|Year : 2022 | Volume
| Issue : 2 | Page : 80-87
A review on electroencephalography (EEG)-controlled upper limb exoskeletons towards stroke rehabilitation
Xin Gao, Robert Clarke, Dingguo Zhang
Centre for Autonomous Robotics (CENTAUR), University of Bath, Bath, UK
|Date of Submission||06-Jun-2022|
|Date of Decision||08-Jun-2022|
|Date of Acceptance||15-Jun-2022|
|Date of Web Publication||29-Jun-2022|
Centre for Autonomous Robotics (CENTAUR), University of Bath
Source of Support: None, Conflict of Interest: None
Stroke is a significant cause of disability in both developing and developed countries. This can cause a severe financial burden on families and society. With the development of robotics and brain-computer interfaces (BCIs), robotic exoskeletons and BCIs have received increasing clinical attention on stroke rehabilitation. Electroencephalography (EEG) is a method of recording brain signals non-invasively, which can be used as a BCI to control exoskeletons. This review focuses on rehabilitation systems of EEG-controlled upper limb exoskeletons, including the newest research progress and clinical evaluation in recent years. From the review, we find EEG-controlled exoskeletons can positively contribute to stroke rehabilitation. However, there are some issues that should be well investigated. More efforts are needed on EEG signal decoding algorithms such as deep learning methods in the clinical context. Practical applications must also bridge the gap between offline experiment and online control. In addition, this review also discusses the impact and significance of shared control, virtual reality/augmented reality, and other ways of human-computer interaction to improve EEG-controlled exoskeletons.
Keywords: brain-computer interface; electroencephalography; exoskeleton; stroke rehabilitation; upper limb
|How to cite this article:|
Gao X, Clarke R, Zhang D. A review on electroencephalography (EEG)-controlled upper limb exoskeletons towards stroke rehabilitation. Brain Netw Modulation 2022;1:80-7
|How to cite this URL:|
Gao X, Clarke R, Zhang D. A review on electroencephalography (EEG)-controlled upper limb exoskeletons towards stroke rehabilitation. Brain Netw Modulation [serial online] 2022 [cited 2022 Dec 8];1:80-7. Available from: http://www.bnmjournal.com/text.asp?2022/1/2/80/348253
| Introduction|| |
Currently, the number of people whose daily lives are affected by stroke is considerable. According to incomplete statistics from the World Stroke Organization (Lindsay et al., 2019), about 185.01 in every 100,000 people per year have a stroke for the first time globally. Among people over the age of 25 years, about a quarter will have a stroke in their lifetime. When a stroke occurs, part of the brain will lose the ability to control specific functional areas of the human body. This will affect the daily life of stroke survivors, and neurological complications such as convulsions and phantom pain will occur in those affected functional areas (American Stroke Association, 2021).
According to the studies of humans and animals on learning adaptations (Gupta et al., 1993; Mattson et al., 2018; Price and Duman, 2020), the activation of experience-based neuronal plasticity enhanced functional recovery after stroke. Neuronal plasticity is generally defined as the ability of the brain to change its neuron connections and pathways in response to a given goal for humans or animals (Zimerman and Hummel, 2014). The stroke rehabilitation of the upper limb is considered a vital direction to improve patients’ quality of daily life. Functional training, graded repetitive arm supplementary program, active and weighted exercise are the most comment upper limb stroke rehabilitation therapies (Stockley et al., 2019). Although the adjuvant therapy for stroke with human intervention has proven to be effective, it has two major shortcomings. One is that the rehabilitation therapy conducted by therapists cannot fully match the patient’s exercise intention; the other is that there is a lack of qualified therapists (Mohd Nordin et al., 2014).
There are already therapies that use robotic exoskeletons as an aid. However, rehabilitation using exoskeletons solely does not allow patients to be actively engaged. The patient’s intent is not well linked to the motion of the exoskeleton, which hinders efficient motor rehabilitation. Brain-computer interfaces (BCIs) are emerging technologies that can provide a link between brain and computer systems. BCIs can recognize human motion intention through a user’s brain signals and be used to control exoskeletons. These have been shown to solve the aforementioned limitations of rehabilitation purely based on exoskeletons (Chen et al., 2013; Iqbal and Baizid, 2015; Louie and Eng, 2016). The exoskeleton will operate according to the command signals from the BCI, helping stroke patients accomplish the training exercise that matches their own intention.
BCIs can bypass the damaged pathway between the central nervous system and peripheral nerves. Electroencephalography (EEG) measures the electrical signals of the brain via electrodes placed on a person’s scalp. EEG is widely used as a non-invasive method of brain-computer interfacing. There are many paradigms for BCIs, most notably, motor imagery (MI) and steady-state visual evoked potentials (Zhu et al., 2010; Liu et al., 2014; Hong and Qin, 2021). During a paradigm, the participant performs enough trials in the same category to facilitate the training and testing of the decoding algorithms. The EEG signals are recorded in these trials. After the EEG data is collected, the EEG signals are preprocessed, and features are extracted using various methods. Following, decoding algorithms are used to classify the EEG data and recognize human motion intention.
