injury and
application of BCI is one of the most exciting and cognitive disaster. We also clarify some of the basic
prospective aspects. BCI has huge potential in mechanism of neural rehabilitative therapies.
the recovery of motor diseases like stroke, spinal Finally, we discuss some prevalent defects in
cord injury (SCI) and amyotrophic lateral sclerosis existing research and the direction of future
(ALS). studies.
As was reported by the American Heart
Association, for patients with complete and chronic 2 Principle of BCIs
motor injuries, the prognosis for recovery is
poor [6]. Traditional therapies for those severely A standard BCI system comprises five functional
paralyzed patients often rely on pharmacology, modules, namely, brain signal acquisition, signal
yet they cannot restore or enhance the neural pre-processing, feature extraction, feature
pathways [6]. BCI intervention, however, may classification, and output devices. Using these
become a promising alternate neurorehabilitative modules, BCI systems can decode a user’s brain
approach. For people with neuromuscular signal and translate it into computer commands
diseases, BCI-controlled prosthetic devices can for the control of external devices or neural
compensate for their loss of motor abilities, and rehabilitation systems.
assist the patients to carry out simple movements The signal acquisition module records, amplifies,
[4]. Besides, a growing body of evidence has and digitizes the user’s brain signal, which can be,
shown that the presence of BCI reduces for example, the specified EEG rhythm closely
neurological deficits and restore the damaged related to motor execution/imagination. Brain
sensorimotor loop [7]. After treatment, patients signals can be recorded from the central nervous
with stroke [8] and acute SCI [7] showed system (CNS) via two main approaches: invasive
great functional and neurological recovery. This and non-invasive methods. In addition to motor-
encouraged researches aimed at developing BCI- induced brain signals, visual and/or muscle-
based neurorehabilitative programs, and many induced signals have also been investigated [1].
of them did manage to improve the performance Generally, the recorded brain signals (e.g.
of patients [9–11]. EEG signals) are contaminated with artifacts
BCI can also serve as an assistive therapy for or noise, such as electromyography (EMG),
14 Journal of Neurorestoratology

electrooculography (EOG), electrocardiography EEG-related features [17, 21]. The commonly

(ECG) artifacts, and power-line noise. The main used classifiers are based on linear methods, such
purpose of signal pre-processing is to enhance as linear discriminant analysis and the support
the signal-to-noise ratio (SNR) of the recorded vector machine, as linear methods can perform
brain signals. According to the artifacts’ charac- relatively stable calculations when the labeled
teristics, temporal filtering, spectral filtering, and data and knowledge about the BCI task are
spatial filtering techniques are usually employed limited [22]. Recently, deep neural networks have
to remove artifacts or noise in temporal, spectral, also been applied to classifying EEG features in
and spatial domains, respectively. For example, BCI applications [17].
power-line noise can be removed by spectral Actions are triggered via output devices, which
filtering. EOG artifacts, which are caused by allow users to act on the real world, regardless
blinking and eye movements, generally induce of the participation of peripheral nerves and
high-amplitude patterns in the low-frequency muscles. Actuators include functional electrical
region. Spatial filtering, which linearly combines stimulation (FES), nerve stimulation, prosthetic
multi-channel EEG signals to improve the SNR, devices, and exoskeleton.
can be performed using a data-independent or FES innervates muscles through electrical
data-driven approach. The data-independent stimulation, which also enables proprioceptive
approach uses the common average reference and feedback to the brain. A robotic prosthesis, such
a Laplacian filter, and the data-driven approach as an avatar arm, works better for patients in
uses common spatial pattern, canonical correlation acute phases of motor impairment and initiates
analysis, independent component analysis, or passive movement. In recent years, hybrid-FES
principal component analysis [17, 18]. After signal systems consisting of FES and active orthotic
pre-processing, the SNR of the recorded brain components are being more widely used as an
signals can be enhanced. effective way to restore everyday manipulation
Feature extraction is used to extract features capabilities [23].
from the brain signals, while those features are In addition to being the intended outcome,
representative of the intention. Two widely used movement also provides feedback to the brain,
features are frequency band power features and which in turn modifies the process of motor
temporal features. Frequency band power features execution to improve speed and accuracy [24, 25].
are the power of the EEG signals within a specified In the most desirable case, the brain will also
frequency band (or rhythm), such as alpha, beta, modify signal features itself, forming bi-directional
mu, and gamma bands. For example, changes in adaptation [24].
the mu/beta rhythm over the sensorimotor cortex
are an essential feature in motor imagery-based 3 Clinical application of BCI
BCIs [19, 20]. Temporal features play an important
role in P300-based (or event-related potential) Several studies have demonstrated the potential
BCIs, in which changes in EEG potentials are of BCI-based functional restoration in patients
time-locked to a given event or stimulus [3]. with different levels of motor and cognitive
Feature classification is applied to recognize impairment. The common paradigm of BCI-based
EEG features that are extracted by the feature functional restoration therapy captures motor
extraction module. Over the past few decades, intention as signals, then triggers real movement
various algorithms have been developed to classify through approaches such as FES and robotic

