Clinical Application of Virtual Reality for Upper Limb Motor Rehabilitation in Stroke: Review of Technologies and Clinical Evidence

Neurorehabilitation for stroke is important for upper limb motor recovery. Conventional rehabilitation such as occupational therapy has been used, but novel technologies are expected to open new opportunities for better recovery. Virtual reality (VR) is a technology with a set of informatics that provides interactive environments to patients. VR can enhance neuroplasticity and recovery after a stroke by providing more intensive, repetitive, and engaging training due to several advantages, including: (1) tasks with various difficulty levels for rehabilitation, (2) augmented real-time feedback, (3) more immersive and engaging experiences, (4) more standardized rehabilitation, and (5) safe simulation of real-world activities of daily living. In this comprehensive narrative review of the application of VR in motor rehabilitation after stroke, mainly for the upper limbs, we cover: (1) the technologies used in VR rehabilitation, including sensors; (2) the clinical application of and evidence for VR in stroke rehabilitation; and (3) considerations for VR application in stroke rehabilitation. Meta-analyses for upper limb VR rehabilitation after stroke were identified by an online search of Ovid-MEDLINE, Ovid-EMBASE, the Cochrane Library, and KoreaMed. We expect that this review will provide insights into successful clinical applications or trials of VR for motor rehabilitation after stroke.


Introduction
Stroke is one of the leading causes of disability and socioeconomic burden worldwide [1]. Although the age-standardized stroke incidence has decreased in most regions, the growth of aging populations, who are at risk of stroke, may lead to an increase in the crude incidence of stroke [2]. According to a policy statement by an American Heart Association working group, approximately 4% of US adults will have a stroke by 2030 [3]. Stroke-related mortality has shown a remarkable decline due to better management in the acute phase, which means there are more people living with disabilities after stroke [1,3]. In this comprehensive narrative review of the application of VR in motor rehabilitation after stroke, we will cover (1) the technologies used in VR rehabilitation including sensors, haptic devices, and VR displays; (2) the clinical application and evidence for VR in motor rehabilitation in stroke; and (3) considerations for VR application in stroke rehabilitation. We expect that this review will provide insights into successful clinical applications or trials of VR for motor rehabilitation after stroke.

Definition of VR
VR technology can give users the experience of being surrounded by a computer-generated world. With VR, users experience inclusive and extensive surroundings, with vivid illusions of a virtual computer generated environment in which both realistic and unrealistic events can occur. So, users can interact as though they are in a real environment and may not even recognize that they are existing in a virtual environment [19]. Therefore, in VR, participants can be fully immersed in the surrounding virtual environment and interact naturally with virtual objects in the virtual world [20]. Because VR content responds to a user's movements in a natural and valid manner, such as showing the corresponding scene on the display when the user looks at it, the interaction evokes a feeling of In this comprehensive narrative review of the application of VR in motor rehabilitation after stroke, we will cover (1) the technologies used in VR rehabilitation including sensors, haptic devices, and VR displays; (2) the clinical application and evidence for VR in motor rehabilitation in stroke; and (3) considerations for VR application in stroke rehabilitation. We expect that this review will provide insights into successful clinical applications or trials of VR for motor rehabilitation after stroke.

Definition of VR
VR technology can give users the experience of being surrounded by a computer-generated world. With VR, users experience inclusive and extensive surroundings, with vivid illusions of a virtual computer generated environment in which both realistic and unrealistic events can occur. So, users can interact as though they are in a real environment and may not even recognize that they are existing in a virtual environment [19]. Therefore, in VR, participants can be fully immersed in the surrounding virtual environment and interact naturally with virtual objects in the virtual world [20]. Because VR content responds to a user's movements in a natural and valid manner, such as showing the corresponding scene on the display when the user looks at it, the interaction evokes a feeling of existing in a virtual environment, which is referred to as "presence." Moreover, control of the avatar's body movements by those of the user can even induce a feeling of ownership in which the user regards the avatar's body parts as surrogates of their own, a phenomenon called "virtual embodiment" [21]. Based on these factors, users are fully immersed, which allows them to experience where they are and what they do there in a way that is similar to real/lived experience.