Independently, EEG-based BCIs can be used for stroke rehabilitation (Silvoni et al., 2011; Subramanya et al., 2012; Ang and Guan, 2017; Al-Quraishi et al., 2018). However, combining BCIs with exoskeletons can further improve rehabilitation performance. This paper will review the clinical achievements and development trends of EEG-controlled exoskeletons for upper limb rehabilitation in recent years. Lower limb rehabilitation is not in the scope of this paper. Regarding the upper limb, we consider the arm and hand separately. The strategy used for this review was to search for publications with terms related to the keywords “BCI,” “upper limb,” “EEG controlled exoskeleton,” and “stroke rehabilitation” in the past 10 years. We qualitatively analyzed the advantages and disadvantages of the research and proposed the research gap. The search engines used were Google Scholar and Web of Science. Some early studies did not set a control group. The typical studies are summarized in [Table 1].
| Electroencephalography-Controlled Arm Exoskeletons|| |
Frolov et al. (2017) invited 74 stroke survivors with paralysis of upper limbs to participate in a randomized controlled trial to determine the effectiveness of EEG-controlled exoskeleton therapy in stroke rehabilitation. The trials used four classes of MI tasks to control the exoskeleton (MI states for the left hand, right hand, and relation in each). However, the study reported a relatively low average accuracy of 51.9% in the signal decoding, and the simple rehabilitation exercise parameters in the experimental design may limit the actual effectiveness of rehabilitation. This is as the exercise was for a pinch and grip movement, controlled by the exoskeleton.
Bhagat et al. (2014) recruited 160 first-time subacute and chronic upper limb hemiplegic stroke patients. After filtering, 10 participants completed the experiment. The study used a one degree-of-freedom exoskeleton robot to assist participants with EEG-controlled movements. The EEG signal processing method uses a support vector machine based on mutual information. The study used a systematic study protocol, using the Fugl-Meyer Assessment Stroke (FMA) of motor recovery after stroke (Gladstone et al., 2002) and the Action Research Arm Test (Yozbatiran et al., 2008) metrics to compare participants before and after the rehabilitation sessions. The results showed that the FMA and Action Research Arm Test scores of the participants’ upper limbs improved by an average of 3.92 ± 3.73 and 5.35 ± 4.62 respectively after rehabilitation. According to the results of kinematic measures, it was shown that the participants’ movement speed fluency was faster and more fluent on average than before. This study invited a sufficiently convincing subject population and used quantitative indicators to compare rehabilitation effects objectively. These findings demonstrate the positive effects of EEG-controlled exoskeleton devices for stroke rehabilitation. However, the one-degree of freedom exoskeleton device used in this study may have reduced the rehabilitation effect, and there is still much room for improvement in EEG signal decoding.
Ang et al. (2015) recruited 26 post-stroke hemiplegic subjects to participate in a randomized controlled trial using BCI controlled exoskeletons. They used the upper limb FMA score as an indicator of rehabilitation efficacy. The MIT-Manus shoulder and elbow robot was used in the experiment. A band-pass filter and filter bank common spatial pattern algorithm were used as EEG signal processing methods. After 4 weeks of rehabilitative training, 71.4% of the participants had significantly improved FMA scores. The BCI group performed worse than the group only using the robotic arm assistance. One reason may be the longer task execution delay in the BCI group, which reduces the fitting level between the mental task and the actual movement. On the other hand, the participants could complete fewer rehabilitation tasks in a limited amount of time compared with those with pure exoskeleton assistance. Improving algorithm decoding accuracy and real-time performance is the key to solving this defect. Additionally, this clinical study also showed fascinating information that the robot group could bring noticeable improvement in motor function to the patients in the early stage of rehabilitation exercise. Still, as the rehabilitation progressed, the performance of the BCI group became better. In the actual rehabilitation treatment, it may be possible to consider only robotic exoskeletons in the early stage and the use of BCI-exoskeletons solution in the later periods to maximize the rehabilitation effect.
Another study by Chen et al. (2020) recruited 14 subacute stroke participants for a randomized controlled trial. These participants were divided into two groups (n = 7 per group) for the BCI and control groups and underwent 4 weeks of training. The results showed that stroke rehabilitation using BCI technology could more effectively recover the patient’s exercise ability than the exoskeleton-based rehabilitation therapy without a BCI. The study also showed that the participants who made significant improvements within their chosen rehabilitation indicators also improved event-related desynchronization strength. However, from the description of the BCI part of the study, Chen et al. (2020) only applied two exercise modes for rehabilitation training, which is minimal compared to therapist conducted rehabilitation. The study employed the power spectral density method to extract data features, and a random forest machine learning model was used for classification.
