Clinical Trial of HEXORR II for Robotic Hand Movement Therapy After Stroke

Impaired use of the hand in functional tasks remains dicult to overcome in many individuals after a stroke. This often leads to compensation strategies using the less-affected limb, which allows for independence in some aspects of daily activities. However, recovery of hand function remains an important therapeutic goal of many individuals, and is often resistant to conventional therapies. In prior work, we developed HEXORR I, a robotic device that allows practice of nger and thumb movements with robotic assistance. In this study, we describe modications to the device, now called HEXORR II, and a clinical trial in individuals with chronic stroke. Fifteen individuals with a diagnosis of chronic stroke were randomized to 12 or 24 sessions of robotic therapy. The sessions involved playing several video games using thumb and nger movement. The robot applied assistance to extension movement that was adapted based on task performance. Clinical and motion capture evaluations were performed before and after training and again at a 6 month followup. Fourteen individuals completed the Fugl-Meyer scores improved signicantly over the 3 time points, indicating reductions in upper extremity impairment. Flexor hypertonia (Ashworth) also decreased signicantly due to the intervention. Motion capture found increased nger range of and extension ability when the arm was supported by gravity. However, did not improve during a reach and grasp task, there was no change in a (Action Research Arm Test). At the followup, the high group had gains during a There were no signicant differences between groups. Our results are generally consistent with prior studies of tabletop robots that train the ngers in isolation from the proximal arm. We found a 2.9 point change in the Fugl-Meyer at followup, while therapy using the FINGER robot reported gains of 1.8–3.7 at followup 37 , and a study using the Amadeo robot reported a 5.1 Fugl-Meyer point change. 33 Our previous clinical study using HEXORR I also reported an increase in nger extension ability and signicant gains in the Fugl-Meyer hand section subscore after 18 hours of training. 40 However our prior study also reported grip strength increases and signicant gains in the ARAT in a subgroup of low tone subjects. Our current study did not nd any changes in the ARAT or grip strength. This might be explained by fact that the prior HEXORR I training included a squeezing task that required generation of targeted isometric matching exion forces from the ngers and thumb, followed by releasing of the grip within a certain time interval. This squeeze and release practice might have helped subjects improve in grip force and the ARAT, which involves grasping and releasing objects. We elected to drop the squeezing task from the current study to increase the number of repetitions focused on extension movement. The prior studies with the FINGER and Amadeo also reported gains in functional scales (ARAT, Box-and-Blocks, Jebsen Taylor Hand Function Test), while we did not see any improvement on functional scales in this study. One possible explanation is the low functional level of our subjects. Our mean intake Fugl-Meyer score was lower than these other two studies, and our intake ARAT scores were low (mean of 19/57 points), with 6 of our subjects having an intake ARAT of 6 points or less. In more severely impaired subjects, practice of grip or squeezing tasks might be important to include with nger extension training.