Journal of Neurorestoratology
Journal of Neurorestoratology 15

devices with distinctive features that are tailored design for BCI rehabilitation is the precise coupling
according to the nature of the disease. between the brain regions and functions regarding
the behavioral, clinical, and neurophysiological
3.1 BCI in stroke rehabilitation
changes.
As the second most common cause of death, The motor rehabilitation process also provides
stroke involves about 15 million patients per feedback to the patients’ brains, either through
year worldwide [6, 26], while pharmacological proprioception with muscle movement or action
therapy is the traditional treatment method to observation (Fig. 1). Successful rehabilitation of
reduce symptoms. Encouragingly, recent studies patients with stroke requires brain activity feedback
have shown that BCI is a promising technology to be paired with its initial movement intention,
for rehabilitating volitional motor capacity in thus forming a closed-loop system. BCI-assisted
survivors of chronic stroke [27–29]. BCI inter- movement enhances the closed-loop circuit by
vention after stroke can be used for both BCI- providing valuable information about brain activity
assisted rehabilitation and as a decision-making and physical responses to CNS. This benefits CNS
guide for intervention [30]. Application of BCI plasticity and leads to the restoration of normal
in stroke treatment aims to restore motor and brain function or a relocation of functional control
cognitive abilities, to enhance neuroplasticity to undamaged brain areas [30].
during rehabilitation [25]. While the positive function of proprioceptive
The clinical efficacy of BCI treatment has been feedback in restoring neuroplasticity has been
shown in a cohort of patients with chronic stroke verified, a recent study conducted by Vourvopoulos
through Fugl-Meyer motor assessment [31], proved et al. combined the principles of virtual reality
significant neurophysiological and behavioral (VR) and BCI as a combination of movement
improvement in patients with acute stroke feedback [35]. The authors measured EEG
after BCI intervention. BCI related targeted and EMG as signals of the motor intention to
neurofeedback (NF) impact the brain structure drive commands and initiate the movement of
and function, which could be rapidly detected a virtual presentation of an avatar arm, while
by magnetic resonance imaging (MRI). The the observation and passive movement process
BCI-induced spatial specific brain plasticity of the visible VR also providing feedback to the
promises that therapeutic interventions can be patient’s brain imagination. The study proposed
tailored to post-stroke functional deficits [32]. a useful NF VR-BCI paradigm for efficient motor
EEG-based motor imagery (MI) BCI systems rehabilitation following stroke, and demon-
have also been found to restore motor function. strated its feasibility as an alternative tool for
MI can be evoked by motor attempts without real individuals without proprioceptive feedback
movement, and may thus serve as an adjunct to pathways. Furthermore, patients with more severe
traditional physical rehabilitation for patients motor impairment achieved a better recovery
with severe stroke [33]. During motor tasks, EEG through EEG-based BCI, while those with active
records alpha and beta rhythms that present movement benefited more from EMG feedback
event-related desynchronization/synchronization in a multimodal platform. This finding could
(ERD/ERS) trends, which provides insights into inform the future selection of signals in patients
the mechanisms underlying neuroplasticity in with different stage of stroke to indicate more
stroke recovery [34]. The protocol of personal precise information of brain activities. Additionally,
16 Journal of Neurorestoratology

Fig. 1 Motor feedback in BCI-assisted rehabilitation. The basic parts of BCI-assisted rehabilitation include signal input and
movement output of the BCI-system to help complete the intended action. In the process, the observation or passive
movement could provide feedback to the initial movement intention to the brain which has been proved beneficial to the
neurological rehabilitation and modification of the movement signal.

Fig. 2 Neuronal feedback in neurorehabilitation. The process of movement provides visual, auditory, as well as proprioceptive
feedback to the brain, serving as signals in modulation and adjustment of the next-step movement. The signals also close the
sensorimotor loop, following the principle of neuroplasticity and enhancing the remaining neural pathways, thus
promoting neurorehabilitation.

studies on evaluation of brain connectivity indices mainly as alternative trials to restore motor
also demonstrated the therapeutic possibilities function, whereas for those completely locked-in,
of BCI intervention in cognitive rehabilitation BCI interventions focus on communication and
after stroke [36]. environmental control.
Among the majority of patients suffering from
3.2 BCI in partial motor impairment
partial motor impairment symptoms, upper limb
For partially paralyzed patients, BCI functions disability is particularly critical because of its high