Bhagat et al. (2014) proposed an EEG controlled exoskeleton rehabilitation system, recruiting three healthy individuals and one stroke patient to evaluate the system. For stroke patients, the FMA for upper-extremities was performed to assess the degree of upper limb injury. The exoskeleton used had four active degrees of freedom. Rehabilitation training was divided into three modes: passive, triggered and active which represented the participants using an autonomous exoskeleton, the EEG-based BCI exoskeleton, and the patient themselves, respectively. In terms of EEG-based exoskeleton control, a band-pass filter and surface Laplacian were used as a preprocessing method, with EEG-based spatial averages used as features and a support vector machine used as a classifier. The trial results showed that stroke patients and healthy participants had high classification accuracy in the EEG-controlled mode. However, this study failed to mention any quantitative analysis of patients in upper limb rehabilitation using indicators such as FMA.
Nann et al. (2021) proposed a stroke rehabilitation system combining both shoulder and hand exoskeletons, which were integrated with a wheelchair to allow mobile usage. The arm exoskeleton structure included four tandem elastic drive units, which could assist the patient by providing considerable torque, assisting rehabilitation movements in an extensive range. In terms of the control system, the scheme used EEG and electrooculography, and the data was preprocessed by a band-pass filter and surface Laplacian. EEG decoding used the power method (Pfurtscheller and Lopes da Silva, 1999). A total of five post-stroke patients with severe hemiplegia participated in the experiment. Results showed that more than 75% of the subtasks were completed within three seconds. The system proposed in this study has high user-friendliness and satisfaction. However, although the experiment mentioned that stroke survivors could use the system to restore activities of daily living, the experiment did not use quantitative measures to compare the participants’ motor function changes before and after the application of the system. Alongside this, the EEG decoding algorithm used in this study was proposed in 1999, and we believe more recent algorithms with better performance may enhance the system’s overall rehabilitation efficacy.
In recent years, Frolov et al. (2018) and Wu et al. (2019) reported some evidence of the mechanism for cortical plasticity induced by intervention in functional and structural neuronal reorganization. Furthermore, a study from Ramos-Murguialday et al. (2019) also reported a significant increase in FMA score after BCI robot intervention in subjects with chronic stroke. A six-month follow-up showed that patients retained their FMA score. In addition to the above research, Barsotti et al. (2015), Bhagat et al. (2016), Nann et al. (2021) and Xiao et al. (2014) have also demonstrated statistically significant outcomes in stroke patients using BCIs.
| Electroencephalography-Controlled Hand Exoskeletons|| |
In addition to the arm exoskeletons, hand exoskeletons or robotic gloves are also considered essential for assisting upper limb rehabilitation.
Araujo et al. (2021) used an EEG-based stroke rehabilitation system fused with a flexible exoskeleton glove. Made with three-dimensional printing technology, this system is lightweight, portable, and low-cost. This study used a scheme combining the common spatial pattern feature extraction algorithm and linear discriminant analysis classifier for EEG signal decoding. The exoskeleton glove solution had a complete engineering design. However, the study showed no subject-related quantitative comparisons. Nonetheless, the advantages of this superior design and low cost are worth learning.
Another study was presented by Khan et al. (2021), which used an integrated forearm and wrist exoskeleton to assist movements. For EEG decoding, the study selected the C3, Cz and C4 channels most associated with sensorimotor rhythms. The interquartile range, median absolute deviation and energy were used as signals to classify the identified features. Finally, the feature data were classified under the linear discriminant analysis classifier. Although this study proposed a complete system, further clinical trials are needed to ensure the impact of this system on various aspects of rehabilitation.
Cheng et al. (2020) proposed a BCI-controlled exoskeleton glove to aid stroke patients’ rehabilitation. The experiment was designed with six activities of daily living, with the experimental subjects divided into a “BCI intervention group” and a “no BCI group,” with five participants in each. The non-BCI group only used exoskeleton gloves in rehabilitation. Although FMA and Action Research Arm Test did not differ significantly compared to before and after the rehabilitation exercise, subjects in the BCI group all reported a sense of movement in the upper limbs. The control group did not have this reaction, which illustrates the effectiveness of BCI on rehabilitation with possible effects in the motor region. The experiment used six types of MI tasks to assist patients in rehabilitation. Compared to having only two classification tasks, a device such as this would be more suitable to have rehabilitation activities of daily living. However, the study used the FBCSP algorithm to extract features from the data and used linear discriminant analysis as the classifier. There are stronger-performing algorithms available that may be able to achieve better results.
| Discussion|| |
So far, some EEG-controlled exoskeleton systems have been clinically tested on stroke patients to demonstrate their positive implications in practical therapies. Exoskeletons are also developing to be more lightweight, friendly, and ergonomic. It is a big topic regarding design and control of exoskeletons (Gopura and Kiguchi, 2009; Gopura et al., 2016; Gull et al., 2020; Shen et al., 2020), which is not in the scope of this review. BCI-based stroke rehabilitation therapy mainly has problems that need to be paid attention to in terms of decoding algorithms, offline algorithms versus online control, shared control, and user interactions.