Introduction
There are 795,000 new strokes in the U.S. each year, and there are currently 7.2 million adult Americans living with stroke. 1 The associated costs are 40.1 billion annually. After the acute and subacute recovery phases, individuals with stroke move into the chronic phase (> 6 months post) where they often need continued rehabilitation, ongoing care and emotional support. 2 There's increasing evidence that rehabilitation in this chronic phase can impact quality of life. 3 In many cases, individuals regain skills and return to independent living. However, many do not receive the appropriate amount of rehabilitation therapies needed to maximize recovery due to constraints of the current health care system related to rehabilitation services. 4 In the upper extremity, reaching and grasping movements are often impaired and a focus of rehabilitation. 5,6 At 3 months post stroke, hand impairments are the most commonly reported impairment after stroke. 7 Typical impairments are hypertonia (increased passive resistance to movement), inability to activate extensors, and abnormal co-contraction of exors. 8 Interjoint coordination and modulation of activation patterns can also be impaired. 9 Hand rehabilitation remains very di cult as control of many joints and muscle groups are required to produce a coordinated grasp. Movement therapies include stretching to reduce exor hypertonia and prevent contractures, and practice of grasp and release tasks in different arm postures. However, repetitive practice of grasping tasks is di cult and frustrating for patients with moderate-severe impairments. Technologies, such as robotics, provide assistance via forces applied to the limb that may facilitate more effective practice, allowing completion of movements that would otherwise be impossible. A large body of work now exists in the area of robotic therapy. A recent meta-analysis of 45 studies including 1619 individuals with stroke, reported robotic therapy improved Activities of Daily Living (ADL) ability, function and muscle strength; however it's unclear what fraction of individuals will achieve long-term clinically meaningful gains. 10 Hand therapy robots can be divided into devices designed to be worn and used as part of ADL or devices that focus on hand movement isolated from the proximal arm. Each approach has advantages and disadvantages.
Wearable devices can be used during whole upper extremity tasks, such as reach and grasp tasks, and can take the form of active [11][12][13][14][15][16][17][18] or passive exoskeletons. [19][20][21][22] However, because of space and weight constraints in wearable devices, movement kinematics and control algorithms can often be more precise and sophisticated with desktop devices that isolate nger movements, but don't allow use of the hand with objects or in conjunction with proximal arm joints. [23][24][25][26][27][28][29][30] Many hand robots are still in the proof-of-concept prototyping phase and have not gone through clinical testing. However, the devices that have been tested clinically are showing promising results. For example, the X-Glove is a portable device with 5 linear actuators that independently extend the digits 31 . A clinical trial using the X-Glove in subacute stroke showed signi cant gains in clinical scales of impairment and function after 15 treatment sessions of 30 min of passive stretching followed by active-assisted, task-oriented training. Another clinical trial from this group in chronic stroke with the VAEDA glove using voice and EMG-control showed advantages in functional scales compared with control therapy without the glove. 32 Amadeo (Tyromotion, Austria) is a tabletop hand robot that provides independent motion of all ve ngertips along linear paths. A pilot clinical trial using Amadeo in chronic stroke showed signi cant gains in several clinical scales. 33 A more recent controlled clinical trial in chronic stroke showed greater gains after Amadeo training than dose-matched conventional therapy, along with normalizing some aspects of interhemispheric connectivity after robot training. 34 The Hand-of-Hope is an EMG-controlled exoskeleton with linkages that couple joints within each digit, decreasing the number of needed actuators. Clinical trials with this robot have shown signi cant impairment reductions and functional gains in chronic stroke subjects. 35,36 The FINGER robot provides assistance to the index and middle ngers as the subject plays a video game that simulates playing a guitar. A clinical trial reported signi cant gains in several clinical scales, with authors noting that subjects with impaired proprioception bene ted less from the training. 37 Previously, our lab developed a Hand Exoskeleton Rehabilitation Robot (HEXORR I) 38 to retrain hand control and function. HEXORR I is a tabletop device that allows practice of nger and thumb movement integrated with video games. Compared to other hand robots, HEXORR I is unique in the use of a tone-compensation algorithm that measures the resistance to passive extension movement and applies extension assistance to counter this resistance. 39 An additional novelty is the auto-adaptation algorithm that alters the shape and magnitude of the assistance pro le to achieve a desired target performance level. In theory with this approach, the patient still has control of initiation, maintenance and termination of movement, but does not have to overcome the resistance from increased exor tone during extension movement. In an initial pilot study, nine chronic stroke subjects showed signi cant improvements after 18 treatment sessions in range of motion, grip strength, and the hand component of the Fugl-Meyer score after HEXORR I use. 40 Since then, HEXORR II was developed, which includes several hardware design changes to improve the performance of the robot, reduce the setup time and make the training sessions more engaging by implementation of a larger repertoire of games. In this study, we describe HEXORR II and report the results of a clinical trial that tested for a dosage effect from the robotic therapy.

