User-Driven Functional Movement Training With a Wearable ... › rehab › sites › default ›...

IEEE TRANSACTIONS ON NEURAL SYSTEMS AND REHABILITATION ENGINEERING, VOL. 28, NO. 10, OCTOBER 2020 2265

User-Driven Functional Movement Training Witha Wearable Hand Robot After Stroke

Sangwoo Park , Member, IEEE, Michaela Fraser, Lynne M. Weber, Cassie Meeker , Member, IEEE,Lauri Bishop, Daniel Geller, Joel Stein , and Matei Ciocarlie , Member, IEEE

Abstract— We studied the performance of a robotic ortho-sis designed to assist the paretic hand after stroke. It iswearable and fully user-controlled, serving two possibleroles: as a therapeutic tool that facilitates device-mediatedhand exercises to recover neuromuscular function or asan assistive device for use in everyday activities to aidfunctional use of the hand. We present the clinical out-comes of a pilot study designed as a feasibility test forthese hypotheses. 11 chronic stroke (>2 years) patientswith moderate muscle tone (Modified Ashworth Scale ≤2 inupper extremity) engaged in a month-long training proto-col using the orthosis. Individuals were evaluated usingstandardized outcome measures, both with and withoutorthosis assistance. Fugl-Meyer post intervention scoreswithout robotic assistance showed improvement focusedspecifically at the distal joints of the upper limb, suggestingthe use of the orthosis as a rehabilitative device for thehand. Action Research Arm Test scores post interventionwith robotic assistance showed that the device may servean assistive role in grasping tasks. These results highlightthe potential for wearable and user-driven robotic handorthoses to extend the use and training of the affected upperlimb after stroke.

Index Terms— Wearable robotics, rehabilitation robotics,hand orthosis, stroke rehabilitation, intent detection.

I. INTRODUCTION

HEMIPARESIS of the upper limb (UL) is a commonand debilitating complication after stroke [1]. Approx-imately 50% of survivors with UL paralysis continue topresent with functional deficits four years after stroke [2].Growing evidence demonstrates high quality, highly repetitive,

Manuscript received November 18, 2019; revised April 3, 2020,July 3, 2020, and August 27, 2020; accepted August 29, 2020. Dateof publication September 4, 2020; date of current version October 8,2020. This work was supported in part by the National Science Foun-dation through the National Robotics Initiative program under GrantIIS-1526960. (Corresponding author: Sangwoo Park.)

Sangwoo Park, Cassie Meeker, and Matei Ciocarlie are withthe Department of Mechanical Engineering, Columbia University,New York, NY 10027 USA (e-mail: [email protected]; [email protected]; [email protected]).

Michaela Fraser, Lynne M. Weber, Daniel Geller, and Joel Steinare with the Department of Rehabilitation and Regenerative Medicine,Columbia University, New York, NY 10032 USA (e-mail: [email protected]; [email protected]; [email protected];[email protected]).

Lauri Bishop is with the Division of Biokinesiology and Physical Ther-apy, University of Southern California, Los Angeles, CA 90089 USA(e-mail: [email protected]).

This article has supplementary downloadable material available athttp://ieeexplore.ieee.org, provided by the authors.

Digital Object Identifier 10.1109/TNSRE.2020.3021691

Fig. 1. Top: Exotendon device and EMG armband. Bottom: Functionalmovement training with a wearable hand orthosis.

and task-specific training is beneficial in UL recovery afterstroke [3], [4]. However, there are challenges which impedemany chronic stroke patients from receiving this type ofrehabilitation program, for reasons that include logistical andgeographical barriers of visiting therapy clinics, insuranceand reimbursement limitations, and insufficient availability oftherapists with specialized training [5].

Robotic devices have been designed to address the needfor increased UL repetitions for post-stroke rehabilitation [6].However, these largely target proximal segments of the UL(i.e. shoulder and elbow); in studies with these devices,Fugl-Meyer (FM) - UL and kinematic parameters show thatrobotic training produces improved functional gains mostly inthe proximal joints [7], [8]. This disparity of motor recoverybetween proximal and distal joints (i.e. wrist and fingers)can lead to undesirable compensatory grasp patterns, whichmake long-term rehabilitation more complicated and oftenineffective [9]. In contrast, recent work found clinical effec-tiveness when subjects receive robotic support directly to

1534-4320 © 2020 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.See https://www.ieee.org/publications/rights/index.html for more information.

Authorized licensed use limited to: Columbia University Libraries. Downloaded on October 09,2020 at 00:13:14 UTC from IEEE Xplore. Restrictions apply.

https://orcid.org/0000-0002-9838-2814https://orcid.org/0000-0002-3065-3817https://orcid.org/0000-0001-5527-025Xhttps://orcid.org/0000-0002-8317-4465

2266 IEEE TRANSACTIONS ON NEURAL SYSTEMS AND REHABILITATION ENGINEERING, VOL. 28, NO. 10, OCTOBER 2020

the distal joints compared to support to the proximal jointsalone [10]. However, wearable active assistance for the handis complicated by the limited space available for the neces-sary motor, controller, transmission components, and powerstorage, as well as by the complex anatomy and kinesiologyof the stroke-affected hand. These difficulties have led to thecurrent lack of functional prototypes for wearable hand robotsused and assessed with clinical efficacy [11].

To help address this need, we have developed a wear-able hand orthosis, which provides mechanical assistancefor finger extension. Our previous studies have establishedthe basic operation principles of the device: in limited caseseries with stroke survivors, we have shown that our exoten-don network can facilitate finger extension to enable grossgrasp/release [12] via a small motor relying on effectiveforce transmission mechanisms [13], and that, for a subsetof patients, we can infer the intent to open and close the handwhen the orthosis is used in conjunction with a commodityelectromyography (EMG) armband [14]. However, the clinicalperformance of this device and the importance of trainingeffects over longer-term use have not been investigated to date.

In this pilot study, we focus on the performance of ourhand orthosis as a rehabilitation device and as an assistivedevice. We present clinical outcomes from a 12 session train-ing program, comprising 3 sessions per week for 4 weeks.Each session involved 30 minutes of active training time inwhich participants practiced a variety of grasp and releasetasks with everyday objects to simulate Activities of DailyLiving (ADLs). 11 subjects with chronic stroke completedthe protocol and were evaluated with a battery of clinicalassessments pre- and post-intervention: FM - UL, ActionResearch Arm Test (ARAT), and Box and Block Test (BBT).

In order to determine the efficacy of the orthosis as arehabilitative device, we compare clinical outcomes both pre-and post-intervention without device assistance. To distinguishbetween recovery throughout the entire UL and more localizedimprovement of the distal segments, we subdivide FM intoshoulder/elbow (FM-proximal) and wrist/hand (FM-distal).To determine efficacy when used as an assistive device,we compare baseline performance and post-intervention clin-ical outcomes without assistance to post-intervention perfor-mance while wearing the device. Our goal is to determineif increased competence using the device from a month-longtraining program leads to increased performance in tasksrequiring grasp, transport, and release of objects while wearingthe orthosis. We examine the assistive capability throughARAT scores measured while the users are wearing therobot. Finally, we provide secondary analyses to compareperformance for our two intent inferral methods, namely EMGand shoulder harness controls, used to drive the hand deviceduring both training and post-testing.

