Training and assessment of upper limb motor function with a roboticexoskeleton after stroke

Antonio Frisoli, Member, IEEE,Edoardo Sotgiu, Caterina Procopio,

Massimo Bergamasco, Member, IEEE,*

Carmelo Chisari, Giuseppe Lamola,and Bruno Rossi**

Abstract— This paper presents the results of the evaluationtraining performed in a group of chronic stroke patients withLight-Exoskeleton device. The rehabilitation treatment includedpassive, assisted and active reaching exercises based on aimpedance control strategy with gain modulated by the patient’sability to perform the task. The effects of training were assessedboth by means of clinical evaluation in terms of Fugl-Meyerand Modified Ashworth assessment scales, and of functionalevaluation, by means of Bimanual Activity test. Moreover, ateach session an automatic assessment of performance was madethrough two robotic measures (task time and smoothness ofmovement), and these were analysed in term of correlationwith the outcome of functional evaluation. Interestingly wefound a significant improvement of both clinical and functionalevaluation, and that the automatic assessment performed bythe robot in terms of smoothness represents a strong predictorof transfer of functional ability to activity of every day life.The high correlation observed between functional outcome androbotic measures suggest the last ones can provide a rapidand useful feedback about the patient’s recovery progress,in addition to the pre- and post- clinical and functionalmeasurements.

I. INTRODUCTIONCerebrovascular disease, i.e. stroke, represents the first

cause of acquired permanent disability in adults and thethird leading cause of death in industrialized countries.Even after rehabilitation, about 80% of post-stroke patientsstill suffer upper limb disability, which impairs their dailyliving activities and often leads to major incapacities [1].The mechanisms of functional recovery after stroke remainlargely unknowns, but animal models and correlated humanstudies demonstrate that functional recovery after stroke isobtained through the use-dependent reorganization of neuralmechanisms, exploiting basic properties of neural plasticity[2]. Current practices in standard care after stroke mainlyrely on manual techniques provided by therapists. The basisof such techniques is task-oriented training, which combinescognitive and motor rehabilitation [3]. Importantly, suchrehabilitation has to be intense, to start very early, and tobe individualized depending on patient needs [4]. Recently,the researchers have focused increasing attention on thequantitative evaluation of patient’s motor abilities in order to

This work was supported in part by the the EU project VERE, and bythe IIT Seed national project Bravo

*A. Frisoli, E. Sotgiu, C. Procopio and M. Bergamasco are withthe PERCRO, TeCIP Scuola Superiore Sant’Anna, Pisa, Italy.e-mail:a.frisoli@sssup.it

**C. Chisari and B. Rossi are with Neurorehabilitation Unit,Department of Neurosciences, University of Pisa, Pisa, Italy.c.chisari@ao-pisa.toscana.it

obtain an objective evaluation of the rehabilitation treatmenteffects. Although the reliability and high acceptance, thestandard clinical and functional scales have shown somedrawback mainly due to high subjectivity, long duration andlow resolution. The consequence is that clinicians need moreefficient and more objective strategies for the assessment ofpatient recovery and, on the other hand, for the delivery ofa high intense and controlled of rehabilitation training.In the last two decades, many studies have proved thepotential of robot-assisted rehabilitation arm in providing agreat help to accelerate the recovery of upper limb function,with positive results on strength and motor function ofpatients [5], [6]. In addition, although the lasting effects onthe quality of the activities of daily living of patients remainsto be confirmed [7], robotic technology offers the opportunityof new therapeutic strategies based on recurrent evaluationof motor function to optimize the dose of the training andtailor the rehabilitation program to patients’ needs [8]. Fromthe perspective of therapists, robotic technologies providesomewhat unique opportunities towards higher volume andmore controlled rehabilitation. However, it might be evenmore important that robotic technologies and Virtual Realityallow to monitor the progress of patient recovery, in order togive useful and reliable performance feedback to the patient,and allow for on-line multisensory feedback, that may leadto a better retention of regained functions[9], [10], [11].Many studies have carried out on the robotic rehabilitationtherapy and their effectiveness evaluated at the start and atthe end of protocol by means of the standard clinical andfunctional scales. However, few of the performed researcheshave shown a correlation between the clinical outcomes andthe robot measurements [12], [13].The main difficult to propose some of the robotic measuresused in the past studies as a standard feedback of therapyprogress, depends on their task or device peculiarity [12],[14], [15]. Two of these measures, commonly based on thereaching movement, are the smoothness and the executiontime of movement, which were selected as robotic measuresfor the present study. Moreover, movement kinematics maybe used to assess movement recovery in a more precise way[16], [17], going far beyond clinical tests that often do notclarify how the movement has been built, and what is adue to motor recovery or compensation. As a consequence,measuring movement smoothness not only informs aboutpatient’s capability of executing a given movement, but alsoabout movement quality [18], [19], [20].