Study Design
All testing protocols were approved by the MedStar Health Research Institute human subjects institutional review board and all subjects provided written consent. The inclusion criteria were: 1) a diagnosis of stroke more than 6 months prior to randomization; 2) presence of voluntary hand activity indicated by a score of at least 1 on the nger mass extension/grasp release item of the Fugl-Meyer Test of Motor Function 41 ; 3) adequate cognitive status, as determined by Mini-Mental Status Examination 42 score > 24. Subjects were excluded if they: 1) were under the in uence of antispasticity medications during the study; 2) had MCP and IP passive extension limit > 30 degrees from full extension; 3) had pain that interfered with daily activities; 4) had excessive tone in the ngers and thumb as determined by Ashworth 43 scores > = 3; 5) had severe sensory loss or hemispatial neglect as determined by a neurological clinical exam.
Each subject was randomized to either 12 or 24 training sessions of 1.5 hours each. In each session, the subjects received robotic therapy supervised by a technician. The subjects also completed pre-training, a post-raining, and 6 month follow-up evaluation sessions involving clinical scales and biomechanical motion capture.
Hand Exoskeleton Rehabilitation Robot (HEXORR II) (2nd Generation) HEXORR II maintains the basic functionality of the rst generation device 38,40 , but the mechanism has been completely redesigned to improve usability and performance. A single motor (Maxon RE40, GP42C 26/1, 4.4 N m peak continuous torque) aligned with the MCP of the ngers, assists synchronous movement of the 4 ngers ( Fig. 1). The motor drives a chain and gear mechanism that moves 3 bars that apply forces to the palmar surfaces of the ngers in 3 locations, helping to keep the ngers in natural postures during exion and extension movements. The bars extend into a single plane for full extension, and collapse into a small space for full exion.
The thumb is controlled by a second motor (Maxon RE32, GP32C 33/1, 3.2 N m peak continuous torque) aligned with the thumb CMC. This motor drives a mechanism that rotates two pads that are strapped to the distal and proximal phalange of the thumb. Movement is about the IP and CMC joints in a single plane that can be adjusted for comfort. The forearm is strapped onto a horizontal surface that also restricts motion of the wrist. In this newer version of the device, there was a marked reduction in the time required to position and strap the hand. Also the moving inertia and friction of the robot was reduced compared to the prior version, allowing more natural movement trajectories and less resistance to free movement A Matlab Simulink program (xPCtarget, State ow) controlled the motors and provided feedback during training.

Training Protocol
For the rst session, the hand is placed in HEXORR II and 3 slow passive stretches of the ngers are performed. The movement is constant velocity and very slow (10 deg/s). We retained the motor torque applied during these stretches as the starting point for the torque vs. angle extension assistance pro les provided during training. The subjects then spent the rest of the session playing several different types of games, while assisted by the robot.
The primary therapy mode game was the Gate Game ( Fig. 2), designed to train active nger and thumb extension.
It required the subject to extend and ex the ngers and thumb to guide two balls through two openings in a gate that sweeps across the screen. If the digits were not opened in time to pass the balls through the gate, the digits were moved by the robot to full extension, before the next exion movement was prompted. An adaptation algorithm was implemented where the target performance was achievement of 2 of 3 consecutive gates. If 2 of 3 gates were successfully completed, the assistance pro le was kept unchanged. If the performance was below this level, the assistance was increased by 0.1 N m over the range from the peak extension angle achieved in the prior 3 trials to full extension. If the performance was perfect over 3 trials, the assistance pro le was scaled down by a 10%. The experimenter could change the adaptation rate by increasing or decreasing the increments in assistance via a GUI control panel. This adaptation strategy allowed for the shape of the torque vs. angle pro le to evolve as well as the overall amplitude. 40 Subjects performed 3 blocks of 30 gates in each session.
The remainder of the 90 min session was spent playing secondary video games (Fig. 3). The subject could select from 4 different PC games, which were played by moving the thumb and ngers in HEXORR II with assistance.
These games included three PC commercial games and one custom designed game. All these games had scoring systems and offered easy ways to set the game di culty and to track individual's performance. All of the games were normally controlled by mouse movement. Interface electronics (Arduino) received input of nger and thumb angles from the Matlab robot controller (RS232 communication protocol) and mouse emulator code on the Arduino controlled the PC mouse position on the computer screen through the USB port of the PC. For games that required a mouse click, a push button was controlled by the unaffected hand and provided input to one of the Arduino digital ports and integrated into the mouse control. In this way, no modi cation of the commercially available PC games was needed and an array of games could be integrated into the training. The most up to date assistance pro le was used in this mode, but was not automatically adapted as in the Gate Game. Games For the biomechanical measures, subjects were seated in front of a table at a standardized position and 4 tasks were performed. The tasks were: 1) full digit exion/extension: straightening the ngers as much as possible from a closed st position, with the hand in a pronated position at midline and the forearm supported against gravity; 2) thumb opposition: touching the thumb to the tip of the 5th digit, to test for thumb abduction range of motion; 3) grasp a water bottle placed lateral to a standard starting point at midline and bring to mouth to drink; 4) pick up a small nut placed at midline and put it on the top of a shelf. Each task was done twice, and each trial was 40 sec. Metrics from each trial were averaged across the 2 trials of each task before statistical analysis.
The kinematics were measured using an electromagnetic tracker (MiniBirds®, Ascension Technologies). Sensors were taped to nail of the thumb, index, middle and ring ngers. Sensors are also taped to the dorsum of the hand and forearm. An additional sensor is taped to the C7 vertebrae. Using commercially available biomechanics software (MotionMonitor, Innsport Inc.), anatomical landmarks were digitized and segment coordinate frames calculated for the hand, forearm and trunk. Raw data were exported into a custom Matlab program that calculated several metrics. For the nger markers, the total exion angle was calculated for the nger distal phalange relative to the hand segment. This represents the sum of exion from all three joints of each digit. For the thumb, the abduction and exion angles were both calculated. Standard Euler sequences were used for these calculations. 49 Finger extension de cit was calculated as the smallest exion angle (largest extension angle) achieved during the trial, averaged across the 4 digits measured. Finger range of motion (ROM) was calculated as the difference between the largest and smallest exion angle achieved, averaged across the 4 digits. Trunk ROM was determined by rst calculating the farthest movement of the trunk coordinate frame relative to the starting point at the beginning of the trial, in each of 3 directions (forward, lateral and vertical). A global measure of trunk ROM was calculated by combining ROM in these 3 directions using the Euclidean norm. Hand displacement (due to proximal arm movement during the reaching tasks) was calculated similarly, except trunk movement in each direction was subtracted from hand movement rst, so that the hand displacement metric was associated with arm movement only.