Overall, the main contributions of this paper are as follows:• To the best of our knowledge, it is the first time an active

wearable hand robot was evaluated in clinical assess-ments both with and without robotic assistance, followinguser-driven functional hand exercises over multiple train-ing sessions for chronic stroke patients interacting withvarious real-world objects.

• FM subsection scores suggest that intensive hand func-tional exercises using our robot may improve motorfunction in distal segments on the UL.

• Subsection of ARAT highlight the potential for the assis-tive capability of the device for grasping components ofADLs.

Our aim is to provide wearable, user-driven assistance for thestroke-affected hand, and to investigate its effects on both reha-bilitation and functional performance. These are steps towardsour long-term goal to test user-operated active orthoses outsideof a clinical setting, for rehabilitation or home-assistance, withthe potential to significantly increase the number of motorrepetitions and the intensity of training.

II. RELATED WORK

Due to the complexity of hand movements and vari-able impairment patterns seen in stroke patients, it is onlyrecently that robotic devices for hand rehabilitation havebeen proposed [11]. Robotic workstations for hand rehabil-itation on the market have demonstrated their feasibility andefficacy [15], [16]. But, workstation devices are often bulky,costly, and tethered to a clinical setup requiring extensivesupervision by health care providers. Wearable robots, by con-trast, have potential to enable use outside the clinic allowinga greater number of repetitions of functional tasks. Therefore,this type of robot has become a focus of recent research inrobot-assisted rehabilitation [17]. To facilitate these benefits,we focus on wearable hand devices in this work, though in thisearly pilot stage, the device is utilized in a clinical environmentwith staff supervision.

Traditional wearable hand robots in the early stage aremostly exoskeleton comprised of rigid links [18]. However,such devices are difficult to align axes of biological and roboticjoints. A soft pneumatic actuator based orthosis can be analternative as it provides safe human-robotic interaction dueto natural compliance and flexibility. Polygerinos et al. havedeveloped a pneumatic powered glove that is inexpensive, low-profile, and well adapted to complex finger movement [19],although untethering from air pressure sources can be chal-lenging for complete portability. A wearable hand device canalso take the form of supernumerary robotic finger to provideassistive benefits for practical use of the affected UL [20].

For intuitive and user-driven control of a wearable handdevice, there has been research to develop wearable sen-sors. Multichannel EMG with wireless communication canbe built within a wearable package [21], and this type ofsensor allows intuitive control using a pattern recognitionalgorithm [14], [22]. A low-profile and portable glove withvarious sensors, such as DAGLOVE [23], can also allowuse of pattern recognition utilizing multimodal sensor data.Multimodal intent detection can provide better accuracy forstroke patients compared to an algorithm that uses a singlemodality, such as EMG [24].

Among many hand rehabilitation tools [11], [25], a subsethave been supported by clinical evidence for chronic strokepatients. The PneuGlove, which facilitates manipulation activ-ities in virtual reality and with real-world objects, demon-strated improvements in a pilot study of 14 subjects [26].


PARK et al.: USER-DRIVEN FUNCTIONAL MOVEMENT TRAINING WITH A WEARABLE HAND ROBOT AFTER STROKE 2267

Hand of Hope has also proven the efficacy of robot-assistedtraining through a pilot study with 19 stroke patients [16].A number of passive devices have also been proposed andvalidated [27], [28], with the advantage of being simpler tooperate for unsupervised therapy. However, given that patientswith hemiparesis may exhibit severe muscle weakness in gripstrength [29], even lower level of mechanical interference withfinger flexion via passive mechanisms can adversely affecthand functionality.

While the majority of devices aim to demonstrate clinicaleffectiveness as a rehabilitative training tool, some others aimto develop an assistive device for immediate assisted functionalgains. Yurkewich et al. have shown that three chronic andtwo acute stroke subjects improved in modified BBT, rangeof motion of the fingers, and a subset of CAHAI [30].The other device is the Myomo which is a portable elbow-wrist-hand orthosis controlled by EMG. With the Myomo,18 chronic stroke survivors achieved immediate and significantUL improvements as confirmed by FM, BBT, and a batteryof functional tasks [31]. In terms of intent detection forassistive robots, a mathematical formulation and algorithmcan potentially help improve functionality of users with motorimpairment [32]. However, these devices have not been testedfor long-term use for ADLs. We believe it is important forwearable devices to demonstrate their robustness such thatusers can complete clinical sessions, and eventually use thedevice unsupervised at home. In this work, we investigate thefeasibility of our hand robot through a clinical protocol toevaluate its strengths and limitations, both as a rehabilitativeintervention and as an assistive device.

III. AN ACTIVE HAND ORTHOSIS FOR STROKE PATIENTS

Chronic stroke patients with hemiparesis often experiencefunctional disuse of their hand. Our device aims to addressone of the most common impairment patterns seen in thispopulation: a combination of weakness, spasticity, poor coor-dination, and a flexor synergy pattern where individuals maybe able to actively flex their fingers to form a gross grasp,but are unable to volitionally extend their fingers to releasethe grasp. By assisting finger extension, our device enablesusers to harness their residual function and incorporate theirimpaired hand into functional grasp and release tasks.

A. Exotendon Device

To address many practical and anatomical challenges ofhand robot designs, our device utilizes underactuation mecha-nisms through tendon networks [12]. The anchoring structuresare designed to provide effective force transmission, in orderto overcome hand spasticity while using low motor forces [13].This design allows functional hand movement using a smallactuator while also reducing distal migration, an undesirablephenomenon where the motor components of the device slidedown the forearm towards the hand due to the applied forces.

Our exotendon device (Fig. 1 - top) consists of a rigidforearm splint on which an actuator (Pololu-25D-MP-12V)is mounted, along with 3D printed distal components for thefingers. The splint constrains wrist movement at neutral angle,

Fig. 2. Tendon routes for the thumb (left) and fingertip components(right).

thus allowing motor force to be transmitted through cableroutes to the fingers. The 3D printed fingertip components(Fig. 2 - right) are secured on the dorsal side with Velcrostrapped around each finger. The main role of these com-ponents is to increase the tendon moment arms around theproximal interphalangeal (PIP) joints throughout the entirerange of motion, so that the motor force is effectively appliedto the digits. Also, this component mechanically preventshyper-extension of the distal interphalangeal (DIP) joint toavoid any injury. Similarly, the distal portion of the fin-gertip component creates a mechanical block that preventshyper-extension at the PIP and metacarpophalangeal (MCP)joints. The device does not prevent hyper-extension of thePIP joints for stroke patients with significantly higher stiff-ness on the MCP joints than on the PIP joints; however,in our study we have not encountered any patients exhibitingthis pattern.