In the clinical studies reviewed above, although a lot of signal processing methods are used in EEG decoding, there are almost no deep learning decoding schemes. In recent years, deep learning has been getting more attention. Some deep learning models designed for EEG decoding have been proposed. EEGNet is the most popular one that directly uses the original EEG signals as input for classification (Lawhern et al., 2018). Time-frequency features are used for an image-based deep learning model that transfers the EEG signal into the time-frequency domain and uses the feature images for classification (Xu et al., 2019; Li et al., 2020; Aslan and Akin, 2022). Long short-term memory is another popular deep learning method that is based on temporal information for decoding (Khademi et al., 2022). Similar to long short-term memory, recurrent neural networks also use temporal information for decoding, for example, these studies adopt the recurrent neural network structures to make better use of the time domain information of EEG signals (Lu et al., 2020; Tortora et al., 2020). The deep learning algorithms essentially use neuron nodes with variable weights simulated on the computer to fit the input and output data. As the number of layers of neuron nodes increases, the neural network will appear complex and have valuable characteristics for decoding. Due to the non-stationarity of the EEG signal (Pardey et al., 1996; Gonen and Tcheslavski, 2012; Ang and Guan, 2017), deep learning is considered a technique that can effectively do the entire decoding process. However, no clinical studies were found to use a deep learning decoder. The reason may be that the EEG control scenarios are very different in offline and online control; more about this topic will be discussed in the next section. Another reason may be that the EEG decoding algorithm based on deep learning has high computational time, which may delay the response (Faust et al., 2018; Rivera et al., 2022). However, decoding algorithms based on deep learning still have many advantages over traditional feature extraction schemes for clinical treatment.
Offline algorithms versus online control
Some advanced algorithms have shown high decoding accuracy for EEG signal decoding in offline data set tests recently (Hou et al., 2020; Singh Malan and Sharma, 2021; Khademi et al., 2022). However, in online scenarios, the accuracy and information transmission rate of algorithms are often lower than the results described in their paper. Part of the reason is that the algorithm may overfit the dataset. Another cause may be that the online and offline scenarios are quite different. In an online scenario, a trigger is needed to initiate decoding of the collected EEG signals with the signal decoding algorithm. Some use a continuous discriminator with a sliding window (Li et al., 2016) to achieve this effect. The advantage of this implementation is that it can distinguish between a task more often. However, the disadvantage is also apparent, which leads to a significant computational burden and misclassification of the idle state as an MI state. Another implementation method is to use electrooculography (Nakanishi et al., 2012; Jiang et al., 2014) or a physical trigger to prompt the system to enter the EEG-decoding and control state, which may present a better application effect. However, there are still a lot of engineering and theoretical problems to be faced if the BCI system is triggered to analyze the acquired signal while the ideal system would be able to automatically decode the mental state.
The most significant defect of the EEG signal is its low signal-noise ratio. As a non-stationary signal, it is difficult to extract the obvious and robust features that can be directly used for pattern recognition from EEG signals.
Additionally, EEG signals may exhibit significant changes over time in response to specific mental tasks. Therefore, EEG signal-based control efforts in the clinic may require periodic re-acquisition of the training data. Since the EEG signal is collected from the skin of the participant’s skull, the signal passes through the cerebral cortex, pia mater, dura mater, skull and scalp. While the electrodes receive the signal on the EEG cap, the obtained signal is already mixed with some noise. In addition to EEG, intracranial brain signals such as electrocorticography with less noise may also be considered for acquiring and identifying brain-related physiological signals, and electrocorticography-based BCI has just been applied to control exoskeletons (Benabid et al., 2019). This requires a trade-off between signal quality and participant acceptability.
Due to the low signal-to-noise ratio of EEG signals, the performance of EEG-based BCI is limited. Robotic control technologies may compensate for this deficiency and a shared control strategy is a feasible solution. In the BCI field, shared control generally uses BCI control as the high-level controller of intent, with computer vision or autonomous control as the detailed implementation layer. Using the shared control strategy can improve task completion efficiency with the limited human intentions provided by EEG decoding algorithms. Millan et al. (2004) used BCI system controlled small cars by applying the shared control strategy. With the assistance of shared control, the efficiency of the car to complete the task has been improved. This idea has now been extended further in rehabilitation research on robotic arms/exoskeletons controlled by BCI (Xu et al., 2019b; Quere et al., 2020; Cao et al., 2021).