Data Analysis
For each outcome, a repeated measures ANOVA was used with between subjects factor of dose (12 sessions or 24 sessions) and within subjects factor of time point (pre, post, followup). Signi cant time point effects were followed with paired t-tests to detect changes from pre to post, and from pre to followup.

Results
Fifteen subjects were enrolled in this study and 1 subject withdrew due to non-compliance with the training schedule. The remaining 14 subjects were randomized to the 12 session dosage (7 subjects) or the 24 session dosage (7 subjects). The mean age was 62.3 (11.7) years, and the mean time since stroke was 28.7 (18.7) months. Seven males and 7 females completed the study, and the right limb was more affected in 8 subjects. At baseline, the mean Fugl-Meyer score was 34 ± 12, and the mean ARAT score was 19 ± 17. The two dosage groups were not signi cantly different at baseline in the FM, ARAT or Ashworth scores (p > 0.14) Table 1     and also there was a group x time interaction (p = 0.004). This was due to no increase in the low-dose subjects (p = 0.539) and a signi cant increase in the high-dose group (p = 0.005). Compared to baseline, the high-dose group increased hand displacement by 11.8 ± 0.10 cm at the followup timepoint (p = .020). There were no other signi cant group, time or interaction factors found in the RM-ANOVA of kinematic variables ( Table 2).