The main role of the orthosis is to provide assistance for fin-ger extension. As size and weight are of critical importance forbuilding a wearable device, we rely on a heavily underactuateddesign: a single motor provides assistance for extending all thedigits except the thumb. When we detect the wearer’s intent toopen the hand (as detailed in the next subsection), the actuatorretracts, applying extension torques to the digit joints via theexotendon network. Conversely, when we detect the intentto close the hand, the motor extends and relaxes the forcesin the tendon network, allowing the digits to flex. We relyon the patient being able to generate sufficient finger flexiontorques, which is common for the impairment pattern we focuson. Throughout this study, we use a PID position control todrive the motor, and the range of motion is determined atthe beginning of protocol depending on the size of the user’shand. Additional specifications include the following - weight:365 g; motor gear ratio: 47:1; extension/retraction time: 1.8 s;max extension force: 100 N; donning and doffing time: approx.15 mins and 1 min respectively.

The natural movements of the thumb are complex and dis-tinct from the other digits [33]. In addition to flexion/extensionand abduction/adduction, the thumb moves in opposition tothe other digits to grasp and pinch. To ensure that the thumbapproximately opposes the other fingers, we use two passivecable routes, one for extension and another for abduction(Fig. 2 - left). The tensions of these two cables are calibratedat the beginning of each session, and are fixed for the durationof the session. The thumb is thus not actively assisted by



Fig. 3. EMG armband or SH, depending on the assigned group, sendsbiophysical data to a computer through bluetooth or microcontroller unit(MCU). In the computer, intent detection algorithm classifies the intentionand generate a motor command. Then, the command is transmitted tothe motor through the MCU.

the orthosis, but statically splinted into opposition for theduration of the protocol, as seen previously in the litera-ture [34]. We have found this procedure to enable graspingwhile reducing the load on the actuator, which must only assistthe remaining digits.

B. Intent Detection

Our goal for this robotic device is for the patient toinitiate robotic assistance by signaling the intent to open thehand (when the motor retracts, providing assistance for fingerextension), or to close (when the motor extends, allowing thefingers to flex). We provide two methods for detecting theintent of the patient, and compare them here.

1) Intent Detection via Ipsilateral EMG Signals: The firstmethod utilizes ipsilateral forearm surface EMG signals todetect the user intent as the patient attempts to use the affectedhand. If it can be realized, it has appealing characteristics: it ishighly intuitive (the patient simply attempts to open/close thehand as needed for the task) and can facilitate neuroplastictyas it closes sensorimotor loop on the impaired UL.

In this study, we use an EMG-based intent inferral methodintroduced previously [14], [24]. This approach relies on apattern recognition algorithm to detect user intention based ondata collected by commercial armband (Myo by Thalmic Labs)equipped with eight sensors, and does not require precisesensor placement. The armband is placed approximately oneinch proximal to the splint to avoid contact with the motor.

While previous work has shown that intuitive control isindeed possible, it has also highlighted a number of challenges.EMG signals are inherently abnormal in patients with hemi-paresis and can be distorted by spasticity and fatigue [35].As a result, when working with stroke patients while engagedin functional tasks, we found our current EMG-based intentdetection method to be effective only for a subset of patients.

2) Intent Detection via Contralateral Shoulder Movement:To account for this phenomenon, we introduce here a secondintent inferral method, using contralateral shoulder movement.This approach, often used for body powered UL prosthe-ses [36], provides a more robust control compared to EMG,as it relies on the unaffacted side. Additionally, compared toother non-EMG control methods, such as a button switch,it enables bimanual tasks since the unaffected hand is notrequired to operate the control. Thus, the focus with thismethod was on restoration of functionality, rather than neuro-recovery. It has the disadvantage of requiring the patient toengage in additional movement (elevating the contralateral

Fig. 4. Exotendon device with SH method.

shoulder) with the only purpose of providing a signal forour device. Such movement can be unintuitive, and of limitedrehabilitative value. However, it may follow that improvementsin functional use result in improved proximal strength andbilateral integration.

We use a shoulder harness worn on the unimpaired UL andused to detect shoulder movement. When shoulder depres-sion is detected, the device retracts to trigger hand openingthrough finger extension (Fig. 4). Shoulder flexion was ruledout to control release, as it promoted a flexor synergy inthe affected limb, while shoulder depression (often coupledwith relaxation/exhalation) appeared more favorable to facil-itate release. Conversely, when shoulder elevation (shrug) isdetected, the device extends to allow hand closure via fingerflexion.

A load cell (Futek, FSH00097) is installed in series witha suspender to measure the tension in the harness, and anextension spring next to the load cell connects a waist beltand the suspender to ease discomfort caused by the tension.Two different thresholds on the load cell signal are used todetect shoulder elevation and depression, in order to preventunnecessary motor oscillation. The thresholds are manuallycalibrated at the beginning of each session. In the rest ofthis study, we will refer to this intent inferral method as SH,shorthand for shoulder harness.

The main advantage of the contralateral SH intent inferralmethod over EMG is its robustness to differences in impair-ment patterns, since it relies exclusively on the unimpairedside. Still, we believe that the potential advantages of ipsilat-eral EMG control (more intuitive motor commands, closingthe loop on the affected side) outweigh the SH robustnessadvantage, as long as EMG control is applicable.

IV. CLINICAL INTERVENTION

In order to evaluate our active hand orthosis, we performeda clinical study aiming to quantify its performance as either arehabilitative or assistive device. The three main characteristicsof the study included the following: first, each training sessionconsisted of user-controlled functional interaction with every-day objects and simulated ADLs. This is intended to emulateuse of the impaired UL outside of clinical settings, which isour directional goal for the project. Second, each patient under-went twelve 30 minutes training sessions, distributed over thecourse of one month. The relatively large number of sessions(compared to our previous feasibility studies) was requiredboth to study rehabilitative effects, and to allow patients to



develop familiarity with the device and its controls, in orderto quantify performance as an assistive device. Third, our out-come measures post-intervention included clinical assessmentsperformed both assisted (with the device on, in order to studyassistive performance) and unassisted (without the device,in order to study rehabilitation effects). Post-intervention clin-ical outcomes without robotic assistance were compared tobaseline to evaluate rehabilitative effect while post-testingresults with robotic assistance were compared to baselineperformance and post-testing results without assistance toexamine assistive capability of the device. We present thedetails of our clinical intervention next.