Journal of Neurorestoratology
Journal of Neurorestoratology 17

prevalence and significant effects on patients’ array with invasive device, an exciting animal
daily life [37, 38]. Patients with chronic stroke, trial of epidural electrical stimulation (EES) on
SCI, and traumatic brain injury often experience interfaced leg motor cortex has showed the
motor impairment of the upper extremities (UE) recovery of their adaptive locomotion after
[39, 40]. Addressing the upper limb disability SCI [49].
will help individuals to improve grasp and The advantage of BCI-based rehabilitation is
precise movement abilities, which will facilitate that neurorecovery is correlated with the gain of
independent living and even a return to work. function, which seems to be a promising approach
The current treatment of upper extremity for future long-term recovery method compared
disfunction is primarily an implanted BCI- with conventional symptom-only treatment.
controlled neuroprosthesis, which deciphers neural A recent longitudinal analysis showed stronger
signals and translates them into outputs such as correlations between ERD/ERS in beta rhythms
FES, external prosthetic devices, and a robotic and the time since stroke onset compared with
exoskeleton [3, 41–43]. BCI triggers FES-based those seen in alpha rhythms. This could be
muscle movement or the replacement of arm explained by the previous findings that beta
function using a robotic prosthetic arm, thus activations are more closely related to a closed-
restoring upper limb function. A previous study loop communication between the sensorimotor
with sub-analysis reported that a BCI interven- cortex and the paralyzed upper limb [50], while
tion regimen resulted in more improvement alpha rhythms are more closely related to motor
and proved effective for the treatment of UE learning mechanisms, which indicates the sen-
impairment in stroke survivors [44]. Another sorimotor cortex is being activated and rebuilt
preliminary study has also demonstrated the during the BCI-based treatment. Therefore,
therapeutic feasibility of BCI in patients with EEG-based BCI intervention cast light on the
persistent neuropathic pain [11]. mechanisms underlying neuroplasticity during
In addition, BCI can be applied in lower limb upper limb recovery; namely, the related findings
motor assistance and rehabilitation. Some studies reinforce the hypothesis that more cortical regions
proved the usefulness of exoskeletons for walking such as frontal temporal and parietal regions, are
assistance, mainly for SCI population [45, 46]. involved in movement tasks to compensate for
Since 2014, several exoskeletons have been severe motor impairment [34].
approved by the USA Food and Drug Adminis- Driven by the technological advances in
tration (FDA) for clinical use, including the most multidisciplinary areas such as virtual and
studied powered exoskeletons [47]. Compared augmented reality (VR/AR), robotics, various BCIs,
to traditional physical maneuvers, BCIs can as well as medication, motor rehabilitation is
theoretically better enhance the robotic systems now a fast-growing field. However, BCI-based
because they work in line with natural human treatment paradigm for motor impairment still
movement. They can be used to bypass the lesion needs to be refined, improved, and personalized
and reconnect the brain and lower extremities. in many ways. First, 50% of functional electrical
Presently, EEG is the primary approach to capture stimulation does not activate the targeted muscle
motor intention in human research. Yet, the due to damage of the innervating motor neurons.
decoder in BCI system must be able to recognize Second, because of the nature of the academic
false alarm, like nodding, to avoid undesired research environment, some motor imaginaries
movement [48]. Using intracortical microelectrode are required to be made rather than volitional
18 Journal of Neurorestoratology

daily movement, making precise signal acquisition as ADHD. Unlike stimulant treatment, which
impractical for real-life application. Therefore, commonly triggers side effects, NF therapy is
future research should aim to identify more associated with relatively few adverse events
realistic control commands, and decode more [56]. Furthermore, a recent analysis demonstrated
precise and complex movement signals to reassure that BCI-based NF therapy showed longer-term
everyday usage. effects than medication [57].
While a previous review refuted the efficacy of
3.3 BCI in cases of severe loss of motor
NF therapy [58], the better effects could be a result
function
of the small sample sizes and methodological
For patients that are completely locked-in, BCIs defects. Further randomized controlled trials have
allow patients to convey messages and commands revealed a significant effect of neurofeedback
to the external world. Preliminary studies have training (NFT) on inattention, impulsivity, and
shown the efficacy of BCI as a useful tool to aid hyperactivity, and NFT is therefore considered
patients with ALS in daily communication and to result in clinically meaningful improvements
basic movement [30, 51–53]. A follow-up study [56, 59].
of 42 patients with ALS also demonstrated the At present, there are two main types of NF
feasibility of independent home use of BCI protocols, as follows: the reduction of the TBR,
for patients with ALS, providing relief for both and the self-regulation of slow cortical potentials
caregivers and patients [52]. (SCPs).
Furthermore, recent findings have reported the Some studies have focused on detecting the
therapeutic efficacy of BCI-FES in children with ratio of theta and beta rhythms for the diagnosis
spastic cerebral palsy. This highlights the clinical of ADHD. A meta-analysis looked into quantitive
possibilities of BCI in provoking contraction of EEG studies that evaluated ADHD by the same
denervated muscles, restoring paralyzed motor criteria of Diagnostic and Statistical Manual of
function, and treating muscle weakness [51]. Mental Disorders, Fourth Edition (DSM-IV), and
There are two major limitations in using a confirmed that an elevated TBR is a typical trait
BCI system with locked-in patients. First, the in ADHD [60]. Nevertheless, a similar change in
insufficient BCI accuracy, particularly in spelling, theta power is found in many other cognitive
means that the basic communicational needs disorders, and cannot be regarded as pathogno-
of patients are still not met. Ongoing studies to monic of any specific disorder [61]. As such,
increase the speed and accuracy of BCI-based further research to investigate the combination
communication are conducted [54]. In addition, of EEG techniques and other diagnosis is needed,
long-term use of BCI devices may cause side in order to meet the need of clinical application.
effects such as impaired vision. One way to Recent studies have also revealed that NF
overcome this limitation could be to use auditory which aims to reduce the TBR can increase focused
rather than visual stimuli [52]. attention and improve alertness, although it does
not have a specific effect on hyperactivity or
3.4 BCI in ADHD
impulsivity [14, 15]. Yet, the efficacy and specificity
Medication and behavioral therapy are the most need to be measured by additional larger-scale
common treatments for ADHD [55]. In recent studies.
years, BCI-based NF therapies have been proposed Some studies have investigated the ability for
as a new treatment approach for disorders such NFT to modify patients’ abilities to control SCP.