In the simultaneous sharing control mode, humans and robots can complete tasks synchronously and collaboratively, which can effectively increase the success rate of tasks while ensuring the user’s sense of participation and also significantly improve patients’ confidence in rehabilitation (Philips et al., 2007; Mane et al., 2020). Sharing strategy is the core issue in shared control and the most critical case to be considered in current related research. For example, the control system needs to judge when to use the machine vision automation system, when to use the BCI control, determine the trigger signal between the BCI control and the autonomous control, etc. The optimization of these problems will bring a better rehabilitation experience and rehabilitation effects to patients.
Virtual reality/augmented reality-based user interaction
It is important to enhance the user’s participation and make EEG-controlled exoskeleton systems interesting. The use of interaction systems such as virtual reality, augmented reality and video games in stroke rehabilitation has been studied in recent years (Alimanova et al., 2017; Zheng et al., 2018; Bouteraa et al., 2019). These human-computer interactions can increase patient engagement, immersion, and motivation (Mubin et al., 2019). Gauthier et al. (2018) and Petrie (2018) investigated how to use gamification to increase repetition, engagement, and care coverage in the context of rehabilitation.
In addition to improving a patient’s focus on rehabilitation, games also positively reduce patient stress (Jannink et al., 2008). According to the study by Pittenger and Duman (2008), neuroplasticity may be disrupted in animal models of emotional disorders and stress, and games can significantly relieve patient stress. This effect is positive for recovery. For more information on the impact of virtual reality, augmented reality, and video games on stroke rehabilitation, please refer to these review papers (Saleh et al., 2017; Mubin et al., 2019; Huygelier et al., 2021). We think it is a promising direction to introduce augmented reality/virtual reality technologies into EEG-controlled exoskeleton rehabilitation training.
| Conclusions|| |
This paper reviewed the progress of EEG-controlled upper limb exoskeletons in stroke rehabilitation over recent years. The typical studies are introduced and summarized. The current challenges are analyzed, and the future directions are discussed. We believe EEG-controlled exoskeletons are a cutting-edge technology with great potential to shine in the field of stroke rehabilitation in the future. However, this requires close collaboration and continuous effort from researchers, developers, and therapists in related fields.
XG drafted the paper. RC and DZ reviewed and revised the paper. DZ conceived of this study. All authors approved the final version for publication.
Conflicts of interest
There are no conflicts of interest.
Editor note: DZ is an Editorial Board member of Brain Network and Modulation. He was blinded from reviewing or making decisions on the manuscript. The article was subject to the journal’s standard procedures, with peer review handled independently of this Editorial Board member and his research group.
Open access statement
This is an open access journal, and articles are distributed under the terms of the Creative Commons AttributionNonCommercial-ShareAlike 4.0 License, which allows others to remix, tweak, and build upon the work non-commercially, as long as appropriate credit is given and the new creations are licensed under the identical terms.
| References|| |
Al-Quraishi MS, Elamvazuthi I, Daud SA, Parasuraman S, Borboni A (2018) EEG-based control for upper and lower limb exoskeletons and prostheses: a systematic review. Sensors (Basel) 18:3342.
Alimanova M, Borambayeva S, Kozhamzharova D, Kurmangaiyeva N, Ospanova D, Tyulepberdinova G, Gaziz G, Kassenkhan A (2017) Gamification of hand rehabilitation process using virtual reality tools: using leap motion for hand rehabilitation. In: 2017 First IEEE International Conference on Robotic Computing (IRC), pp 336-339.
Ang KK, Guan C (2017) EEG-based strategies to detect motor imagery for control and rehabilitation. IEEE Trans Neural Syst Rehabil Eng 25:392-401.
Ang KK, Chua KS, Phua KS, Wang C, Chin ZY, Kuah CW, Low W, Guan C (2015) A randomized controlled trial of EEG-based motor imagery brain-computer interface robotic rehabilitation for stroke. Clin EEG Neurosci 46:310-320.
Araujo RS, Silva CR, Netto SPN, Morya E, Brasil FL (2021) Development of a low-cost EEG-controlled hand exoskeleton 3D printed on textiles. Front Neurosci 15:661569.
Aslan Z, Akin M (2022) A deep learning approach in automated detection of schizophrenia using scalogram images of EEG signals. Phys Eng Sci Med 45:83-96.
Barsotti M, Leonardis D, Loconsole C, Solazzi M, Sotgiu E, Procopio C, Chisari C, Bergamasco M, Frisoli A (2015) A full upper limb robotic exoskeleton for reaching and grasping rehabilitation triggered by MI-BCI. In: 2015 IEEE International Conference on Rehabilitation Robotics (ICORR), pp 49-54.
Benabid AL, Costecalde T, Eliseyev A, Charvet G, Verney A, Karakas S, Foerster M, Lambert A, Morinière B, Abroug N, Schaeffer MC, Moly A, Sauter-Starace F, Ratel D, Moro C, Torres-Martinez N, Langar L, Oddoux M, Polosan M, Pezzani S, et al. (2019) An exoskeleton controlled by an epidural wireless brain-machine interface in a tetraplegic patient: a proof-of-concept demonstration. Lancet Neurol 18:1112-1122.