Non-Immersive and Immersive VR
Non-immersive VR allows users to experience a virtual environment as observers and interact with the virtual environment by using devices that cannot fully overwhelm sensory perceptions [22], which results in a lesser feeling of immersion in the virtual world. Non-immersive VR systems are mainly characterized by users' ability to control their surroundings while perceiving stimuli around them, such as sounds, visuals, and haptics. Non-immersive VR systems are primarily based on a computer or video game console, flat screen, or monitor and input devices such as keyboards, mice, and controllers. Non-immersive VR systems can also use other physical input devices, such as racing wheels, pedals, and speed shifters, to augment users' realistic experiences. Using various input devices, users can interact with VR content on a display. To enhance the level of immersion, some non-immersive VR systems provide a first-person view for users to associate themselves with their virtual avatar. To allow users to perceive objects as being 3D, stereoscopic vision technology, with which stereo images are provided so that each eye of the user, who wears special goggles, receives the same scene but from a slightly different angle, could be used, which would allow the user to feel the third dimension from a 2D monitor or screen [23].
Immersive VR, on the other hand, improves the feeling of presence, enabling people to feel more like they are actually in the virtual environment, which means that users are more likely to interact with the stimuli given by the computer and related devices providing visual, auditory, and haptic sensations. The main goal of immersive VR is to make it possible for users to experience the illusion of being in the computer-generated environment rather than the real-world environment. By wearing a head-mounted display (HMD), tracking devices, haptic devices, and data gloves and by using wireless controllers, users can be placed in virtual environments and interact with a computer-generated world. However, the real world has a greater variety of senses including smell, taste, the feeling of warm and cold, etc., which may increase the gap between the virtual and real worlds. These could be further covered by complete immersive VR, but the need for a sophisticated artificial stimulator to provide variable sensations may require more space and have a higher cost. HMD-based immersive VR could also be enriched using physical objects or devices placed in the physical space by tracking their positions precisely in relation to where the user stands. By using this paradigm, the user could perceive the texture or temperature of objects without any awkwardness when touching it because the physical object is tracked to be placed at the same position as that in the virtual space; thus, the user touches the physical object when they touch the virtual object. Another issue to overcome is that the user must be placed in a limited space; therefore, their walking area is constrained. Using a VR treadmill allows users to physically walk or run toward any place in a virtual environment by solving two problems: realistic synchronized simulation of the user's walk and no requirement for a large space. The Cave Automatic Virtual Environment (CAVE) has been introduced as another way to provide visual information for immersive VR, instead of using an HMD [24]. CAVE uses six large walls on which scenes are displayed so that the participant can be placed in the CAVE and experience the surrounding virtual environment with a large field of view.
Immersive and non-immersive environments can be better differentiated by their level of immersion. Immersive VR strengthens the level of immersion because less mental effort is required to be immersed in the virtual environment since the hardware systems cover most sensory perceptions. In contrast, non-immersive VR requires more mental effort to be immersed in the virtual environment. Therefore, non-immersive VR may reduce the level of spatial presence, which is defined as "the sense of being in an environment" [25,26].