Journal of Neurorestoratology
Journal of Neurorestoratology 19

Positive and negative SCP-shifts are related to BCI paradigm concerns two different ways:
cortical inhibition and activation, respectively. detouring around breaks in neural pathways
Most training is focused on practicing and rein- and muscles, and directly using brain signals for
forcing the negative polarity concerning neural communication and control; and following the
changes in the brains of paitients with ADHD. principle of neuroplasticity and enhancing the
A preliminary study detected an enhancement remaining neural pathways [62].
of contingent negative variation amplitude in an In this part of the review, we will describe
attention task [10]. Moreover, it has been reported several basic concepts concerning the use of
that after 30 sessions of training about SCP self- BCI in neurorehabilitation. First, we will explain
control, children with ADHD managed to regulate that neuroplasticity is the internal mechanism
negative slow cortical potentials, and all impro- underlying functional recovery. Then, we
vements in behavior, attention and IQ score demonstrate how the implementation of motor
were still evident at the 6-month follow up [16]. imagery can promote the recovery of movement
Following this progress, researchers are now disorders. Finally, we will discuss the role of BCI
looking to develop an independent home-used signals in the closure of the sensorimotor loop.
training program. Some researchers have designed
4.1 Neuroplasticity
a feasible BCI-based training method for ADHD
[9]. In this training program, participants’ Neurofeedback is a sort of biofeedback that
brainwaves were recorded before and after training depends on real-time cortical activity. NF aims to
to create an individualized EEG pattern. According teach patients to regulate brain function, which
to the assessment of ADHD rating scale and child is reflected in certain EEG features, and can
behavior checklist, inattentiveness was slightly enhance neuroplasticity and restore cognitive
improved in the intervention group, as expected. abilities. BCI can analyze neural activities and
Furthermore, the researchers stated that the provide real-time feedback, and have therefore
program is feasible; 24 sessions over an 8-week been widely employed in NF therapy as a
period was manageable for parents and resulted rehabilitation tool for motor injuries [63].
in a low dropout rate. Neuroplasticity is grounded in the concept of
NF therapy triggers few adverse effects and Hebbian plasticity, that is, synapses increase their
little discomfort. It’s relatively safe and proved efficiency when they are persistently stimulating
to be effective, and so is a promising potential the postsynaptic cells. Neuroplasticity includes
adjunctive therapy for ADHD. However, NF multiple regenerative processes, such as axonal
devices that are easier to use, standardized training sprouting of surviving neurons, molecules released
programs and uniformed regulations should be by neurons, and the formation of new synapses
developed before BCI-NF therapy is used in [64]. It is widely acknowledged that functional
clinical practice. Further research should investigate impairment results in a reorganization of the
the combination of NF, pharmacology, and central nervous system, and the success of a new
behavioral therapy in clinical treatment. therapy is linked to its ability to interact with the
new brain architecture after neural injury [65].
BCI-based interventions combined with feedback
4 Mechanisms of BCI-based
give rise to a persistent change in cortical connec-
neurorehabilitation
tivity, and are thus applicable in cognitive and
Like all other neurorehabilitative methods, the motor recovery, such as stroke [66, 67] and SCI
20 Journal of Neurorestoratology

[68]. Especially, robot-assisted rehabilitation serves short enough to synchronize with brain activities,
as a way of neuroplasticity improvement, which which makes it hard to result in long-term neural
is currently the main recovery of stroke [34]. changes [63]. Further research should investigate
Furthermore, BCI-NF has also been found to the maximum feedback delay that meets the
normalize functional networks in children with requirement of clinical application.
ADHD [5]. There are many possible feedback pathways in
the application of neuroprostheses [69]. To date,
4.2 Closed-loop neurostimulator: a compensation most research interest is focused on implanted
for the damaged loop
devices, especially deep brain stimulation (DBS).
To date, there are two main protocols for using DBS uses high-frequency electrical pulses to
BCI in neurorehabilitation. One is to elicit neural normalize the neuronal activity of a specific
activity restoration, and the other is to control nucleus. It is a widely used therapy to treat
external devices, such as a neuroprosthesis. These movement and psychiatric disorders, such as
protocols both rely on the dynamical closed- Parkinson’s disease, dystonia, and major depressive
loop system, whereby feedback is essential to disorder. A recent trial developed a closed-loop
respond to the input stimulation and control the system with implanted electrodes (both recording
output effect [69]. This closed-loop in neural and stimulating), whereby the feedback electrical
rehabilitation has inspired individualized therapies signal was transmitted directly from the robotic
that adapt to the BCI user’s body situation and effector to the controller [74]. However, neural
brain activity. tissue damage is unavoidable when using invasive
At present, multiple methods to reestablish the electrodes, and can include bleeding, inflammation,
damaged sensorimotor loop have been adopted. physical trauma, and neuron damage [75].
Many BCI systems use visual or auditory signals Research into non-invasive recovery therapies
as feedback [70]. Visual elements, including a that are based on a closed-loop system have
computer cursor and flickering lights, have been also made progress. Using an augmented-reality
widely used as a feedback signal. However, the approach [69], some BCI techniques, such as EEG
high brightness and long training period can cause or fMRI, can acquire cortical signals from the
fatigue and inattention among participants, which surface or from the scalp, which is safer and more
thus undermines the performance of the BCI likely to be accepted by users.
system [71]. Furthermore, these conventional
modalities are not suited to patients who are 5 Conclusion
completely paralyzed and can’t get their eyes
moved, or patients with advanced ALS who may The preliminary studies presented here demon-
have visual problems. strate the therapeutic and clinical potential of
However, haptic information channels, such BCI-based applications, despite the presence of
as vibrotactile feedback, can overcome such neutral and negative findings. BCI could serve as
limitations. Unlike visual signals, haptic feedback an alternative therapy for neurological disorders,
is simple and doesn’t need much attention [72]. with fewer side effects compared with traditional
Peripheral nerve interfaces can now provide clinical treatment.
long-term, natural touch sensations, and enable We have also summarized some common
subjects to manipulate objects [73]. Yet, in many problems of current existing BCI systems and
current studies, the haptic feedback delay is not addressing these questions will expand the clinical