A. Participants

Total twelve community-dwelling individuals with chronicstroke volunteered to participate in the study and met inclusioncriteria. Inclusion criteria were:

• Stroke diagnosis at least 6 months prior to start of study• Passive range of motion: wrist to neutral, digits within

normal limits• Moderate muscle tone, i.e., Modified Ashworth Scale

(MAS) ≤ 2 in digits, wrist, and elbow• Active range of motion: At least 30 degrees shoulder

flexion, 20 degrees shoulder abduction, 20 degrees elbowflexion, finger flexion within functional limits

• Strength: At least trace palpable finger extension• Able to successfully flex the fingers to form a grasp• Unable to extend all fingers fully without assistance• Intact cognition to provide informed written consentExclusion criteria were:• Concurrent participation in another study• Comorbid orthopedic condition/pain limiting functional

use of the impaired upper extremity• History or neurological disorder other than stroke• Excessive spasticity (MAS > 2)• Recent botox injection to the affected limb (< 13 weeks)5 participants had prior experience with the exotendon

device in varying capacities, but not within 6 months beforestart of the protocol. All subjects gave informed writtenconsent to participate and the protocol was approved bythe Columbia University Medical Center Institutional ReviewBoard. Participants were primarily recruited through a pre-existing, IRB-approved, research registry of stroke patients.Additionally, physiatrists in our clinic referred some oftheir patients. The trial was registered on ClinicalTrials.gov(NCT03767894). All training and testing sessions wereperformed under the supervision of an occupational or physicaltherapist.

B. Outcome Measures

1) Baseline Assessments: All clinical assessments were per-formed by an occupational therapist who was not involved inthe training protocol, though blinding was not possible in thisstudy design. For all testing sessions, the MAS was performedfirst since other measurement tools can cause fatigue. Afterthe MAS, the FM, ARAT, and BBT, were administered in

a randomized order to limit order effect. FM for UL is animpairment level measure of body structures that evaluatesreflexes, motor function, and joint range of motion of theUL [37], ARAT is an activity level assessment that involvesspecific grasp, grip, pinch, and gross motor tasks for theUL [38], and BBT is an activity level assessment that testsunilateral pinch and manual dexterity in a timed manner [39].The maximum score on the FM for UL is 66, and the FM canbe subscaled into FM-proximal (42/66), and FM-distal (24/66).Its estimated Minimal Clinically Important Difference (MCID)ranges from 4.25 to 7.25 [40]. The maximum score on theARAT is 57, and anchor-based MCID of the ARAT for chronicstroke is 5.7 [41].

2) Post Assessments: To evaluate rehabilitative and assis-tive effects of the training, participants completed post-testingassessments, split over the course of two sessions to avoidfatigue. One session involved administration of the FM, ARAT,and BBT without robotic assistance, while the other sessioninvolved administration of the ARAT and BBT with roboticassistance. The order of post-testing days was also random-ized. The FM was only performed at post-testing withoutrobotic assistance because the FM assesses capacity of the armprimarily through gross motor tasks, and comparatively fewgrasping and pinching tasks. We thus presumed that roboticassistance would have minimal influence on FM scores.

Post-testing without the device assesses motor recovery afterrobot-assisted training whereas post-testing with robotic assis-tance evaluates the assistive aspects of the device. We assumedthat the proposed intent detection methods, particularly EMG-based, will take time for users to learn, so the clinicalassessments are performed after 12 training sessions in orderto allow competent use of the device.

3) Statistical Analysis: Our primary outcome analysesinclude the FM, ARAT, and their subscales rated on allsubjects. The FM has two sub-scales: FM-proximal evaluatesthe shoulder and elbow, while FM-distal evaluates the wristand hand. The ARAT has 4 sub-scales: grasp, grip, pinch,and gross movement. Secondary outcome analyses include theFM and ARAT broken down by control method (EMG andSH), and BBT broken down by level of functionality of theparticipants. For all data, we report mean gains of unassistedpost-training score compared to baseline score to examinethe rehabilitative effect as well as assisted post-training scorecompared to baseline score and unassisted post-training scoreto assess the assistive effect.

We tested the primary clinical outcome data and gains fornormal distribution based on the residuals of our dependentvariables with Shapiro-Wilk and with Q-Q plots. We alsotested the homogeneity of variance using the Levene test.FM and its subsections, ARAT, ARAT-grasp, and ARAT-grippassed the normality and homogeneity test, and we appliedpaired sample t-tests. ARAT-pinch, ARAT-gross movementfailed the normality test, and we applied a nonparametricpaired sample Wilcoxon test which does not make assumptionsof normality or homogeneity of variance.

To determine statistical significance, we computedp-values for all of our primary measures. We utilized theBenjamini-Hochberg (BH) procedure to control the False



Discovery Rate (FDR) for mutiple comparisons [42], using0.05 as the overall FDR control level. We note that, whenusing the BH procedure, the threshold used for rejecting thenull hypothesis is different for each test, even if the overallcontrol level is the same. We thus report, for every test, boththe p-value and the BH significance threshold for rejectingthe null hypothesis. Our secondary measures are consideredpost-hoc analyses, and we report only mean gains with nostatistical analysis.

C. Protocol

After participating in the informed consent process, allparticipants were screened for inclusion, and those includedperformed baseline measurements. During the next screeningvisit, each participant was fitted with the exotendon deviceto avoid hyper-extension and was screened with the EMGclassifier to determine which control method they would usefor the study. During the control screening, a classifier wastrained with the subject’s EMG signals, and the user wasinstructed on how to use the control. The user was asked toopen, relax, and close their hand three times each, with theforearm on and off the table. If our EMG method was able toclassify user intention correctly under all conditions, and theuser could maintain each signal for at least 2 seconds on everyattempt, the user was assigned to the EMG group. Otherwise,the participant was assigned to the SH group.

We allowed the use of a mobile arm support (Saebo MAS)for participants who were clinically observed to have signif-icant difficulty performing the training protocol even withfrequent rest. Criteria for use of the arm support includedweakness (2-/5 to 3-/5 muscle grades for elbow flexionand shoulder flexion, abduction, and rotation) and significantfatigue limiting functional performance as observed by thetherapist. The subjects who met the criteria used the armsupport for the duration of training at a set level of support.The level of support was customized for each subject by thetherapist in order to optimize their ability to perform functionaltasks and limit the impact of shoulder fatigue on grasp training.

Each participant completed 12 training sessions, three timesper week for four weeks. Each training session was between60-90 minutes including time for set up, system classification,donning/doffing the device, and rest breaks as needed. Par-ticipants completed 30 minutes of active training during eachvisit. After the 12 sessions, participants completed two daysof post-testing, as described above, as well as a questionnairefor subjective feedback on their experience.

D. Training

The series of selected tasks reflect best-practice in ULprosthetic training (controls training, repetitive drills, andbimanual functional skills training) [43]. During controls train-ing, participants were educated on the device operation anddemonstrated proficiency in the control absent any objects orfunctional tasks. Participants then advanced to repetitive drills,which incorporated an array of objects of various shapes, sizes,and densities (Fig. 5).

Fig. 5. Real world objects used for treatment.

TABLE IBASELINE CHARACTERISTICS

Before donning the device, the therapist performed5 minutes of passive range of motion to all joints of the ULto help mitigate fluctuations in tone across sessions. Then,the patient donned the device. During each session, partici-pants completed 30 minutes of active training. Occasionally,breaks were provided upon patient request due to fatigue,or if any technical issues arose with the prototype device thatrequired an adjustment be made during the session.