Technologies for Motion Tracking and Feedback for Virtual Rehabilitation
Virtual rehabilitation is a method of rehabilitation via gamification through synchronization with software content or by providing a motion guide. Various studies have been conducted to investigate the application of VR for upper limb rehabilitation (Table 1).
For motor rehabilitation of limbs, the patient's body part must be captured by motion tracking sensors and synchronously transferred to an object in VR. Sensors to track the patient's motion are mandatory for movement visualization and can be selected from a mouse and joystick, depth-sensing cameras, electromagnetic sensors, inertial sensors, bending sensors, data gloves, and so on. The sensor performance is important to precisely track the motion, but the subjective perception and preferences are also important factors to be considered, in addition to cost [27].
A sensor technology that recognizes motion is essential for virtual rehabilitation. Such technologies are divided into wearable and nonwearable devices that recognize upper limb rehabilitation motions. Nonwearable devices are further divided into those using a vision sensor and those using a robot-based controller or a controller with three degrees of freedom (DOF) either alone or in combination. Wearable devices are usually divided into those using data gloves and those using an exoskeleton. Some studies have used both types together. Sensing using cameras in nonwearable devices has recently changed from tracking markers or color patches using webcams to tracking body or hand signals through depth sensing methods. In this way, the users' movements are sensed within a limited space without obstacles. With wearable devices, the sensor is attached to collect high frequency data and force or torque can be tracked as well as position and movement.
Most studies have primarily used visual and auditory feedback through content and some studies have applied tactile and force feedback (which are haptic feedback). We divided the studies into visuomotor and visuohaptic feedback. Visuomotor feedback provides visual information by applying measured movements through sensors to content in real time. Visuohaptic feedback refers to providing haptic feedback with visual information. Haptic feedback could be divided into tactile or force feedback, depending on whether resistance is present. Tactile feedback provides feedback to users through the sense of touch using vibration, skin deformation, or small forces. Force feedback, or kinesthetic force feedback, simulates real-world physical touch using motorized motion or resistance rather than by fine touch [28]. Research on virtual rehabilitation can be categorized according to the use of fine motor tracking during upper limb rehabilitation, which is distinguished by the use of wearable or nonwearable devices. In the case of camera methods among nonwearable devices, hand tracking is possible with a HMD for VR, and this has been released as a commercial product (e.g., Oculus Quest, Facebook Technologies, LLC, Menlo Park, CA, USA). The sensor types used in previous studies are summarized in Figure 2. In studies using only visual feedback, auditory feedback could possibly be used. Abbreviations: I, immersive; NI, non-immersive; SI, semi-immersive; T, tactile; F, force; V, visual; A, auditory; EMG, electromyography; DOF, degrees of freedom; VR: virtual reality.

Examples of Commercialized VR Upper Limb Rehabilitation Systems
Commercialized virtual rehabilitation devices that can provide gamification through content or rehabilitation guides are similar to those used in VR rehabilitation research, but they are simplified and more focused on ease of use. Sensing methods are divided into wearable and nonwearable methods using cameras, joysticks, and robots ( Table 2).

Examples of Commercialized VR Upper Limb Rehabilitation Systems
Commercialized virtual rehabilitation devices that can provide gamification through content or rehabilitation guides are similar to those used in VR rehabilitation research, but they are simplified and more focused on ease of use. Sensing methods are divided into wearable and nonwearable methods using cameras, joysticks, and robots ( Table 2).

Literature Search
Studies for upper limb VR rehabilitation after stroke were identified by an online search of Ovid-MEDLINE, Ovid-EMBASE, the Cochrane Library, and KoreaMed on 18 June 2020. The search queries are presented in Supplementary Table S1. Titles and abstracts were reviewed for screening by Y.K. and non-English papers, animal studies, commentaries, case series, narratives, book chapters, editorials, nonsystematic reviews, and conference papers were excluded. Duplicated publications between databases were also excluded. A total of 339 studies were included for the full text review and Y.K. and W.S.K. selected systematic reviews and meta-analyses for review. Six meta-analyses were included for our evidence summary [63][64][65][66][67][68].