Journal of Neurorestoratology
Journal of Neurorestoratology 21

potential of BCIs and guide future research:

Conflict of interests
First, due to small sample sizes, some studies
did not meet the traditional mathematical standard The authors declare no conflict of interests in this
for statistical significance, and some only found work.
weak positive evidence after BCI intervention
which were not persuasive enough to demonstrate
Acknowledgments
the clinical efficiency. More studies with larger
sample sizes and more diverse patient populations This work is supported by High Level-Hospital
[35] are required to confirm and increase the Program, Health Commission of Guangdong
reliability of previous results. Province, China (HKUSZH201902033).
Second, clinical trials have adopted various
experimental designs and included patients with References
heterogeneous types of motor impairment, which
makes it harder to assess the efficiency of BCI [1] Shih JJ, Krusienski DJ, Wolpaw JR. Brain-computer
interventions [34]. To overcome this limitation, interfaces in medicine. Mayo Clin Proc. 2012, 87(3):
more comprehensive BCI designs that include a 268–279.
[2] Brumberg JS, Nieto-Castanon A, Kennedy PR, et al.
more balanced sample of patients with different
Brain-computer interfaces for speech communication.
degrees of impairment are needed through
Speech Commun. 2010, 52(4): 367–379.
multi-center collaboration [52, 76], which would
[3] Hochberg LR, Bacher D, Jarosiewicz B, et al. Reach
eliminate irrelevant variables. Moreover, because and grasp by people with tetraplegia using a neurally
of ethical or practical obstacles, it is difficult to controlled robotic arm. Nature. 2012, 485(7398):
carry out a blinded experiment. Subsequent trials 372–375.
should increase the level of blinding to eliminate [4] Bertani R, Melegari C, De Cola MC, et al. Effects
the possibility of confounding variables, for of robot-assisted upper limb rehabilitation in stroke
example the placebo effect. patients: a systematic review with meta-analysis.
Third, previous data have been analyzed using Neurol Sci. 2017, 38(9): 1561–1569.
different scales and standards, making it difficult [5] Qian X, Loo BRY, Castellanos FX, et al. Brain-
computer-interface-based intervention re-normalizes
to perform post hoc analyses [76]. Concerning
brain functional network topology in children with
patient recruitment, strict inclusion and exclusion
attention deficit/hyperactivity disorder. Transl Psychiatry.
criteria should be followed. Forthcoming studies
2018, 8(1): 149.
are expected to improve the quality through [6] Benjamin EJ, Blaha MJ, Chiuve SE, et al. Heart
standardization of trials design and measurement disease and stroke statistics-2017 update: A report from
to assist their translation into clinical practice. the American Heart Association. Circulation. 2017,
Finally, the despite positive results from phase 135(10): e146–e603.
I/II trials, the lack of phase III trials has prevented [7] Curt A, van Hedel HJ, Klaus D, et al. Recovery from
BCI-based therapies from being widely used. a spinal cord injury: significance of compensation,
Therefore, more work should be done to optimize neural plasticity, and repair. J Neurotrauma. 2008,
25(6): 677–685.
the BCI system and further confirm its clinical
[8] Pichiorri F, Morone G, Petti M, et al. Brain-computer
feasibility. There should also be a focus on more
interface boosts motor imagery practice during stroke
personalized BCI systems that are tailored to
recovery. Ann Neurol. 2015, 77(5): 851–865.
the patients’ clinical and neurophysiological [9] Lim CG, Poh XWW, Fung SSD, et al. A randomized
characteristics. controlled trial of a brain-computer interface based
22 Journal of Neurorestoratology