The Fourth IEEE RAS/EMBS International Conferenceon Biomedical Robotics and BiomechatronicsRoma, Italy. June 24-27, 2012

978-1-4577-1198-5/12/$26.00 ©2012 IEEE 1782

A. Active Rehabilitation Exoskeletons

A class of rehabilitation robots that deserve a particularinterest in upper limb rehabilitation is constituted by activeexoskeletons: robotic devices with kinematic isomorphic tothe human arm, that can be worn on the user upper limb.They present several significant advantages compared to end-effector based system [21]. They may track the full armkinematics, so not only the hand, but the full kinematicchain of the arm, and consequently apply a force assist at thelevel of each joint independently, e.g. shoulder, elbow andwrist, offering the unique opportunity to re-train interjointsynergies. The workspace covered by these systems is threedimensional and large, so that the training can be conductedon fully spatial movements and along different directions. Inthe simplest way the assistance provided by the robot canconsist in a gravity counterbalancing of weight of the arm,as this has been proved to enlarge the workspace of the arm[22] and to be effective for motor recovery. Alternativelythe robot can provide a guided assistance to the task to beperformed, according to an impedance-based model or moreadvanced controls [23].In this work we present the last results of the robotic-basedtraining and clinical assessment for upper limb rehabilitationperformed with a robotic exoskeleton training system inhemiplegic patients after stroke. The aim of this work isto define some automatic measurements of motor functionassociated to the training exercises, and to analyse thecorrelation of these indexes with the outcome of standardclinical measures.

II. METHODS

A. Instruments

The rehabilitation platform used for this clinical treatmentis composed of the the Light-Exoskeleton device (L-Exos)mounted on a mobile support of device, a control systemand a graphic workstation with associated display [24]. TheL-Exos [25], [21] is a right-arm rehabilitation robot used forstroke patients rehabilitation; it is composed of five degree-of-freedom (DoF), of which four active, i.e. shoulder andelbow flexion/extension, shoulder internal/external rotationand, abduction/adduction Fig.1a. This device is capable toexert a controlled force up to 100N on the hand palm ofthe patient. The weight of each moving link is totally com-pensated by the controller, while the inertia and dynamicsof the device are assumed to be negligible, due to thelow masses (each actuator is posed on the fixed base, asshown in [25]) and low speeds involved in the exercises. Thehigh level control system provides three different strategiesfor movement training: Passive Mode (Impedance BasedAssistance), Assisted mode (Triggered Gain Position Mode)and Free Mode (Direct Force Mode and Counterbalancingassistance). The Assisted Mode implement an “assistanceas needed” training strategy based on a gain scheduling. Aset of VR-based exercises (shown in Fig.1b) can specificallytrain arm movements in the frontal, ipsilateral and contralat-eral areas. The system is endowed with the virtual reality

(a) The Light-Exoskeleton device worn by a patient duringa rehabilitation session and schematic representation of activejoints.

(b) Snapshots of two virtual exercises.

Fig. 1. Arm joints controlled by the L-exos (1a) and the exercises designedfor the recovery of function in tasks of achieving objects (1b).

training scenarios projected on a stereoscopic projection wallembedding the rehabilitation exercises and the graphic userinterface for controlling the system parameters and adaptingthe exercises to the specific patients needs.

B. Outcome and Clinical Assessment

The patients were evaluated by means of functional andclinical tests at both enrollment and discharge. Assessmenttook place during one week before the start of training,within one week after training had stopped, and at 5 monthsfollow-up. The primary outcome measure was the BimanualActivity Test (BAT), consisting in evaluating in terms oftime and quality of movement (estimated on 0-4 scale) theexecution by the patient of a set of Activities of DailyLife (ADL) tasks requiring bimanual coordination. The Fugl-Meyer Assessment scale (FMA) [26] was used as clinical testfor the evaluation of motor function, sensation qualities andjoint function in hemiplegic patients. Only the upper limbsubsection (66 points) was considered for this study. TheModified Ashworth Scale (MAS) [27] was used as qualitativescale for the assessment of spasticity (measure resistance topassive stretch). The chosen set of clinical measures wasjustified by providing both motor impairment and functionalability measures. In addition, the clinical reliability of thesescales (mainly the FMA) with the stroke population has beenproved in past researches [28], [29].