The training protocol was always carried out in the sameorder, with basic tasks first, progressing to more complextasks. See Appendix for the list of tasks performed by sub-jects during each session. Some participants were not ableto complete the full training protocol during each session.In that case, they stopped after 30 minutes of training andthe last completed task was recorded. Some participants wouldcomplete the full protocol in less than 30 minutes. In that case,they continued working on grasp, transport, and release tasksof their choosing (to be client-centered) with oversight fromthe therapist for the duration of the 30 minutes.

V. RESULTS

Among 12 enrolled individuals, 11 subjects (6 males and5 females) completed the training and evaluations. One par-ticipant dropped out prior to the first training session due to amedical issue unrelated to the study, therefore all analysis isof 11 subjects. Years since stroke ranged from 2 to 22 years.9 patients had ischemic, 1 had hemorrogic, and 1 had unknowntype of stroke. There were 4 left affected and 7 right affectedpatients. 10 patients were right hand dominants and 1 wasleft hand dominant. 6 participants were screened to use EMGmethod and 5 subjects were assigned in SH group. Note thatthe subjects screened to the EMG group tended to have morefunctional use of their impaired UL at baseline, as notedby pre-testing scores compared to those assigned to the SHgroup (Table I). We examine the difference in clinical outcomebetween the two groups through secondary analyses. Onesubject used the arm support for training.



Fig. 6. FM(left) and ARAT(right) scores. Subject 5 dropped out due to a medical issue unrelated to the study.

TABLE IIGAINS FOR FM POST-INTERVENTION. BOLD DATA ARE STATISTICALLY

SIGNIFICANT (P-VALUE < BH SIGNIFICANCE THRESHOLD)

1) Fugl-Meyer Upper Extremity Scale: The FM results areshown in Table II. We note that, at baseline, ten subjectshad ‘no to poor’ UL capacity (


TABLE IVMEAN GAINS FROM FM, BBT AND ARAT POST-INTERVENTION FOR

SECONDARY ANALYSES. (A), (B) AND (C) ARE AS IN TABLE III

use my hand more. It gave me the feeling of freedom touse my hand again.” “I am able to fold and wring out awashcloth.” “If I could, I would wear it at home for most ofthe day for everything.” Participants also offered feedback fordevice improvements such as reduced wiring, less bulk aroundfingertips, increased training intensity (time and duration), andactuation of the thumb tendon for powered pinch.

VI. DISCUSSION

Overall, we identified trends in the data that suggest thisdevice might serve two distinct purposes for different subsetsof the stroke population - namely as a rehabilitation or assistivedevice. But, the results also highlight limitations of the device,and point towards possible areas for future improvements.

A. An Active Hand Orthosis for Rehabilitative Effects

From a rehabilitation perspective, we discuss resultsobtained post-intervention without using the device, and com-pared to baseline performance. Positive gains noted on theFM-distal subtest suggest that training with the device canserve as a rehabilitative tool to remediate some functional usein the affected UL, especially to help improve hand function-ality. In our secondary analyses, we note that the magnitudeof gains on the FM-distal was larger in the EMG group thanthe SH group, suggesting that the restorative training effectmay be greater in participants with some residual baselinefunctioning.

Based on the observation of a positive trend on the ARAT,we posit that increasing the intensity and duration of the inter-vention in future studies may lead to increased gains quantifiedusing this measure. For example, small gains captured on theFM (e.g. ability to actively flex or extend the fingers) maynot be captured on the ARAT because the improvement inrange of motion was not sufficient to translate into increasedfunctional ability (e.g. ability to pick up a small object).

B. An Active Hand Orthosis for Assistive Effects

We focus here on performance measured post-interventionwith the participants actively using the device, and compareagainst pre-training baseline or post-training without roboticassistance. ARAT results suggest that, for stroke patients, usingthe robot as an assistive device for long term compensation toincrease functionality in daily life may be feasible, althoughefficacy of the assistive capacity of this device has yet to bedemonstrated.

The improvement in the Grasp category of the ARATwas expected as the device specifically assists with graspingtasks. However, the magnitude of improvement was differ-ent depending on control method, as seen in the secondaryanalyses. We speculate that the differences between EMG andSH groups may be related to baseline differences in handfunctioning. EMG group participants tended to have moreresidual functioning at baseline, often employing compen-satory patterns to achieve grasp and/or pinch, whereas whenwearing the robotic device, the bulky finger components mayhave impaired their performance by making it more difficultto pick up and place objects in tight spaces. In contrast,those in the SH group had less functionality at baseline, andtherefore the device assisted their ability to pick up objects,though they still had similar difficulty placing objects intight spaces.

The outcome measures highlighted a few limitations of thedevice. Improvement in Grasp category of the ARAT andnegative mean gains in all other categories implies that thedevice facilitated grasping of mid-size objects, but not smallobjects which require pinching. We speculate this was becausethe thumb was splinted into a stable opposition position againstthe other digits. This had the advantage of reducing actuatorload, and we found this pose to be effective when graspingmid-size objects. However, a static thumb also made it difficultfor subjects to stably hold small objects in a pinching pattern.This behavior likely affected the results in both the ARATand BBT. As in recent studies, assisting the thumb has beenshown effective in small object grasping [45], [46], we areplanning to address this issue in future studies by designingan actuated thumb component which enables assisted pinch inaddition to enveloping larger objects.

Another limitation was poor performance of the EMGmethod due to abnormal muscle synergy in unregistered ULpostures. The BBT score drop in the EMG group with roboticassistance was likely because of unstable EMG classificationwhen the user had to lift the arm higher due to the partition andheight of the box. We note that the training sessions containedno action item that involved lifting an object higher than theheight of the partition (15.2 cm), thus the participants did notget an opportunity to practice their proximal muscles or learnto control the device while lifting the arm high in our protocol.

The current design is not fully self-contained. Power isprovided externally, and motor commands are sent from thecomputer over a wire. We are envisioning the next version con-taining a belt pack (weighing approximately 300 g) housing abattery pack and other electronics, including a microcontrollerfor issuing motor commands.



C. Limitations

It is important to point out a number of limitations related tostudy design in this pilot case series. In particular, we did notuse a control group consisting of patients receiving treatmentof similar intensity and duration, but without robotic assis-tance. However, meaningful motor recovery with traditionalphysical therapy for chronic stroke patients with moderateto severe motor impairments is considered rare [47], [48].In addition, our robotic-assistance enabled training tasks thatwere not possible for most participants otherwise, and thus cannot be replicated with traditional therapy. Our study also didnot comprise a follow up assessment to observe the durabilityof gains. We plan to address these limitations in the future.

We also note that assignments between the EMG and SHgroups was not performed in randomized fashion, but ratherbased on the ability of our intent inferral algorithm to classifyEMG signals. As a result, we noticed systematic differencesbetween the groups, with SH participants generally havinglower baseline functionality. This limits our ability to interpretdifferences in the results obtained by the two groups.