attention training program for ADHD. PLoS One. [21] Lotte F, Congedo M, Lécuyer A, et al. A review of
2019, 14(5): e0216225. classification algorithms for EEG-based brain-computer
[10] Takahashi J, Yasumura A, Nakagawa E, et al. Changes interfaces. J Neural Eng. 2007, 4(2): R1–R13.
in negative and positive EEG shifts during slow [22] Müller KR, Anderson CW, Birch GE. Linear and
cortical potential training in children with attention- nonlinear methods for brain-computer interfaces.
deficit/hyperactivity disorder: a preliminary investigation. IEEE Trans Neural Syst Rehabil Eng. 2003, 11(2):
Neuroreport. 2014, 25(8): 618–624. 165–169.
[11] Yoshida N, Hashimoto Y, Shikota M, et al. Relief of [23] Rupp R, Kreilinger A, Rohm M, et al. Development
neuropathic pain after spinal cord injury by brain- of a non-invasive, multifunctional grasp neuroprosthesis
computer interface training. Spinal Cord Ser Cases. and its evaluation in an individual with a high spinal
2016, 2: 16021. cord injury. In 2012 Annual International Conference
[12] Friedrich EV, Suttie N, Sivanathan A, et al. Brain- of the IEEE Engineering in Medicine and Biology
computer interface game applications for combined Society, San Diego, USA, 2012, pp 1835–1838.
neurofeedback and biofeedback treatment for children [24] Wolpaw JR, Birbaumer N, McFarland DJ, et al.
on the autism spectrum. Front Neuroeng. 2014, 7: 21. Brain-computer interfaces for communication and
[13] Lim CG, Lee TS, Guan C, et al. A brain-computer
control. Clin Neurophysiol. 2002, 113(6): 767–791.
interface based attention training program for treating
[25] McFarland DJ, Daly J, Boulay C, et al. Therapeutic
attention deficit hyperactivity disorder. PLoS One.
applications of BCI technologies. Brain Comput
2012, 7(10): e46692.
Interfaces (Abingdon). 2017, 47(1/2): 37–52.
[14] Hillard B, El-Baz AS, Sears L, et al. Neurofeedback
[26] Béné R, Beck N, Vajda B, et al. Interface providers in
training aimed to improve focused attention and
stroke neurorehabilitation. Period Biol, 2012, 114(3):
alertness in children with ADHD: a study of relative
403–407.
power of EEG rhythms using custom-made software
[27] Song J, Young BM, Nigogosyan Z, et al. Characterizing
application. Clin EEG Neurosci. 2013, 44(3): 193–202.
relationships of DTI, fMRI, and motor recovery in
[15] Bakhshayesh AR, Hänsch S, Wyschkon A, et al.
stroke rehabilitation utilizing brain-computer interface
Neurofeedback in ADHD: a single-blind randomized
technology. Front Neuroeng. 2014, 7: 31.
controlled trial. Eur Child Adolesc Psychiatry. 2011,
[28] Song J, Nair VA, Young BM, et al. DTI measures
20(9): 481–491.
[16] Strehl U, Leins U, Goth G, et al. Self-regulation of track and predict motor function outcomes in stroke
slow cortical potentials: a new treatment for children rehabilitation utilizing BCI technology. Front Hum
with attention-deficit/hyperactivity disorder. Pediatrics. Neurosci. 2015, 9: 195.
2006, 118(5): e1530–e1540. [29] Muralidharan A, Chae J, Taylor DM. Extracting
[17] Lotte F, Bougrain L, Cichocki A, et al. A review of attempted hand movements from EEGs in people with
classification algorithms for EEG-based brain-computer complete hand paralysis following stroke. Front
interfaces: a 10 year update. J Neural Eng. 2018, Neurosci. 2011, 5: 39.
15(3): 031005. [30] Huggins JE, Guger C, Ziat M, et al. Workshops of the
[18] Bashashati A, Fatourechi M, Ward RK, et al. A Sixth International Brain–Computer Interface Meeting:
survey of signal processing algorithms in brain- brain–computer interfaces past, present, and future.
computer interfaces based on electrical brain signals. Brain-Comput Interfaces. 2017, 4(1/2): 3–36.
J Neural Eng. 2007, 4(2): R32–R57. [31] Mrachacz-Kersting N, Jiang N, Stevenson AJ, et al.
[19] Ramoser H, Müller-Gerking J, Pfurtscheller G. Optimal Efficient neuroplasticity induction in chronic stroke
spatial filtering of single trial EEG during imagined patients by an associative brain-computer interface. J
hand movement. IEEE Trans Rehabil Eng. 2000, Neurophysiol. 2016, 115(3): 1410–1421.
8(4): 441–446. [32] Nierhaus T, Vidaurre C, Sannelli C, et al. Immediate
[20] Pfurtscheller G, Neuper C. Motor imagery and direct brain plasticity after one hour of brain-computer
brain-computer communication. Proc IEEE. 2001, interface (BCI). J Physiol. 2019, in press, DOI
89(7): 1123–1134. 10.1113/JP278118.