Clinical Evidence
The general characteristics of the included meta-analyses are presented in Table 3. Two studies included randomized controlled trials (RCTs) and quasi-randomized controlled trials [64,68] and three other studies only included RCTs [63,65,67]. Karamians et al. included RCTs and prospective studies [66]. The number of studies and participants included in each meta-analysis ranged from 21 to 72 and 562 to 2470, respectively.    The interventions were VR rehabilitation and the controls were either conventional therapy (dose-matched or not) or no intervention (Table 4). VR rehabilitation included rehabilitations using both custom-built virtual environments and commercial video gaming consoles (e.g., Nintendo Wii or Xbox Kinect) in the selected meta-analyses. Outcomes were usually the composite outcomes of upper limb function or activities and Karamians et al. only included studies using one of the following outcome measures: Fugl-Meyer Assessment (FMA), Wolf Motor Function Test (WMFT), and Action Research Arm Test (ARAT) [66]. Mekbib et al. only included studies using one of the three following outcomes: FMA, Box and Block Test (BBT), and Motor Activity Log (MAL) [65]. Methodological quality of the included meta-analyses was assessed using the Assessment of Multiple Systematic Reviews (AMSTAR 2) instrument by two authors (S.C. and W.S.K.) and was categorized as high, moderate, low, or critically low [69]. Any disagreements were resolved through the discussion for consensus. Most of the included meta-analyses showed moderate to high methodological quality (Table 4).
In one high-quality meta-analysis from 2014, VR rehabilitation showed better improvements in body function (standardized mean difference (SMD) = 0.43, 95% confidence interval (CI) = 0.22 to 0.64) and activities (SMD = 0.54, 95% CI = 0.28 to 0.81) when compared to conventional therapy [64]. However, the commercially available gaming failed to show a significant beneficial effect due to the small number of studies (Table 4). In a recent Cochrane systematic review with high methodological quality, VR rehabilitation for the composite outcome of upper limb function (primary outcome) was not superior to conventional therapy, but upper limb function measured by FMA was significantly improved in VR rehabilitation (SMD = 2.85, 95% CI = 1.06 to 4.65) [68]. When VR rehabilitation was applied in addition to conventional therapy, VR rehabilitation showed significant beneficial effects on the composite outcome of upper limb function (SMD = 0.49, 95 CI = 0.21 to 0.77). Two metanalyses by Aminov et al. [63] and Lee et al. [67] also showed similar moderate effect sizes for upper limb function in VR rehabilitation (Table 4). Mekbib et al. [65] only included RCTs using dose-matched conventional therapy and calculated the mean differences of FMA, BBT, and MAL, which all represent upper limb function. Although VR showed better improvements in all outcomes when compared to conventional therapy, they were less than the minimal clinically important difference [70,71].

HMDs and Motion Sickness
HMDs give users a more immersive experience in a 3D artificial world and allow interaction with virtual objects using motion tracking sensors. Considering the therapy time and active motion during the rehabilitation, the HMD must be light, comfortable to wear, positioned stably on the head, and cool enough during operation (HMD typically generate heat). The HMD may also benefit from being wireless (with enough battery life). Although VR rehabilitation can induce eye strain or physical fatigue during extended therapy, the most common issue to overcome is motion sickness. Motion sickness can be elicited when there is a lag in processing the visual response to user input interactions, resulting in conflicting signals to the brain from the eyes, vestibular systems in the inner ear, and proprioceptive sensory receptors (sensory conflict theory) [72]. Motion sickness can be affected by the system (e.g., head tracking, rendering, field of view, optics); application and user interaction (lack of controlling visual motion, visual acceleration or deceleration, longer duration of VR experience, frequent head movement during VR play); and individual perceptual factors (age, motion sickness history, lack of VR experience). The following approaches can be employed to reduce motion sickness when designing VR rehabilitation programs [73]: "(1) to make patients actively control their view points and be responsible for initiating movement, (2) to avoid or limit linear or angular accelerations or decelerations without corresponding vestibular stimulation, (3) to display visual indicators or motion trajectories, (4) to display visual cues that remain stable as the patient moves, and (5) to perform dynamic blurring of unimportant areas."

Differences in Movements in VR
The movement kinematics of the upper limb in patients with stroke differ between VR and real environments. Viau et al. reported that patients with hemiparesis used less wrist extension and more elbow extension at the end of the placing phase during reaching, grasping, and performing tasks in VR than in a real environment [74]. Similarly, several studies using reaching tasks also demonstrated that the movements in VR using HMDs were slower than those in the real environment and that spatial and temporal kinematics differ between VR and real environments [75][76][77]. Lott et al. reported that the range of the center of pressure during reaching in standing (usually used for balance training) was different between real environments, non-immersive VR with 2D flat-screen displays, and immersive VR with HMDs [78]. Considering the rehabilitation purpose of improving independence in real-world living, these different movement kinematics can affect the transfer of learning in VR to real environments and therefore must be considered when designing a VR-based rehabilitation program.