1783

C. Participants

Stroke survivors who volunteered to participate in thestudy were people who met the following inclusion criteria:aged between 18 and 80 years, able to understand the studypurpose and procedure, and provide informed consent, mod-erate upper limb emiparesis, MAS greater then 2, stabilizedbefore the beginning of the treatment.Ten patients, of which 7 males and 3 females and with strokeevent occurred at least six months ahead, were enrolled inthe experimental group that was administered the robotictreatment. Two of the patients were unable to complete thetreatment and they were excluded from the study. The resteight patients (2 females and 6 males, 62.9±9.9 years old)were enrolled in the treatment and scheduled to perform sixweeks of training with three sessions per week. Six of theeight patients underwent also a follow-up clinical assessment5 months after the end of treatment to evaluate the retentionof regained function. All participants provided informedconsent in accordance with the Declaration of Helsinki priorto participation in this study, which was approved by theEthical Committee of the Cisanello hospital in Pisa.

D. Protocol

Patients who satisfied the inclusion criteria were instructedto desist from different rehabilitation treatment during thestudy period. The rehabilitation training consisted in twotraining exercises and one evaluation test, performed withthe assistance of the L-Exos. The first exercise consisted ina reaching task (see the left side of Fig.1b) performed withan impedance assistance by the robot with level of assistancemodulated by the robot according to the position error(Passive Mode), or with variable gain adjusted according tothe force input by the patient (Assisted Mode), to enhancepatient’s active participation in the task. The second trainingexercise consisted in a training scenario in which the patienthas to compose a virtual puzzle with the gravity assistanceonly provided by the robot (Free Mode)). The exercise wasthen followed by an evaluation session (see the right sideof Fig.1b) allowing to assess the performance in movementexecution over different direction of space, since a set oftarget to be reached where placed in the vertical plane, alongtwelve possible positions arranged equally spaced along acircumference, resembling the hours position of a clock.Difficulty of sessions was tailored to the patients ability. Inorder to reduce the influence of fatigue and to avoid thestarting of spasticity, 1 minute rest period was administeredafter every trial. The absolute hand position and the angularjoints displacements were recorded at a sampling rate of1 kHz. The tangential velocity were calculated by derivingthe hand position and digitally filtered by using a low-passsecond order forward and backward Butterworth filter withcut-off frequency at 5 Hz. To assess the improvement ofmotor recovery, two robotic performance indexes, measuredin the evaluation session, were selected:

• Execution Time (ET), computed as the time to movefrom the start position in the centre to the peripheral

target. The movement time was defined as the differencebetween the movement onset and offset, which weredefined as the times when the tangential velocity re-spectively rose above and fell down 5% of the maximumvelocity.

• Smoothness Index (SI), computed by counting the num-ber of peaks in the velocity profile for each movement.A peak was counted if the increase from a minimum tothe next maximum of the norm of the tangential speedwas above 15% of the global maximum speed. As aconsequence a decrease of the smoothness index meansan increase of smoothness of movement.

III. RESULTS

We observed a significant clinical improvement in allpatients. Clinical assessment at the enrollment the averageFMA assessment score was of 24.6±12.9 and the averageMAS of 14±8.8; at the end of treatment the outcomerevealed an increase of FMA to 34.3±17.2 (p<0.01), withno significant change in spasticity (MAS after treatment10.8±6.9). On the functional level (BAT), we observed asignificant reduction of time in the execution of movementfrom 17.05±3.59 s to 13.15±3.77 s (p<0.005) with anoverall increase of quality of movement from 1.99±0.77to 2.94±0.92 (p<0.05). The results of the clinical testsperformed by the subjects pre- and post-treatment are shownin Tab.I.

The assessment of the two indexes of performanceachieved with the robot revealed a similar trend, as shown inFig.2: a marked decrease of movement time from 2.55±1.15to 1.02±0.68 s (p<0.005) and a marked increase of smooth-ness going from 6.63±0.92 to 2.75±1.54 (p<0.005).

Fig. 2. Assessment of movement performance with the L-Exos.