Bhagat NA, French J, Venkatakrishnan A, Yozbatiran N, Francisco GE, O’Malley MK, Contreras-Vidal JL (2014) Detecting movement intent from scalp EEG in a novel upper limb robotic rehabilitation system for stroke. Annu Int Conf IEEE Eng Med Biol Soc 2014:4127-4130.
Bhagat NA, Yozbatiran N, Sullivan JL, Paranjape R, Losey C, Hernandez Z, Keser Z, Grossman R, Francisco GE, O’Malley MK, Contreras-Vidal JL (2020) Neural activity modulations and motor recovery following brain-exoskeleton interface mediated stroke rehabilitation. Neuroimage Clin 28:102502.
Bhagat NA, Venkatakrishnan A, Abibullaev B, Artz EJ, Yozbatiran N, Blank AA, French J, Karmonik C, Grossman RG, O’Malley MK, Francisco GE, Contreras-Vidal JL (2016) Design and optimization of an EEG-based brain machine interface (BMI) to an upper-limb exoskeleton for stroke survivors. Front Neurosci 10:122.
Bouteraa Y, Abdallah IB, Elmogy AM (2019) Training of hand rehabilitation using low cost exoskeleton and vision-based game interface. J Intell Robot Syst 96:31-47.
Cao L, Li G, Xu Y, Zhang H, Shu X, Zhang D (2021) A brain-actuated robotic arm system using non-invasive hybrid brain-computer interface and shared control strategy. J Neural Eng 18:046045.
Chen G, Chan CK, Guo Z, Yu H (2013) A review of lower extremity assistive robotic exoskeletons in rehabilitation therapy. Crit Rev Biomed Eng 41:343-363.
Chen S, Cao L, Shu X, Wang H, Ding L, Wang SH, Jia J (2020) Longitudinal electroencephalography analysis in subacute stroke patients during intervention of brain-computer interface with exoskeleton feedback. Front Neurosci 14:809.
Cheng N, Phua KS, Lai HS, Tam PK, Tang KY, Cheng KK, Yeow RC, Ang KK, Guan C, Lim JH (2020) Brain-computer interface-based soft robotic glove rehabilitation for stroke. IEEE Trans Biomed Eng 67:3339-3351.
Faust O, Hagiwara Y, Hong TJ, Lih OS, Acharya UR (2018) Deep learning for healthcare applications based on physiological signals: A review. Comput Methods Programs Biomed 161:1-13.
Frolov AA, Bobrov PD, Biryukova EV, Silchenko AV, Kondur AA, Dzhalagoniya IZ, Massion J (2018) Electrical, hemodynamic, and motor activity in BCI post-stroke rehabilitation: clinical case study. Front Neurol 9:1135.
Frolov AA, Mokienko O, Lyukmanov R, Biryukova E, Kotov S, Turbina L, Nadareyshvily G, Bushkova Y (2017) Post-stroke rehabilitation training with a motor-imagery-based brain-computer interface (BCI)-controlled hand exoskeleton: a randomized controlled multicenter trial. Front Neurosci 11:400.
Gauthier LV, Richter TA, George LC, Schubauer KM (2018) Chapter 33 - Gaming for the brain: video gaming to rehabilitate the upper extremity after stroke. In: Neuromodulation (Second Edition) (Krames ES, Peckham PH, Rezai AR, eds), pp 465-476. Academic Press.
Gladstone DJ, Danells CJ, Black SE (2002) The Fugl-Meyer assessment of motor recovery after stroke: a critical review of its measurement properties. Neurorehabil Neural Repair 16:232-240.
Gonen FF, Tcheslavski GV (2012) Techniques to assess stationarity and gaussianity of EEG: an overview. Int J Bioautomation 16:135-142.
Gopura RARC, Kiguchi K (2009) Mechanical designs of active upper-limb exoskeleton robots: State-of-the-art and design difficulties. In: 2009 IEEE International Conference on Rehabilitation Robotics, pp 178-187.
Gopura RARC, Bandara DSV, Kiguchi K, Mann GKI (2016) Developments in hardware systems of active upper-limb exoskeleton robots: A review. Robot Autonomous Syst 75:203-220.
Gull MA, Bai S, Bak T (2020) A review on design of upper limb exoskeletons. Robotics 9:16.
Gupta MM, Rao DH, Nikiforuk PN (1993) Neuro-controller with dynamic learning and adaptation. J Intell Robot Syst 7:151-173.
Hong J, Qin X (2021) Signal processing algorithms for SSVEP-based brain computer interface: State-of-the-art and recent developments. J Intell Fuzzy Syst 40:10559–10573.
Hou Y, Jia S, Lun X, Zhang S, Chen T, Wang F, Lv J (2020) GCNs-Net: a graph convolutional neural network approach for decoding time-resolved EEG motor imagery signals. arXiv:2006.08924.