In addition, we note that rehabilitation studies strive toassess progress using outcome measures that are at theparticipation-level, observing and rating task performance inreal-life environments. However, the FM measure, consideredas the gold-standard in stroke research due to well establishedpsychometric properties and MCID [40], only assesses bodystructures at the impairment level, focusing on the capacity ofthe UL to move. ARAT and BBT are activity-level assessmentsthat involve observing and scoring participants performingsimulated functional tasks that are shorter in duration and morehighly scripted compared to ADLs. We believe customizedoutcome measures that capture higher task variation and allowlonger completion times might be better suited for capturingprogress when using robotic devices in an assistive fashion.

While it is important for patients to be able to don and doffa wearable device without assistance, we are still far fromquantifying this characteristic. With the current prototype,supervision is required as it would be for traditional therapy,but in the future as the device is further refined, it is hopedthat patients will be able to use the device independently athome after initial training with a therapist.

Finally, we did not conduct any structured interviews orask the users to rate the device using standard usability scales.However, participants were asked to provide open-ended, qual-itative feedback. Furthermore, a trained occupational therapistsupervised each training session and thoroughly monitoredfor pressure, redness, pain, and any other signs of distressthroughout the study. There have been no complaints frompatients reported in this study. As the system is furtherrefined, we plan to use standard questionnaires, such as SystemUsability Scale (SUS) or Likert scale.

VII. CONCLUSION

In this work, we presented clinical outcomes after 12 train-ing sessions for a study using a robotic hand orthosis. Ourdevice is designed to assist the paretic hand after stroke,focusing primarily on an impairment pattern characterized by

difficulty with active finger extension. Two main design goalsfor our device are wearability and user-driven operation: weuse two different methods to infer the intent of the user, andthus to determine when to provide assistance.

Post-intervention FM sub-scores suggest the grasp exerciseshelped improve distal movements of UL whereas it did nothave a significant impact on proximal segments. This resultsuggests the possibility for using our orthosis as a rehabilitativedevice for the hand. Assisted ARAT scores show that thedevice can indeed function in an assistive role for strokepatients. However, the results should be cautiously interpretedbecause of the limited sample size, lack of a control group,and the fact that the outcomes did not meet MCID for eitherFM or ARAT. We are planning to address this aspect throughboth improvements to the device and extended training periodsin future work.

Our study also underscored limitations of the device. In par-ticular, the device disrupts compensatory grasp patterns devel-oped by stroke survivors, leading to an immediate decrease infunctionality when the device is removed. It is likely that the12 sessions were not long enough to enable learning of newgrasp patterns for participants. Furthermore, our current designrelies on a static, passively splinted thumb, which enablesgross grasp but is not suited for pinching smaller objects.

Nevertheless, we believe that this work can highlightthe potential and feasibility for wearable and user-drivenrobotic hand orthoses. Such devices may enable roboticbased-hand rehabilitation during daily activities (as opposedto isolated hand exercises with limited UL engagement) andover extended periods of time, even in a patient’s homeenvironment. Numerous challenges must still be overcomein order to achieve this vision, related to design (compactdevices with easier donning/doffing), control (robust yet intu-itive intent inferral), and effectiveness (improved functionalityin a wider range of metrics). However, if these challenges canbe addressed, wearable robotic devices have the potential togreatly extend the use and training of the affected UL afterstroke, and help improve the quality of life for a large patientpopulation.

APPENDIXTRAINING PROCEDURE

Participants picked up and released selected objects 5 timeswith their forearm resting on the table top (supported reach)and then 5 times with their forearm lifted off the table(unsupported reach) to simulate functional reach. The objectsused in this task include (1) 2.5 cm wooden cube, (2) 5 cmwooden cube, (3) tennis ball, (4) 4 cm diameter toiletrybottle, and (5) 13 cm tall, tapered, hard plastic cup. Objectswere positioned in various locations to optimize the functionalenvelope. The objects were then arranged on a tray with 1 inchraised lip and participants removed all 5 items from the traytwice, and then replaced all items onto the tray twice.

Next, participants picked up and released the followingirregularly shaped objects twice each: (1) cotton ball, (2) 1 inchrubber ball, and (3) washcloth.

Lastly, participants advanced to bimanual functional skillstraining to best simulate real-life conditions and the additional



physical and cognitive challenge of operating the device with-out exclusive attention to its performance. The tasks were com-pleted twice each with participants using their affected handto stabilize and their unaffected hand to perform manipulation:(1) removing and replacing the cap from a broad line marker,(2) removing and replacing the cap from a standard tube oftoothpaste (screw off), (3) removing and replacing the capfrom a screw top beverage bottle, (4) removing and replacingthe wide mouth screw cap from large coffee container, (5)using a wooden spoon to stir in a small bowl for 10 seconds(affected hand stabilized bowl), (6) using a butter knife tomake 2 cuts in a ‘log’ of theraputty to mimic cutting food(affected hand stabilized theraputty), (7) opening a lock witha key (affected hand held lock), and (8) opening a sealedsandwich-size ziploc bag.

REFERENCES

[1] E. S. Lawrence et al., “Estimates of the prevalence of acute strokeimpairments and disability in a multiethnic population,” Stroke, vol. 32,no. 6, pp. 1279–1284, Jun. 2001.

[2] J. G. Broeks, G. J. Lankhorst, K. Rumping, and A. J. H. Prevo,“The long-term outcome of arm function after stroke: Results of afollow-up study,” Disability Rehabil., vol. 21, no. 8, pp. 357–364,Jan. 1999.

[3] C. E. Lang et al., “Dose response of task-specific upper limb trainingin people at least 6 months poststroke: A phase II, single-blind, ran-domized, controlled trial,” Ann. Neurol., vol. 80, no. 3, pp. 342–354,2016.

[4] E. J. Schneider, N. A. Lannin, L. Ada, and J. Schmidt, “Increas-ing the amount of usual rehabilitation improves activity after stroke:A systematic review,” J. Physiotherapy, vol. 62, no. 4, pp. 182–187,Oct. 2016.

[5] G. Levanon, B. Colijn, B. Cheng, and M. Paterra, “From not enoughjobs to not enough workers: What retiring baby boomers and thecoming labor shortage mean for your company,” in Proc. Conf.Board Market Rep., 2014. [Online]. Available: https://www.conference-board.org/topics/publicationdetail.cfm?publicationid=2819

[6] H. I. Krebs, “Twenty + years of robotics for upper-extremity reha-bilitation following a stroke,” in Rehabilitation Robotics. Amsterdam,The Netherlands: Elsevier, 2018.

[7] P. S. Lum, C. G. Burgar, P. C. Shor, M. Majmundar, andM. Van der Loos, “Robot-assisted movement training compared withconventional therapy techniques for the rehabilitation of upper-limbmotor function after stroke,” Arch. Phys. Med. Rehabil., vol. 83, no. 7,pp. 952–959, Jul. 2002.