Journal of Neurorestoratology
Journal of Neurorestoratology 23

[33] Crosbie JH, McDonough SM, Gilmore DH, et al. The 12: 752.
adjunctive role of mental practice in the rehabilitation [45] Guanziroli E, Cazzaniga M, Colombo L, et al. Assistive
of the upper limb after hemiplegic stroke: a pilot study. powered exoskeleton for complete spinal cord injury:
Clin Rehabil. 2004, 18(1): 60–68. correlations between walking ability and exoskeleton
[34] Carino-Escobar RI, Carrillo-Mora P, Valdés-Cristerna control. Eur J Phys Rehabil Med. 2019, 55(2):
R, et al. Longitudinal analysis of stroke patients’ brain 209–216.
rhythms during an intervention with a brain-computer [46] Louie DR, Eng JJ, Lam T, et al. Gait speed using
interface. Neural Plast. 2019, 2019: 7084618. powered robotic exoskeletons after spinal cord injury:
[35] Vourvopoulos A, Pardo OM, Lefebvre S, et al. Effects a systematic review and correlational study. J Neuroeng
of a brain-computer interface with virtual reality (VR) Rehabil. 2015, 12: 82.
neurofeedback: a pilot study in chronic stroke patients. [47] Contreras-Vidal JL, A Bhagat N, Brantley J, et al.
Front Hum Neurosci. 2019, 13: 210. Powered exoskeletons for bipedal locomotion after
[36] Toppi J, Mattia D, Anzolin A, et al. Time varying spinal cord injury. J Neural Eng. 2016, 13(3): 031001.
effective connectivity for describing brain network [48] Kilicarslan A, Prasad S, Grossman RG, et al. High
changes induced by a memory rehabilitation treatment. accuracy decoding of user intentions using EEG to
Conf Proc IEEE Eng Med Biol Soc. 2014, 2014: control a lower-body exoskeleton. Conf Proc IEEE
6786–6789. Eng Med Biol Soc. 2013, 2013: 5606–5609.
[37] Lawrence ES, Coshall C, Dundas R, et al. Estimates [49] Capogrosso M, Milekovic T, Borton D, et al. A
of the prevalence of acute stroke impairments and brain–spine interface alleviating gait deficits after
disability in a multiethnic population. Stroke. 2001, spinal cord injury in Primates. Nature. 2016, 539(7628):
32(6): 1279–1284. 284–288.
[38] Lang CE, Beebe JA. Relating movement control at 9 [50] Aumann TD, Prut Y. Do sensorimotor β-oscillations
upper extremity segments to loss of hand function in maintain muscle synergy representations in primary
people with chronic hemiparesis. Neurorehabil Neural motor cortex? Trends Neurosci. 2015, 38(2): 77–85.
Repair. 2007, 21(3): 279–291. [51] Kim TW, Lee BH. Clinical usefulness of brain-
[39] Langhorne P, Coupar F, Pollock A. Motor recovery computer interface-controlled functional electrical
after stroke: a systematic review. Lancet Neurol. 2009, stimulation for improving brain activity in children with
8(8): 741–754. spastic cerebral palsy: a pilot randomized controlled
[40] Marino RJ, Ditunno JF Jr, Donovan WH, et al. trial. J Phys Ther Sci. 2016, 28(9): 2491–2494.
Neurologic recovery after traumatic spinal cord injury: [52] Wolpaw JR, Bedlack RS, Reda DJ, et al. Independent
data from the Model Spinal Cord Injury Systems. Arch home use of a brain-computer interface by people
Phys Med Rehabil. 1999, 80(11): 1391–1396. with amyotrophic lateral sclerosis. Neurology. 2018,
[41] Do AH, Wang PT, King CE, et al. Brain-computer 91(3): e258–e267.
interface controlled robotic gait orthosis. J Neuroeng [53] Carelli L, Solca F, Faini A, et al. Brain-computer
Rehabil. 2013, 10: 111. interface for clinical purposes: cognitive assessment
[42] Collinger JL, Wodlinger B, Downey JE, et al. High- and rehabilitation. Biomed Res Int. 2017, 2017:
performance neuroprosthetic control by an individual 1695290.
with tetraplegia. Lancet. 2013, 381(9866): 557–564. [54] Santhanam G, Ryu SI, Yu BM, et al. A high-
[43] Aflalo T, Kellis S, Klaes C, et al. Neurophysiology. performance brain-computer interface. Nature. 2006,
Decoding motor imagery from the posterior parietal 442(7099): 195–198.
cortex of a tetraplegic human. Science. 2015, [55] ADHD: clinical practice guideline for the diagnosis,
348(6237): 906–910. evaluation, and treatment of attention-deficit/
[44] Remsik AB, Dodd K, Williams L Jr, et al. Behavioral hyperactivity disorder in children and adolescents.
outcomes following brain-computer interface inter- Pediatrics. 2011, 128(5): 1007–1022.
vention for upper extremity rehabilitation in stroke: a [56] Arns M, de Ridder S, Strehl U, et al. Efficacy of
randomized controlled trial. Front Neurosci. 2018, neurofeedback treatment in ADHD: the effects on
24 Journal of Neurorestoratology