Transfer of Learning in VR to the Real World
The transfer of improved function after rehabilitation to the performance of activities of daily living is important in upper limb rehabilitation after stroke. Constraint-induced movement therapy (CIMT) comprises repetitive tasks/shaping practice with constraint of the hemiparetic upper limb, emphasizing the transfer package to foster compliance and use of the hemiparetic upper limb in the real world as a key component to improve function following CIMT [79]. Therefore, the transfer of novel rehabilitation therapeutic approaches based on repetitive movements to the real environment, such as robot-assisted arm rehabilitation [80] and VR-based rehabilitation [81], is an important issue to be discussed.
The transfer of learning effects in VR to real environments is inconclusive. Rose et al. showed that the effect of simple sensorimotor task training is comparable between VR and real environments [82]. However, several recent studies have shown that training in VR did not translate to better performance in the real environment [83][84][85][86]. In the virtual BBT simulated using a 2D flat screen and depth-sensing camera, the number of boxes moved in VR showed good correlation (a high correlation coefficient) with that in the real BBT, but the actual number of boxes moved was much less in the VR condition [33]. The weak transfer of effects from VR to real environments may be associated with different sensory-motor symptoms and spatiotemporal organization, especially the differences in depth perception in VR during upper limb rehabilitation (reach, touch, grasp, and release tasks). Although an HMD improves depth perception compared to a 2D flat screen display [87], further improvements in VR depth perception, and thereby fidelity, is needed. Possible strategies include object occlusion; effects of lighting and shadow; color shading; and relative scaling of objects by considering depth, perspective projection, and motion parallax [88]. Other methods to improve the interaction can be visual (e.g., color change) or auditory feedback when touching objects in VR. Haptic feedback can further improve the interaction and thereby the fidelity of the VR training. Ebrahimi et al. demonstrated that the errors and time to complete the task during reaching and pointing tasks using a stylus in immersive VR with a HMD were decreased with the addition of visuohaptic feedback compared to the condition without it [89]. It has also been suggested that matching the VR interaction dimensions with the control dimension of the task in the real world could improve the transfer of the VR rehabilitation effect [90].

Gamification
Gamification has been broadly and clearly defined as the "use of game design principles in non-game contexts" by Deterding et al. [91]. Gamification of VR-based rehabilitation systems can motivate patients to participate in rehabilitation actively with enjoyment, which could lead to more movements of the hemiparetic arm and better recovery [92]. The strategies to apply gamification to virtual rehabilitation design have been thoroughly reviewed by Charles et al. [88] and Mubin et al. [93].
Briefly, the VR rehabilitation system must give the patients clear feedback for meaningful play, such as the therapist's verbal and emotional encouragement, with a clear goal to be achieved during the occupational therapy. The difficulty level or challenge during the rehabilitation game should be adapted according to the patient's ability to facilitate meaningful play and handle failures [94,95] as patients with stroke may experience multiple failures and can be frustrated during upper limb rehabilitation due to motor impairments. Various types of feedback, including visual, auditory, and haptic feedback, can be applied and approaches to possibly promote motor learning should be considered (inducing variability of tasks, amplification of visual errors, and manipulating task physics for implicit behavioral guidance) [81].

Barriers
In addition to the barriers caused by patients (physical and cognitive disabilities, low adoptability, and compliance to technology), it has been suggested that there are also barriers at the therapist level, which can lead to underuse of VR rehabilitation [96]. Glegg et al. recently reviewed the barriers and facilitators influencing the adoption of VR rehabilitation, which include "the ability to grade the degree of training, transfer of training to real life, knowledge about how to operate the VR clinically, therapist self-efficacy and perceived ease of use, technical and treatment space issues, access to the technology, and time to learn practice for VR rehabilitation" [97]. They gave three recommendations to promote the use of VR rehabilitation, which were "(1) enhance collaboration, (2) ensure knowledge transfer interventions are system-and context-specific, and (3) optimize VR effectiveness through an evidence-based approach" [97].