Discussion
HEXORR II therapy produced reductions in upper extremity impairments, as measured by signi cant gains in the Fugl-Meyer score. The largest changes were at the 6 month time point and included a signi cant reduction of exor tone, increased nger ROM, and decreased nger extension de cit with the arm supported against gravity (Task 1). The improvement in nger extension is noteworthy, as we are aware of only one prior study of hand robotics that has reported an increase is extension range that was retained 6 months after the intervention. 32 However, there were no changes on a measure of upper extremity function (ARAT). This was consistent with the kinematic analysis of a reach and grasp task (Task 3), which reported no changes in nger ROM and only a non-signi cant trend of decreased nger extension de cit. No dosage effects were found, with the exception of increased hand displacement during the task requiring forward reach (Task 4) in the high-dose group, and no change in the low-dose group.
We observed signi cant long-term reductions in hypertonia at the followup, as measured by the Modi ed Ashworth Scale. To our knowledge, this is a novel result not previously reported for hand robotic devices. While the passive stretching performed at the beginning of each session was limited to only a few repetitions, we applied a stretch and hold movement immediately after each active extension attempt during the Gate game. The possibility of co-contraction of exors during this stretch would have led to eccentric contraction of exors, which may decrease in hypertonia following neurologic injury. 50 Studies with the X-Glove have shown that a 30 min period of cyclic passive stretching can transiently improve active motor performance in stroke patients, with effects carrying over across sessions in subacute stroke. 51 Improvements in subacute stroke subjects were reported in measures of impairment and function following training that included 30 min of passive cyclic stretching followed by active-assisted, task practice. 31 Authors attributed the passive stretching to facilitating the effectiveness of the active training and preventing any increases in spasticity. There is also some evidence that orthotic-based static stretching can decrease upper extremity spasticity, although there is no evidence this alone will improve motor performance. 52,53 Thus, our results contribute to the evidence supporting further study into the use of robotics to integrate stretching protocols into active motor retraining.
One the unique aspects of this study was the detailed biomechanical analysis that reported the kinematics of nger and arm movements under several conditions. Results support the use of HEXORR II in combination with practice of functional upper extremity tasks. The HEXORR II focuses on hand movement with the forearm and wrist immobilized and the arm supported against gravity. Hand movements in conjunction with proximal arm movements were not practiced, as is required for functional use of the upper extremity. This might explain the gains in an impairment scale (Fugl-Meyer), but no gains in a test of function that tests the ability to pick up and place objects (ARAT). There is strong evidence that control of the ngers degrades when proximal muscles must support the arm against gravity. [54][55][56] These studies are consistent with our kinematic results, as nger extension did improve signi cantly when tested with the arm supported against gravity, but nger extension during reach and grasp tasks did not improve signi cantly. In the water bottle task, there were mean improvements in nger extension, but the time factor in the RM-ANOVA only approached signi cance (p = .057). Thus, it appears the training of distal hand control did produce gains in the training task, but did not generalize strongly to improved function in reach and grasp tasks. Two large multisite clinical trials of whole arm and hand robotic training also found similar results. The Armin was found to produce greater gains in the Fugl-Meyer scale than conventional therapy, but had no advantages in a motor function scale. 57 In the RATULS study, robotic therapy produced greater gains in the Fugl-Meyer compared to usual and customary care, but had no advantage on the ARAT. 58 In contrast, studies which combined robotic hand training with functional task practice have reported gains in functional scales. A recent study with Amadeo reported gains in the 9-hole Peg test, when subjects received the robotic training after a 3 hour session of physiotherapy that included 45 min of occupational therapy and 45 min of biomechanical training of upper and lower limbs. 34 Several other studies have used wearable hand robots (X-Glove 31 , VAEDA 32 , Hand-of-Hope 35,36 , HandSOME 59 ) that enabled practice of reach, grasp and release tasks with robotic assistance to hand movement. All of these studies reporting signi cant gains on a variety of functional scales. Thus, functional gains with devices similar to HEXORR II that focus on distal control only, might be achieved by integrating practice of coordinated proximal and distal limb control, as is often done during conventional therapy. Robotic and conventional therapies promote distinct patterns of motor recovery 60 , and there is evidence from clinical trials that the addition of conventional task practice to robotic therapy is superior to robotic therapy alone. [61][62][63] Our results are generally consistent with prior studies of tabletop robots that train the ngers in isolation from the proximal arm. We found a 2.9 point change in the Fugl-Meyer at followup, while therapy using the FINGER robot reported gains of 1.8-3.7 at followup 37 , and a study using the Amadeo robot reported a 5.1 Fugl-Meyer point change. 33 Our previous clinical study using HEXORR I also reported an increase in nger extension ability and signi cant gains in the Fugl-Meyer hand section subscore after 18 hours of training. 40 However our prior study also reported grip strength increases and signi cant gains in the ARAT in a subgroup of low tone subjects. Our current study did not nd any changes in the ARAT or grip strength. This might be explained by fact that the prior HEXORR I training included a squeezing task that required generation of targeted isometric matching exion forces from the ngers and thumb, followed by releasing of the grip within a certain time interval. This squeeze and release practice might have helped subjects improve in grip force and the ARAT, which involves grasping and releasing objects. We elected to drop the squeezing task from the current study to increase the number of repetitions focused on extension movement. The prior studies with the FINGER and Amadeo also reported gains in functional scales (ARAT, Box-and-Blocks, Jebsen Taylor Hand Function Test), while we did not see any improvement on functional scales in this study. One possible explanation is the low functional level of our subjects. Our mean intake Fugl-Meyer score was lower than these other two studies, and our intake ARAT scores were low (mean of 19/57 points), with 6 of our subjects having an intake ARAT of 6 points or less. In more severely impaired subjects, practice of grip or squeezing tasks might be important to include with nger extension training.
Gains were largest at the followup, with some metrics even showing no change immediately after training, but signi cant improvements at the 6 month time point (Fugl-Meyer, Ashworth, hand displacement). These improvements at the followup might have been due to more repetitions of hand and arm practice during the 6 month period between the end of training and the followup test. This practice during the followup period might have been more effective because of the improved nger control afforded by the training, or the subjects may have been encouraged to use the upper extremity more after noting the improvements in hand function during the training. The only signi cant between-group difference was as increase in hand displacement in the high dose group during a forward reaching task that appeared between post training and followup, which suggests practice of reaching tasks during this period. However, MAL scores did not indicate increased use of the more-affected arm within ADL tasks. Future studies may consider using objective methods to assess upper extremity activity as an outcome measure. 64 It is unlikely the gains during the followup period were due to encouragement or guidance from therapists, since the interaction with therapists was limited to the clinical evaluations and subjects were not given a home therapy plan during the followup period.
This study randomly assigned participants to either 12 sessions or 24 sessions. Based on RM-ANONA analysis, the group*time factors were not signi cant for our clinical outcomes measures (Fugl-Meyer and ARAT). A strong dosage effect has been di cult to show in interventions that rely on repetitive task practice. Lang et al. found no differences in functional gains in chronic stroke patients randomized to different levels of movement repetitions of task practice, even as the dosage ranged from 3200 to 10,808 repetitions. 65 The ICARE study in subacute stroke found no differences in functional gain between intensive task practice (28.3 hrs), conventional occupational therapy (26.7 hrs) and usual and customary care (11.2 hrs of therapy). 66 Robotics has been touted as a means of increasing the number of movement repetitions per treatment session, however a large study of 770 stroke subjects did not nd a signi cant difference between robotic therapy, repetitive task practice and usual care in terms of functional gains. 58 More study is needed to understand why a dosage effect is often not present in neuro-rehabilitation clinical trials of the upper extremity.