Huygelier H, Mattheus E, Abeele VV, van Ee R, Gillebert CR (2021) The use of the term virtual reality in post-stroke rehabilitation: a scoping review and commentary. Psychol Belg 61:145-162.
Iqbal J, Baizid K (2015) Stroke rehabilitation using exoskeleton-based robotic exercisers: mini review. Biomed Res India 26:197-201.
Jannink MJ, van der Wilden GJ, Navis DW, Visser G, Gussinklo J, Ijzerman M (2008) A low-cost video game applied for training of upper extremity function in children with cerebral palsy: a pilot study. CyberPsychol Behav 11:27-32.
Jiang J, Zhou Z, Yin E, Yu Y, Hu D (2014) Hybrid Brain-Computer Interface (BCI) based on the EEG and EOG signals. Biomed Mater Eng 24:2919-2925.
Khademi Z, Ebrahimi F, Kordy HM (2022) A transfer learning-based CNN and LSTM hybrid deep learning model to classify motor imagery EEG signals. Comput Biol Med 143:105288.
Khan BA, Usmani AR, Athar S, Hashmi A, Farooq O, Muzammil M (2021) EEG-based exoskeleton for rehabilitation therapy. In: Ergonomics for improved productivity (Muzammil M, Khan AA, Hasan F, eds), pp 645-653. Singapore: Springer Singapore.
Lawhern VJ, Solon AJ, Waytowich NR, Gordon SM, Hung CP, Lance BJ (2018) EEGNet: a compact convolutional neural network for EEG-based brain-computer interfaces. J Neural Eng 15:056013.
Li Y, Wei H-L, Billings SA, Sarrigiannis PG (2016) Identification of nonlinear time-varying systems using an online sliding-window and common model structure selection (CMSS) approach with applications to EEG. Int J Syst Sci 47:2671-2681.
Lindsay MP, Norrving B, Sacco RL, Brainin M, Hacke W, Martins S, Pandian J, Feigin V (2019) World Stroke Organization (WSO): Global Stroke Fact Sheet 2019. Int J Stroke 14:806-817.
Liu Q, Chen K, Ai Q, Xie S (2014) Review: recent development of signal processing algorithms for SSVEP-based brain computer interfaces. J Med Biol Eng 34:299-309.
Louie DR, Eng JJ (2016) Powered robotic exoskeletons in post-stroke rehabilitation of gait: a scoping review. J Neuroeng Rehabil 13:53.
Lu W, Wei Y, Yuan J, Deng Y, Song A (2020) Tractor assistant driving control method based on EEG combined with RNN-TL deep learning algorithm. IEEE Access 8:163269-163279.
Mane R, Chouhan T, Guan C (2020) BCI for stroke rehabilitation: motor and beyond. J Neural Eng 17:041001.
Mattson MP, Moehl K, Ghena N, Schmaedick M, Cheng A (2018) Intermittent metabolic switching, neuroplasticity and brain health. Nat Rev Neurosci 19:63-80.
Millán Jdel R, Renkens F, Mouriño J, Gerstner W (2004) Noninvasive brain-actuated control of a mobile robot by human EEG. IEEE Trans Biomed Eng 51:1026-1033.
Mohd Nordin NA, Aziz NA, Abdul Aziz AF, Ajit Singh DK, Omar Othman NA, Sulong S, Aljunid SM (2014) Exploring views on long term rehabilitation for people with stroke in a developing country: findings from focus group discussions. BMC Health Serv Res 14:118.
Mubin O, Alnajjar F, Jishtu N, Alsinglawi B, Al Mahmud A (2019) Exoskeletons with virtual reality, augmented reality, and gamification for stroke patients’ rehabilitation: systematic review. JMIR Rehabil Assist Technol 6:e12010.
Nakanishi M, Mitsukura Y, Wang Y, Wang YT, Jung TP (2012) Online voluntary eye blink detection using electrooculogram. In: Proceedings of the 2012 International Symposium on Nonlinear Theory and its Applications.
Nann M, Cordella F, Trigili E, Lauretti C, Bravi M, Miccinilli S, Catalan JM, Badesa FJ, Crea S, Bressi F, Garcia-Aracil N, Vitiello N, Zollo L, Soekadar SR (2021) Restoring activities of daily living using an EEG/EOG-controlled semiautonomous and mobile whole-arm exoskeleton in chronic stroke. IEEE Syst J 15:2314-2321.
Pardey J, Roberts S, Tarassenko L (1996) A review of parametric modelling techniques for EEG analysis. Med Eng Phys 18:2-11.
Petrie R (2018) Designing an augmented reality video game to assist stroke patients with independent rehabilitation. Wellington: Victoria University of Wellington.
Pfurtscheller G, Lopes da Silva FH (1999) Event-related EEG/MEG synchronization and desynchronization: basic principles. Clin Neurophysiol 110:1842-1857.