[8] R. S. Calabrò et al., “Who may benefit from armeo power treatment?A neurophysiological approach to predict neurorehabilitation outcomes,”PM&R, vol. 8, no. 10, pp. 971–978, Oct. 2016.

[9] G. Gillen, Stroke Rehabilitation: A Function-Based Approach.Amsterdam, The Netherlands: Elsevier, 2015.

[10] Q. Qian et al., “Distal versus proximal—An investigation on differentsupportive strategies by robots for upper limb rehabilitation after stroke:A randomized controlled trial,” J. NeuroEng. Rehabil., vol. 16, no. 1,p. 64, Dec. 2019.

[11] O. Lambercy, R. Ranzani, and R. Gassert, “Robot-assisted rehabil-itation of hand function,” in Rehabilitation Robotics. Amsterdam,The Netherlands: Elsevier, 2018, pp. 205–225.

[12] S. Park, L. Bishop, T. Post, Y. Xiao, J. Stein, and M. Ciocarlie, “On thefeasibility of wearable exotendon networks for whole-hand movementpatterns in stroke patients,” in Proc. IEEE Int. Conf. Robot. Autom.(ICRA), May 2016, pp. 3729–3735.

[13] S. Park, L. Weber, L. Bishop, J. Stein, and M. Ciocarlie, “Design anddevelopment of effective transmission mechanisms on a tendon drivenhand orthosis for stroke patients,” in Proc. IEEE Int. Conf. Robot. Autom.(ICRA), May 2018, pp. 2281–2287.

[14] C. Meeker, S. Park, L. Bishop, J. Stein, and M. Ciocarlie, “EMG patternclassification to control a hand orthosis for functional grasp assistanceafter stroke,” in Proc. Int. Conf. Rehabil. Robot. (ICORR), Jul. 2017,pp. 1203–1210.

[15] F. Vanoglio et al., “Feasibility and efficacy of a robotic device forhand rehabilitation in hemiplegic stroke patients: A randomized pilotcontrolled study,” Clin. Rehabil., vol. 31, no. 3, pp. 351–360, Mar. 2017.

[16] E. A. Susanto, R. K. Tong, C. Ockenfeld, and N. S. Ho, “Efficacyof robot-assisted fingers training in chronic stroke survivors: A pilotrandomized-controlled trial,” J. NeuroEng. Rehabil., vol. 12, no. 1, p. 42,Dec. 2015.

[17] C. E. Lang, J. R. MacDonald, and C. Gnip, “Counting repetitions: Anobservational study of outpatient therapy for people with hemiparesispost-stroke,” J. Neurologic Phys. Therapy, vol. 31, no. 1, pp. 3–10,Mar. 2007.

[18] P. Heo, G. M. Gu, S.-J. Lee, K. Rhee, and J. Kim, “Current handexoskeleton technologies for rehabilitation and assistive engineering,”Int. J. Precis. Eng. Manuf., vol. 13, no. 5, pp. 807–824, May 2012.

[19] P. Polygerinos, Z. Wang, K. C. Galloway, R. J. Wood, and C. J. Walsh,“Soft robotic glove for combined assistance and at-home rehabilitation,”Robot. Auto. Syst., vol. 73, pp. 135–143, Nov. 2015.

[20] I. Hussain, G. Salvietti, G. Spagnoletti, and D. Prattichizzo, “The soft-SixthFinger: A wearable EMG controlled robotic extra-finger for graspcompensation in chronic stroke patients,” IEEE Robot. Autom. Lett.,vol. 1, no. 2, pp. 1000–1006, Jul. 2016.

[21] D. Brunelli, A. M. Tadesse, B. Vodermayer, M. Nowak, andC. Castellini, “Low-cost wearable multichannel surface EMG acquisitionfor prosthetic hand control,” in Proc. 6th Int. Workshop Adv. SensorsInterfaces (IWASI), Jun. 2015, pp. 94–99.

[22] F. Ryser, T. Butzer, J. P. Held, O. Lambercy, and R. Gassert, “Fullyembedded myoelectric control for a wearable robotic hand orthosis,” inProc. Int. Conf. Rehabil. Robot. (ICORR), Jul. 2017, pp. 615–621.

[23] P. Weber, E. Rueckert, R. Calandra, J. Peters, and P. Beckerle, “A low-cost sensor glove with vibrotactile feedback and multiple finger jointand hand motion sensing for human-robot interaction,” in Proc. 25thIEEE Int. Symp. Robot Hum. Interact. Commun. (RO-MAN), Aug. 2016,pp. 99–104.

[24] S. Park, C. Meeker, L. M. Weber, L. Bishop, J. Stein, and M. Ciocarlie,“Multimodal sensing and interaction for a robotic hand orthosis,” IEEERobot. Autom. Lett., vol. 4, no. 2, pp. 315–322, Apr. 2019.

[25] R. A. Bos et al., “A structured overview of trends and technologies usedin dynamic hand orthoses,” J. NeuroEng. Rehabil., vol. 13, no. 1, p. 62,Dec. 2016.

[26] L. Connelly, Y. Jia, M. L. Toro, M. E. Stoykov, R. V. Kenyon, andD. G. Kamper, “A pneumatic glove and immersive virtual realityenvironment for hand rehabilitative training after stroke,” IEEE Trans.Neural Syst. Rehabil. Eng., vol. 18, no. 5, pp. 551–559, Oct. 2010.

[27] S. Ates, C. J. W. Haarman, and A. H. A. Stienen, “SCRIPT passiveorthosis: Design of interactive hand and wrist exoskeleton for rehabili-tation at home after stroke,” Auto. Robots, vol. 41, no. 3, pp. 711–723,Mar. 2017.

[28] N. A. Lannin et al., “Upper limb motor training using a Saeboorthosis is feasible for increasing task-specific practice in hospital afterstroke,” Austral. Occupational Therapy J., vol. 63, no. 6, pp. 364–372,Dec. 2016.

[29] P. Boissy, D. Bourbonnais, M. M. Carlotti, D. Gravel, andB. A. Arsenault, “Maximal grip force in chronic stroke subjects and itsrelationship to global upper extremity function,” Clin. Rehabil., vol. 13,no. 4, pp. 354–362, Aug. 1999.

[30] A. Yurkewich, D. Hebert, R. H. Wang, and A. Mihailidis, “Handextension robot orthosis (HERO) glove: Development and testing withstroke survivors with severe hand impairment,” IEEE Trans. Neural Syst.Rehabil. Eng., vol. 27, no. 5, pp. 916–926, May 2019.

[31] H. T. Peters, S. J. Page, and A. Persch, “Giving them a hand: Wear-ing a myoelectric Elbow-Wrist-hand orthosis reduces upper extremityimpairment in chronic stroke,” Arch. Phys. Med. Rehabil., vol. 98, no. 9,pp. 1821–1827, Sep. 2017.