Limitations
This study has several limitations that should be noted. HEXORR II allows practice of isolated thumb movement, but the other 4 ngers are coupled together. Inability to isolate these 4 ngers may have limited the therapy's effectiveness to target weak ngers, since a weak nger can be carried along by the actions of the other 3.
Another limitation was that the automatic adaptation algorithm only operated during the Gate Game, and not the secondary games, which were commercially available PC games chosen for their professional graphics and potential to entertain the subject. However, the downside of this approach is that the robot controller has no knowledge of current performance during the game, so automatic adaptation of assistance was not possible. At times, the subject would be unable to play the video game, and the experimenter would have to try and manually adjust the assistance level via a GUI menu. This trial and error process was not always successful, and detracted from the therapy. Future efforts using this approach should incorporate feedback of performance during all of the games, so the adaptation algorithm can operate during all of the training. The device currently is not portable, but getting into the device was straightforward and the potential for a home based portable device that can be used independently by patients seems possible if the overall size and footprint of the device can be reduced.

Conclusions
Overall, HEXORR II training reduced impairment levels, increased nger extension ability and decreased exor hypertonia at the 6-month followup. The increased nger extension was achieved with the forearm supported against gravity, similar to the training task, but this did not translate to increased nger extension during a reach and grasp task. Future work with HEXORR II should focus on integrating it with functional task practice and incorporating grip and squeezing tasks. Notably, the easy setup and gaming interface make HEXORR II a potential home therapy device that could be used in conjunction with outpatient therapy, where they would receive functional upper extremity task practice.

Consent for publication
Not applicable Availability of data and material The datasets during and/or analysed during the current study available from the corresponding author on reasonable request.

Competing interests
The authors declare that they have no competing interests.

Funding
This work was supported by the National Institutes of Health under grant R15 HD075166.
Authors' contributions JC assisted with the software development, data collection, interpretation of results and was a major contributor to writing the manuscript.
IB designed the mechanical aspects of the robotic device.
DN performed subject recruitment, data collection and assisted with interpretation of results.
TC assisted with the mechanical design, assembly of the device and data collection.
MS assisted with data collection and writing of the manuscript.
RC assisted with data collection and writing of the manuscript PL conceived of the work, designed the clinical trial, assisted with the mechanical design of the robotic device, assisted with software development, performed data analysis, assisted with interpretation of results and was a major contributor to writing the manuscript.
All authors read and approved the nal manuscript. Figure 1 Pictures of the HEXORR II. The thumb exion/extension plane can be adjusted by rotating the thumb actuator about 2 independent axes through the thumb CMC and locked in place.

Figure 3
Page 19/20 The secondary video games implemented were Angry Birds, Bubble Shooter, Pong and Shopping.