Philips J, Millan JdR, Vanacker G, Lew E, Galan F, Ferrez PW, Brussel HV, Nuttin M (2007) Adaptive shared control of a brain-actuated simulated wheelchair. In: 2007 IEEE 10th International Conference on Rehabilitation Robotics, pp 408-414.
Pittenger C, Duman RS (2008) Stress, depression, and neuroplasticity: a convergence of mechanisms. Neuropsychopharmacology 33:88-109.
Price RB, Duman R (2020) Neuroplasticity in cognitive and psychological mechanisms of depression: an integrative model. Mol Psychiatry 25:530-543.
Quere G, Hagengruber A, Iskandar M, Bustamante S, Leidner D, Stulp F, Vogel J (2020) Shared control templates for assistive robotics. In: 2020 IEEE International Conference on Robotics and Automation (ICRA), pp 1956-1962.
Ramos-Murguialday A, Curado MR, Broetz D, Yilmaz Ö, Brasil FL, Liberati G, Garcia-Cossio E, Cho W, Caria A, Cohen LG, Birbaumer N (2019) Brain-machine interface in chronic stroke: randomized trial long-term follow-up. Neurorehabil Neural Repair 33:188-198.
Rivera MJ, Teruel MA, Maté A, Trujillo J (2022) Diagnosis and prognosis of mental disorders by means of EEG and deep learning: a systematic mapping study. Artif Intell Rev 55:1209-1251.
Saleh S, Fluet G, Qiu Q, Merians A, Adamovich SV, Tunik E (2017) Neural patterns of reorganization after intensive robot-assisted virtual reality therapy and repetitive task practice in patients with chronic stroke. Front Neurol 8:452.
Shen Y, Ferguson PW, Rosen J (2020) Chapter 1 - Upper limb exoskeleton systems—overview. In: Wearable robotics (Rosen J, Ferguson PW, eds), pp 1-22. Academic Press.
Silvoni S, Ramos-Murguialday A, Cavinato M, Volpato C, Cisotto G, Turolla A, Piccione F, Birbaumer N (2011) Brain-computer interface in stroke: a review of progress. Clin EEG Neurosci 42:245-252.
Singh Malan N, Sharma S (2021) Time window and frequency band optimization using regularized neighbourhood component analysis for Multi-View Motor Imagery EEG classification. Biomed Signal Process Control 67:102550.
Stockley R, Peel R, Jarvis K, Connell L (2019) Current therapy for the upper limb after stroke: a cross-sectional survey of UK therapists. BMJ Open 9:e030262.
Subramanya K, Pinto AJP, Kumar MKA, Arya BK, Mahadevappa M (2012) Surface electrical stimulation technology for stroke rehabilitation: a review of 50 years of research. J Med Imaging Health Inform 2:1-14.
Tortora S, Ghidoni S, Chisari C, Micera S, Artoni F (2020) Deep learning-based BCI for gait decoding from EEG with LSTM recurrent neural network. J Neural Eng 17:046011.
Wu Q, Yue Z, Ge Y, Ma D, Yin H, Zhao H, Liu G, Wang J, Dou W, Pan Y (2019) Brain functional networks study of subacute stroke patients with upper limb dysfunction after comprehensive rehabilitation including BCI training. Front Neurol 10:1419.
Xiao ZG, Elnady AM, Webb J, Menon C (2014) Towards a brain computer interface driven exoskeleton for upper extremity rehabilitation. In: 5th IEEE RAS/EMBS International Conference on Biomedical Robotics and Biomechatronics, pp 432-437.
Xu G, Shen X, Chen S, Zong Y, Zhang C, Yue H, Liu M, Chen F, Che W (2019a) A deep transfer convolutional neural network framework for EEG signal classification. IEEE Access 7:112767-112776.
Xu Y, Ding C, Shu X, Gui K, Bezsudnova Y, Sheng X, Zhang D (2019b) Shared control of a robotic arm using non-invasive brain–computer interface and computer vision guidance. Robot Autonomous Syst 115:121-129.
Yozbatiran N, Der-Yeghiaian L, Cramer SC (2008) A standardized approach to performing the action research arm test. Neurorehabil Neural Repair 22:78-90.
Zheng J, Shi P, Yu H (2018) A virtual reality rehabilitation training system based on upper limb exoskeleton robot. In: 2018 10th International Conference on Intelligent Human-Machine Systems and Cybernetics (IHMSC), pp 220-223.
Zhu D, Bieger J, Garcia Molina G, Aarts RM (2010) A survey of stimulation methods used in SSVEP-based BCIs. Comput Intell Neurosci 2010:702357.
Zimerman M, Hummel FC (2014) Chapter 12 - Brain stimulation and its role in neurological diseases. In: The stimulated brain (Cohen Kadosh R, ed), pp 333-369. San Diego: Academic Press.