[32] S. Jain and B. Argall, “Recursive Bayesian human intent recognitionin shared-control robotics,” in Proc. IEEE/RSJ Int. Conf. Intell. RobotsSyst. (IROS), Oct. 2018, pp. 3905–3912.

[33] F. J. Valero-Cuevas, M. E. Johanson, and J. D. Towles, “Towards arealistic biomechanical model of the thumb: The choice of kinematicdescription may be more critical than the solution method or the vari-ability/uncertainty of musculoskeletal parameters,” J. Biomech., vol. 36,no. 7, pp. 1019–1030, Jul. 2003.

[34] J. Arata, K. Ohmoto, R. Gassert, O. Lambercy, H. Fujimoto, and I. Wada,“A new hand exoskeleton device for rehabilitation using a three-layeredsliding spring mechanism,” in Proc. IEEE Int. Conf. Robot. Autom.,May 2013, pp. 3902–3907.



[35] B. Cesqui, P. Tropea, S. Micera, and H. Krebs, “EMG-based patternrecognition approach in post stroke robot-aided rehabilitation: A feasi-bility study,” J. NeuroEng. Rehabil., vol. 10, no. 1, p. 75, 2013.

[36] A. van Lunteren, G. H. M. van Lunteren-Gerritsen, H. G. Stassen,and M. J. Zuithoff, “A field evaluation of arm prostheses for unilat-eral amputees,” Prosthetics Orthotics Int., vol. 7, no. 3, pp. 141–151,Dec. 1983.

[37] A. R. Fugl-Meyer, L. Jääskö, I. Leyman, S. Olsson, and S. Steglind,“The post-stroke hemiplegic patient. 1. a method for evaluation ofphysical performance,” Scand. J. Rehabil. Med., vol. 7, no. 1, pp. 13–31,1975.

[38] R. C. Lyle, “A performance test for assessment of upper limb functionin physical rehabilitation treatment and research,” Int. J. Rehabil. Res.,vol. 4, no. 4, pp. 483–492, Dec. 1981.

[39] V. Mathiowetz, G. Volland, N. Kashman, and K. Weber, “Adult normsfor the box and block test of manual dexterity,” Amer. J. OccupationalTherapy, vol. 39, no. 6, pp. 386–391, Jun. 1985.

[40] S. J. Page, G. D. Fulk, and P. Boyne, “Clinically important differencesfor the upper-extremity fugl-meyer scale in people with minimal tomoderate impairment due to chronic stroke,” Phys. Therapy, vol. 92,no. 6, pp. 791–798, Jun. 2012.

[41] J. H. van der Lee, V. de Groot, H. Beckerman, R. C. Wagenaar,G. J. Lankhorst, and L. M. Bouter, “The intra- and interrater reliabilityof the action research arm test: A practical test of upper extremityfunction in patients with stroke,” Arch. Phys. Med. Rehabil., vol. 82,no. 1, pp. 14–19, Jan. 2001.

[42] Y. Benjamini and Y. Hochberg, “Controlling the false discovery rate:A practical and powerful approach to multiple testing,” J.Roy. Stat. Soc., B (Methodol.), vol. 57, no. 1, pp. 289–300,Jan. 1995.

[43] S. S. Johnson and E. Mansfield, “Prosthetic training: Upper limb,” Phys.Med. Rehabil. Clinics, vol. 25, no. 1, pp. 133–151, 2014.

[44] M. H. Hoonhorst, R. H. Nijland, J. S. van den Berg, C. H. Emmelot,B. J. Kollen, and G. Kwakkel, “How do fugl-meyer arm motor scoresrelate to dexterity according to the action research arm test at 6 monthspoststroke?” Arch. Phys. Med. Rehabil., vol. 96, no. 10, pp. 1845–1849,Oct. 2015.

[45] A. Yurkewich, I. J. Kozak, D. Hebert, R. H. Wang, and A. Mihailidis,“Hand extension robot orthosis (HERO) grip glove: Enabling indepen-dence amongst persons with severe hand impairments after stroke,”J. NeuroEng. Rehabil., vol. 17, no. 1, pp. 1–17, Dec. 2020.

[46] S. R. Soekadar et al., “Hybrid EEG/EOG-based brain/neuralhand exoskeleton restores fully independent daily living activi-ties after quadriplegia,” Sci. Robot., vol. 1, no. 1, Dec. 2016,Art. no. eaag3296.

[47] P. W. Duncan, L. B. Goldstein, D. Matchar, G. W. Divine, andJ. Feussner, “Measurement of motor recovery after stroke. Outcomeassessment and sample size requirements,” Stroke, vol. 23, no. 8,pp. 1084–1089, Aug. 1992.

[48] H. Y. Jung, J. S. Yoon, and B. S. Park, “Recovery of proximaland distal arm weakness in the ipsilateral upper limb after stroke,”NeuroRehabilitation, vol. 17, no. 2, pp. 153–159, Jun. 2002.


/ColorImageDict > /JPEG2000ColorACSImageDict > /JPEG2000ColorImageDict > /AntiAliasGrayImages false /CropGrayImages true /GrayImageMinResolution 150 /GrayImageMinResolutionPolicy /OK /DownsampleGrayImages true /GrayImageDownsampleType /Bicubic /GrayImageResolution 600 /GrayImageDepth -1 /GrayImageMinDownsampleDepth 2 /GrayImageDownsampleThreshold 1.50000 /EncodeGrayImages true /GrayImageFilter /DCTEncode /AutoFilterGrayImages true /GrayImageAutoFilterStrategy /JPEG /GrayACSImageDict > /GrayImageDict > /JPEG2000GrayACSImageDict > /JPEG2000GrayImageDict > /AntiAliasMonoImages false /CropMonoImages true /MonoImageMinResolution 300 /MonoImageMinResolutionPolicy /OK /DownsampleMonoImages true /MonoImageDownsampleType /Bicubic /MonoImageResolution 900 /MonoImageDepth -1 /MonoImageDownsampleThreshold 1.33333 /EncodeMonoImages true /MonoImageFilter /CCITTFaxEncode /MonoImageDict > /AllowPSXObjects false /CheckCompliance [ /None ] /PDFX1aCheck false /PDFX3Check false /PDFXCompliantPDFOnly false /PDFXNoTrimBoxError true /PDFXTrimBoxToMediaBoxOffset [ 0.00000 0.00000 0.00000 0.00000 ] /PDFXSetBleedBoxToMediaBox true /PDFXBleedBoxToTrimBoxOffset [ 0.00000 0.00000 0.00000 0.00000 ] /PDFXOutputIntentProfile (None) /PDFXOutputConditionIdentifier () /PDFXOutputCondition () /PDFXRegistryName () /PDFXTrapped /Unknown

/CreateJDFFile false /Description >>> setdistillerparams> setpagedevice

Date post:	26-Jan-2021
Category:	Documents
Upload:	others
View:	1 times
Download:	0 times

User-Driven Functional Movement Training With a Wearable ... › rehab › sites › default ›...

Documents