Combinational Approaches with VR in Stroke Rehabilitation
Neuroplasticity is the ability of the human brain to adapt to certain experiences, environments, and extreme changes, including brain damage [98][99][100]. Several novel therapeutic approaches to enhance neuroplasticity can be considered as combinational approaches to VR rehabilitation. The brain-computer interface (BCI) is one method used to improve neuroplasticity after stroke; it is based on motor imagery, which is defined as the mental simulation of a kinesthetic movement. The BCI provides sensory feedback of ongoing sensorimotor brain activities, thereby enabling stroke survivors to self-modulate their sensorimotor brain activities [101]. BCI for motor rehabilitation involves the recording and decoding of brain signals generated in the sensorimotor cortex areas. The recorded brain signals can be used (1) to objectify and strengthen motor imagery-based training by providing stroke patients with real-time feedback on an imagined motor task; (2) to generate a desired motor task by producing a command to control external rehabilitative tools, such as functional electrical stimulation, robotic orthoses attached to the patient's limb, or VR; and (3) to understand cerebral reorganizations of lesioned areas by quantifying plasticity-induced changes in brain networks and power spectra in motor-related frequency bands (i.e., alpha and beta) [102]. A previous meta-analysis reported that BCI had an SMD of 0.79, which represented a medium to large effect size comparable with those of conventional rehabilitation therapy such as CIMT (SMD = 0.33), mirror therapy (SMD = 0.61), and mental practice (SMD = 0.62) [101]. Pichiorri et al. showed that BCI combined with VR may further improve upper limb rehabilitation outcomes and may be used to predict motor outcomes by analyzing brain activity in patients with stroke [103]. A more recent study also demonstrated the clinical feasibility of using a combination of BCI and VR in post-stroke motor rehabilitation and confirmed that this combinatory method may benefit patients with severe motor impairments who have little ability for volitional movement [104].
Another novel strategy to increase neuroplasticity using noninvasive brain stimulation, such as transcranial magnetic stimulation (TMS) and transcranial direct current stimulation (tDCS), has also been suggested by various researchers [105][106][107]. Noninvasive brain stimulation methods can be used to (1) enhance the ipsilesional brain activity by high-frequency rTMS [108] or anodal tDCS [109,110]; (2) inhibit contralesional brain activity by low-frequency rTMS [111,112] or cathodal tDCS [113]; (3) produce an additive effect by simultaneously applying anodal tDCS over the ipsilesional area and cathodal tDCS over the contralesional area, referred to as bihemispheric tDCS [114]; or (4) modulate the somatosensory input from nerve fibers to the brain [115,116]. These noninvasive brain stimulation methods have been reported to have acceptable tolerability and safety with no significant adverse effects in various populations, including patients with stroke [117,118]. Several studies have shown a positive effect of combinational approaches (noninvasive brain stimulation plus VR-based rehabilitation) in patients following stroke [119][120][121].
Together with BCI and noninvasive brain stimulation, a telerehabilitation approach may also be important for motor rehabilitation after stroke in terms of better accessibility and prolonged usage at home. Telerehabilitation, by definition, can provide the inputs of multidisciplinary skilled personnel for rehabilitation, including physiatrists, physiotherapists, and occupational therapists, which are often unavailable at home or challenged by transportation restrictions for disabled patients [122]. A recent Cochrane systematic review showed moderate-quality evidence that there was no difference in activities of daily living in patients with stroke between those who received telerehabilitation and those who received usual care (SMD = −0.00, 95% CI = −0.15 to 0.15) [123]. There was also low-quality evidence of no difference in upper limb functions between the use of a computer program to remotely retrain upper limb function and in-person therapy (mean difference = 1.23, 95% CI = −2.17 to 4.64) [123]. Several studies have shown that VR based telerehabilitation can be used for motor rehabilitation of upper extremity functions with improvements in FMA of the upper extremity, Brunnstrom stage, manual muscle test, and action research arm test [124,125].

Summary
VR-based rehabilitation is a promising tool to actively engage patients in the rehabilitation program and can lead to better motor recovery. Although current clinical evidence shows that VR-based rehabilitation is beneficial as an adjunct therapy to conventional rehabilitation therapy, the interventions in the studies included in the meta analyses were heterogeneous and it is unclear who benefits more from VR rehabilitation (e.g., severity of impairment, time since onset of stroke) and what type of VR (e.g., immersive vs. non-immersive) and feedback is more effective. Further research including large well-designed RCTs to find the factors influencing the effects of VR rehabilitation are required.
To improve the efficacy of VR-based rehabilitation, VR rehabilitation is designed to improve the transfer of VR training to real environments, gamification, and feedback to promote active patient participation and neuroplasticity is necessary. The user interface and user experience must be designed to be more user-friendly to patients and therapists, considering both the patient's physical and cognitive impairments and therapists' needs. VR can be integrated into novel therapeutic modalities that can enhance neuroplasticity (e.g., BCI and noninvasive brain stimulation) and is expected to induce better recovery by combinational approaches, which warrant further investigation.