inattention, impulsivity and hyperactivity: a meta- for chronic stroke. Stroke. 2008, 39(3): 910–917.
analysis. Clin EEG Neurosci. 2009, 40(3): 180–189. [67] Soekadar SR, Birbaumer N, Slutzky MW, et al.
[57] van Doren J, Arns M, Heinrich H, et al. Sustained Brain-machine interfaces in neurorehabilitation of
effects of neurofeedback in ADHD: a systematic review stroke. Neurobiol Dis. 2015, 83: 172–179.
and meta-analysis. Eur Child Adolesc Psychiatry. [68] King CE, Wang PT, Chui LA, et al. Operation of
2019, 28(3): 293–305. a brain-computer interface walking simulator for
[58] Vollebregt MA, van Dongen-Boomsma M, Buitelaar individuals with spinal cord injury. J Neuroeng
JK, et al. Does EEG-neurofeedback improve neuro- Rehabil. 2013, 10: 77.
cognitive functioning in children with attention- [69] Broccard FD, Mullen T, Chi YM, et al. Closed-loop
deficit/hyperactivity disorder? A systematic review brain-machine-body interfaces for noninvasive
and a double-blind placebo-controlled study. J Child rehabilitation of movement disorders. Ann Biomed
Psychol Psychiatry. 2014, 55(5): 460–472. Eng. 2014, 42(8): 1573–1593.
[59] Arns M, Heinrich H, Strehl U. Evaluation of [70] Nijboer F, Furdea A, Gunst I, et al. An auditory
neurofeedback in ADHD: the long and winding road. brain-computer interface (BCI). J Neurosci Methods.
Biol Psychol. 2014, 95: 108–115. 2008, 167(1): 43–50.
[60] Snyder SM, Hall JR. A meta-analysis of quantitative [71] Cao T, Wan F, Wong CM, et al. Objective evaluation
EEG power associated with attention-deficit hyperac- of fatigue by EEG spectral analysis in steady-state
tivity disorder. J Clin Neurophysiol. 2006, 23(5): visual evoked potential-based brain-computer interfaces.
440–455. Biomed Eng Online. 2014, 13(1): 28.
[61] Coutin-Churchman P, Añez Y, Uzcátegui M, et al. [72] Chatterjee A, Aggarwal V, Ramos A, et al. A brain-
Quantitative spectral analysis of EEG in psychiatry computer interface with vibrotactile biofeedback for
revisited: drawing signs out of numbers in a clinical haptic information. J Neuroeng Rehabil. 2007, 4: 40.
setting. Clin Neurophysiol. 2003, 114(12): 2294–2306. [73] Tan DW, Schiefer MA, Keith MW, et al. A neural
[62] Daly JJ, Wolpaw JR. Brain-computer interfaces in interface provides long-term stable natural touch
neurological rehabilitation. Lancet Neurol. 2008, 7(11): perception. Sci Transl Med. 2014, 6(257): 257ra138.
1032–1043. [74] Santaniello S, Fiengo G, Glielmo L, et al. Closed-
[63] Grosse-Wentrup M, Mattia D, Oweiss K. Using loop control of deep brain stimulation: a simulation
brain-computer interfaces to induce neural plasticity study. IEEE Trans Neural Syst Rehabil Eng. 2011,
and restore function. J Neural Eng. 2011, 8(2): 025004. 19(1): 15–24.
[64] Wieloch T, Nikolich K. Mechanisms of neural plasticity [75] Fernández E, Greger B, House PA, et al. Acute
following brain injury. Curr Opin Neurobiol. 2006, human brain responses to intracortical microelectrode
16(3): 258–264. arrays: challenges and future prospects. Front Neuroeng.
[65] Ward NS. Neural plasticity and recovery of function. 2014, 7: 24.
In Progress in Brain Research. Amsterdam: Elsevier, [76] Stinear CM. Stroke rehabilitation research needs to
2005. be different to make a difference. F1000Res. 2016, 5:
[66] Buch E, Weber C, Cohen LG, et al. Think to move: a F1000 Faculty Rev–1467.
neuromagnetic brain-computer interface (BCI) system

Miaomiao Zhuang is an undergraduate in Xiangya School of Medicine, Central

South University. She was an exchange student in the University of Hong Kong
during 2019 summer. Her research interest focuses on gene mutation identification
of isolated GnRH deficiency. E-mail: [email protected]

Journal of Neurorestoratology
Journal of Neurorestoratology 25

Qingheng Wu is a sophomore in School of Stomatology, Nanjing Medical University.

Now she is an intern in School of Pharmacy, Nanjing Medical University. She was an
exchange student in the University of Hong Kong during 2019 summer. Her research
interest focuses on neural specification from human pluripotent stem cell. E-mail:
[email protected]

Feng Wan received his Ph.D. degree in electrical and electronic engineering from the
Hong Kong University of Science and Technology, Hong Kong. He is currently an
associate professor in the Department of Electrical and Computer Engineering,
Faculty of Science and Technology, and also a primary faculty in the Centre for
Cognitive and Brain Sciences, Institute of Collaborative Innovation, University of
Macau, Macau, China. His research interests include biomedical signal processing
and analysis, brain–computer interfaces, neurofeedback training, computational intelligence and
intelligent control. E-mail: [email protected]

Yong Hu received his B.Sc. and M.Sc. degrees in the biomedical engineering from
Tianjin University, Tianjin, China, in 1985 and 1988, respectively, and Ph.D. degree
from the University of Hong Kong in 1999. He is currently an associate professor and
the director of Neural Engineering and Clinical Electrophysiology Laboratory in the
Department of Orthopaedics and Traumatology, University of Hong Kong. He is also
the principle investigator of HKU-Shenzhen Hospital and HKU-Shenzhen Institute of
Research and Innovation. His research interests include neural engineering, clinical electrophysiology,
biomedical signal measurement and processing. E-mail: [email protected]