Moreover, we investigated the correlation of the BATfunctional index with the robot assessment scale, to assesswhether the improvement of performance observed with therobot was effectively transferred as ability to perform dailyliving activities. A strong correlation was observed betweenthe two robotic performance indexes and the functionalindexes of the BAT scale (p<0.05), as shown in Fig.3,

1784

TABLE ICLINICAL SCORES OF TREATED STROKE SUBJECTS BEFORE (PRE) AND AFTER TREATMENT (POST).

Subject Birth Year Gender FMA (66pts) MASBAT

time qualityPRE POST PRE POST PRE POST PRE POST

P1 61 M 17 21 12 11 19,36 14,95 1,91 3,58P2 78 M 41 47 0 -3 15,84 11,73 2,95 3,58P3 67 M 16 14 6 13 20,94 19,79 0,95 3,58P4 72 F 20 31 24 17 18,43 14,66 1,83 2,29P5 57 M 12 28 27 19 20,04 16,87 1,74 2,61P6 50 M 35 55 13 10 19,81 11,01 2,42 2,79P7 55 F 13 19 13 13 10,60 8,68 1,08 1,08P8 51 M 43 59 17 6 11,35 7,53 3,00 4,00

Mean 62,86 24,63 34,25 14,00 10,75 17,05 13,15 1,99 2,94SD 9,96 12,88 17,21 8,82 6,86 3,59 3,77 0,70 0,92

Paired t-test two-tailed (p) 0,008 0,143 0,002 0,015

confirming that the improvement of performance observedwith the robot was effectively transferred as ability toperform ADL activity. The performance improvement in

(a)

(b)

Fig. 3. Correlation of BAT scales outcomes with both robotic performanceindexes.

movement execution of the evaluation exercise was analysedover different directions of space by means of the robot-based assessment, in terms of polar diagrams shown in Fig.4.In Fig.4a the means and standard deviations of differencesbetween execution time at first and last session of treatment,averaged over each patient and over all sessions, for eachof twelve direction are shown; while in Fig.4b the sameanalysis is depicted about the smoothness index. A signifi-cant improvement is observed mainly in the ipsilateral space(p<0.05), while no significant improvement is obtained in

the contralateral space.

(a)

(b)

Fig. 4. Improvement of smoothness index and time execution observed(Mean±SD) along different direction from central position (**:p<0.01,*:p<0.05, o:0.05<p<0.1).

In the group of patients that underwent a follow-up eval-uation at month 5 (shown in Fig.5), we found a percentage

1785

improvement of FMA of 39% that was maintained also atthe follow-up assessment with a percentage improvement of22% (p<0.01), indicating a retention of the acquired motorskill.

Fig. 5. Follow-up evaluation (FMA) performed at 5 months.

IV. DISCUSSION

The aim of this clinical study was to evaluate the feasibilityof robotic therapy in chronic stroke patients. Subjects weremotivated to participate to the training sessions and the im-provement observed in smoothness of movement representsa predictive indicator of both clinical assessment and ofrecovered ability to perform activities of everyday life. Thisrepresents an important result, since it allows mapping di-rectly the measured performance to improvements in terms offunctional activity. Furthermore, the first preliminary resultsfrom follow-up evaluation performed at 5 months of distancesince treatment allowed us also to hypothesize that such aregain of function is retained over time. These data wereconfirmed also by analysing the Execution Time requiredto move virtual cubes from the central point of a circlealong different radial directions, performing movements in avertical plane. Changes in Execution Time and SmoothnessIndex, before and after robotic training, where larger formovements performed in the ipsilateral space, which con-stitutes a very important goal in rehabilitation of the upperarm after stroke. This confirms that exoskeleton systemsin combination with specifically design exercises might beused not only to enhance the process of recovery, but alsoto evaluate the progress of therapy. In confirmation of theobtained results, some of the patients declared to have notedlittle increase right about the stabilization and precision ofreaching movements during some common daily life tasks.

V. CONCLUSIONS

Robotic technologies offer the unique opportunity to indi-vidually tailor the physical therapy during neurorehabilitationtraining to the specific patient’s needs, creating the conditionsto enhance the patient involvement in the training program,through the interplay of different factors, such as motivation,feedback, controlled assistance and ecological presentation ofexercises. Results from our studies demonstrate that robotictraining can be effective both in terms of clinical efficacyand regain of motor function, assessed by functional tests in

everyday life actions. Moreover, the recording capabilities ofrobotic technologies make it possible to perform a continuousevaluation of motor performance through quantitative andobjective indexes that correlate well with clinical scales,providing an on-line and dynamic assessment of patient’sstatus and a record of the recovery process.

VI. ACKNOWLEDGMENTS

This research was supported by Integrated Project fundedby UE under the Sixth Framework Programme (FP6)SKILLS (Project IST-2006-035005-SKILLS). The authorsgratefully acknowledge the contribution of the Dr. MartinaVenturi.

REFERENCES

[1] G. Kwakkel, B. Kollen, and J. Twisk, “Impact of time on improvementof outcome after stroke.” Stroke; a journal of cerebral circulation,vol. 37, no. 9, pp. 2348–2353, Sept. 2006.

[2] M. Hallett, “Plasticity of the human motor cortex and recovery fromstroke,” Brain Research. Brain Research Reviews, vol. 36, no. 2-3, pp.169–174, 2001.

[3] P. Langhorne, J. Bernhardt, and G. Kwakkel, “Stroke rehabilitation,”Lancet, vol. 377, no. 9778, pp. 1693–1702, 2011.

[4] S. D. Horn, G. DeJong, R. J. Smout, J. Gassaway, R. James, andB. Conroy, “Stroke rehabilitation patients, practice, and outcomes:is earlier and more aggressive therapy better?” Archives of physicalmedicine and rehabilitation, vol. 86, no. 12 Suppl 2, Dec. 2005.

[5] A. C. Lo, P. D. Guarino, et al., “Robot-Assisted Therapy for Long-Term Upper-Limb Impairment after Stroke,” New England Journal ofMedicine, vol. 362, no. 19, pp. 1772–1783, May 2010.

[6] G. B. Prange, M. J. A. Jannink, C. G. M. Groothuis-Oudshoorn, H. J.Hermens, and M. J. IJzerman, “Systematic review of the effect ofrobot-aided therapy on recovery of the hemiparetic arm after stroke.”Journal of Rehabilitation Research and Development, vol. 43, no. 2,p. 171, 2006.

[7] J. Mehrholz, T. Platz, J. Kugler, and M. Pohl, “Electromechanical androbot-assisted arm training for improving arm function and activitiesof daily living after stroke.” Cochrane database of systematic reviews(Online), no. 4, 2008.

[8] C. E. Han, M. A. Arbib, and N. Schweighofer, “Stroke rehabilitationreaches a threshold.” PLoS computational biology, vol. 4, no. 8, 2008.

[9] D. Jack, R. Boian, et al., “Virtual reality-enhanced stroke rehabilita-tion,” Neural Systems and Rehabilitation Engineering, IEEE Transac-tions on, vol. 9, no. 3, pp. 308–318, 2002.

[10] M. Cameirao, S. Bermudez I Badi, E. Duarte Oller, and P. Verschure,“Augmenting the Rehabilitation Gaming System with haptics andorthosis.” Frontiers: Annual CyberTherapy and CyberPsychology,2009.

[11] M. Johnson, X. Feng, L. Johnson, and J. Winters, “Potential of asuite of robot/computer-assisted motivating systems for personalized,home-based, stroke rehabilitation,” Journal of NeuroEngineering andRehabilitation, vol. 4, no. 1, p. 6, 2007.

[12] R. Colombo, F. Pisano, et al., “Robotic techniques for upperlimb evaluation and rehabilitation of stroke patients.” IEEETransactions on Neural Systems and Rehabilitation Engineering,vol. 13, no. 3, pp. 311–324, 2005. [Online]. Available:http://www.ncbi.nlm.nih.gov/pubmed/16200755

[13] O. Celik, M. K. O?Malley, C. Boake, H. S. Levin, N. Yozbatiran,and T. A. Reistetter, “Normalized movement quality measures fortherapeutic robots strongly correlate with clinical motor impairmentmeasures.” IEEE Transactions on Neural Systems and RehabilitationEngineering, vol. 18, no. 4, pp. 433–444, 2010. [Online]. Available:http://www.ncbi.nlm.nih.gov/pubmed/20388607

[14] R. J. Sanchez, J. Liu, et al., “Automating arm movement training fol-lowing severe stroke: functional exercises with quantitative feedbackin a gravity-reduced environment,” Neural Systems and RehabilitationEngineering, IEEE Transactions on, vol. 14, no. 3, pp. 378–389, 2006.

[15] L. Zollo, L. Rossini, M. Bravi, G. Magrone, S. Sterzi, andE. Guglielmelli, “Quantitative evaluation of upper-limb motor controlin robot-aided rehabilitation.” Med. Biol. Engineering and Computing,vol. 49, no. 10, pp. 1131–1144, 2011.

1786

[16] A. Roby-Brami, A. Feydy, M. Combeaud, E. V. Biryukova, B. Bussel,and M. F. Levin, “Motor compensation and recovery for reaching instroke patients.” Acta Neurologica Scandinavica, vol. 107, no. 5, pp.369–381, 2003.

[17] S. K. Subramanian, J. Yamanaka, G. Chilingaryan, and M. F. Levin,“Validity of movement pattern kinematics as measures of arm motorimpairment poststroke.” Stroke: A Journal of Cerebral Circulation,vol. 41, no. 10, pp. 2303–2308, 2010.

[18] H. I. Krebs, N. Hogan, M. L. Aisen, and B. T. Volpe, “Robot-aidedneurorehabilitation.” IEEE Transactions on Rehabilitation Engineer-ing, vol. 6, no. 1, pp. 75–87, 1998.

[19] F. Amirabdollahian, R. Loureiro, and W. Harwin, “Minimum jerk tra-jectory control for rehabilitation and haptic applications,” in Roboticsand Automation, 2002. Proceedings. ICRA ’02. IEEE InternationalConference on, vol. 4, 2002, pp. 3380–3385.

[20] O. Rohrer, S. Fasoli, et al., “Movement smoothness changes duringstroke recovery,” J Neurosci, vol. 22, pp. 8297–8304, 2002.

[21] A. Frisoli, F. Salsedo, M. Bergamasco, B. Rossi, and M. Carboncini,“A force-feedback exoskeleton for upper-limb rehabilitation in virtualreality,” Applied Bionics and Biomechanics, vol. 6, no. 2, pp. 115–126,2009.

[22] T. Sukal, M. Ellis, and J. Dewald, “Shoulder abduction-inducedreductions in reaching work area following hemiparetic stroke: neuro-scientific implications,” Experimental Brain Research, vol. 183, no. 2,pp. 215–223, Nov. 2007.

[23] A. Frisoli, L. Borelli, et al., “Robot-mediated arm rehabilitation invirtual environments for chronic stroke patients: A clinical study,”Robotics and Automation, 2008. ICRA 2008. IEEE International

Conference on, pp. 2465–2470, May 2008.[24] A. Frisoli, E. Sotgiu, C. Procopio, M. Bergamasco, B. Rossi, and

C. Chisari, “Design and implementation of a training strategy inchronic stroke with an arm robotic exoskeleton,” in RehabilitationRobotics (ICORR), 2011 IEEE International Conference on, 29 2011-july 1 2011, pp. 1–8.

[25] Salsedo F., Bergamasco M., Dettori A., Franceschini M., Frisoli A.,and R. F., “Exoskeleton interface apparatus,” Patent WO2002IB05 66420 021 231, 2004.

[26] A. R. Fugl-Meyer, L. Jaasko, I. Leyman, S. Olsson, and S. Steglind,“The post-stroke hemiplegic patient. 1. a method for evaluationof physical performance.” Scandinavian journal of rehabilitationmedicine., vol. 7, no. 1, pp. 13–31, 1975. [Online]. Available:http://view.ncbi.nlm.nih.gov/pubmed/1135616

[27] R. W. Bohannon and M. B. Smith, “Interrater reliabilityof a modified ashworth scale of muscle spasticity.” PhysicalTherapy, vol. 67, no. 2, pp. 206–207, 1987. [Online]. Available:http://www.ncbi.nlm.nih.gov/pubmed/3809245

[28] F. Malouin, L. Pichard, C. Bonneau, A. Durand, and D. Corriveau,“Evaluating motor recovery early after stroke: comparison of the Fugl-Meyer Assessment and the Motor Assessment Scale.” Archives ofphysical medicine and rehabilitation, vol. 75, no. 11, pp. 1206–1212,nov 1994.

[29] D. J. Gladstone, C. J. Danells, and S. E. Black, “The fugl-meyerassessment of motor recovery after stroke: a critical review of its mea-surement properties.” Neurorehabilitation and Neural Repair, vol. 16,no. 3, pp. 232–240, 2002.

1787