Development of a robotic upper limb assessment to configure a serious game

Abstract

BACKGROUND

ROBiGAME project aims to implement serious games on robots to rehabilitate upper limb (UL) in stroke patients. The serious game characteristics (target position, level of assistance/resistance, level of force) are adapted based on the patient’s assessment before and continuously during the game (measuring UL working area, kinematics and muscle strength).

OBJECTIVE

To develop an UL robotic motor assessment protocol to configure the serious game.

METHODS

32 healthy subjects and 20 stroke patients participated in the study. Subjects were clinically assessed (UL length and isometric force) and using a robot. The robot assessment consisted of the patient’s UL working area (WA), the UL isometric and isokinetic force in three directions and the UL kinematics during a pointing task toward targets placed at different distances.

RESULTS

The WA and the UL isometric force were moderately to highly correlated with clinical measures (respectively ρ= 0.52; p = 0.003, ρ= 0.68–0.73; p < 0.001). Ratios between the UL isokinetic force generated on three directions were established. The velocity and straightness indexes of all subjects increased when subjects had to reach to targets placed more distantly (r= 0.82–0.90; ρ= 0.86–0.90 respectively; p < 0.001).

CONCLUSIONS

This protocol can be integrated into a serious game in order to continuously configure the game characteristics to patient’s performance.

Keywords

Stroke upper limb robotics motor assessment serious game

1 Introduction

Trial registration: The study was registered at ClinicalTrials.gov (NCT02543424) the 12th of August 2015.

Upper limb (UL) rehabilitation of stroke patients remains a challenge (Langhorne, Bernhardt, & Kwakkel, 2011; Pollock et al., 2014). Despite intensive interdisciplinary rehabilitation, many patients have persistent neurological impairments (Mackay & Mensah, 2004) that cause deficits in motor and/or cognitive functions (Jaillard, Naegele, Trabucco-Miguel, LeBas, & Hommel, 2009; Pollock et al., 2014) limiting patient’s activities and restricting their social participation. Rehabilitation robots and serious games constitute two innovative technologies that are increasingly used to rehabilitate these impairments (Mehrholz, Pohl, Platz, Kugler, & Elsner, 2015).

Robot-assisted therapy (RAT) follows current guidelines for neurorehabilitation as it intensifies therapy, provides assistance as needed, quantifies measurements of the patient’s movement performance, and delivers feedback to the patient (Langhorne et al., 2011; Pignolo, 2009). Recent systematic reviews recommend RAT for stroke rehabilitation to reduce UL impairment and improve execution of daily living activities (Mehrholz et al., 2015; Norouzi-Gheidari, Archambault, & Fung, 2012; Pollock et al., 2014).

Serious games also provide new promising technology in neurorehabilitation (Burke et al., 2009). A serious game is a game that does not have entertainment, enjoyment or fun as primary purpose. It can play an important role for increasing motivation during the education (Chen et al., 2003). One of the main interests is the continuous adaptation of the exercise characteristics according to the patient’s individual performances (Borghese, Pirovano, Lanzi, Wuest, & de Bruin, 2013). Indeed, the success rate of an exercise is a key factor in rehabilitation. If the exercise is too difficult, the patient will always fail, and not recover new abilities. Conversely, if the exercise is too easy, the patient will always succeed, without learning new abilities (Cameirao, Badia, Oller, & Verschure, 2010). A balance between the two that is not too difficult and not too easy therefore promotes learning. Some studies have recommended to use a success rate between 50% and 70% to maximize recovery (Cameirao et al., 2010; Perry et al., 2011). The entertaining aspect of serious games is also essential in order to increase the enjoyment and the patient’s involvement in treatment. Studies have shown the serious games potential to improve UL motor control (da Silva Cameirao, Bermudez, Duarte, & Verschure, 2011) and reduce hemineglect (Katz et al., 2005).

The combination of RAT and serious games could be effective for stroke patient’s rehabilitation, providing improvements in both motor control and cognition. A previous study suggested an interest to rehabilitate motor and cognitive impairments together to put patients in real-world settings (Cameirao, Faria, Paulino, Alves, & Bermudez, 2016). According to these observations, the ROBiGAME project was set up to develop a serious game on a rehabilitation robot in order to rehabilitate simultaneously UL motor and cognitive impairments among stroke patients (Heins et al., 2017). The ROBiGAME’s design is shown in Fig. 1 (Heins et al., 2017). During the serious game (e.g., for the target pointing exercise), the robot allows for assistance as needed based on the patient’s performance. The robotic assistance is presented in Fig. 2 (Galinski, 2014; Sapin, Dehez, & Gilliaux, 2017). UL strengthening exercises are considered as an adjunct to traditional post stroke rehabilitation (Winstein et al., 2016). Therefore, the serious game also includes some strengthening exercises while the robot provides a resistance to patient’s movement. To regulate the game characteristics, patients UL are evaluated before and continuously during the game. These characteristics and their regulations are:

The placement of game targets, based on the patient’s UL active and passive working area (WA) assessment.

The robotic assistance, configured according to the patient’s UL active and passive WA, and UL kinematics (velocity, movement smoothness and straightness).

The robotic resistance and the level of force that the patient needs to generate in the game during UL strengthening exercises, configured by the patient’s UL isokinetic and isometric force.

Fig.1

Schema of the ROBiGAME project (Heins et al., 2017). The goal is to adapt game difficulty to individual patient’s performance. The ROBiGAME can include several types of task, with each one being associated with a set of task-specific parameters characterising task difficulty (p). The desired task difficulty is determined within a single regulator, which computes a set of difficulty indicators (y) that are common to all the types of task. An input interpreter translates these indicators into task parameters, based on the patient’s evaluation (s). Then, during task performance, the game receives information from captors (w) and reacts in function of the recorded data. When the task is completed, all information measured during the task (z) is sent to the output interpreter, which translates the information into performance indicators (x). The performance indicators are finally sent to the regulator, which sets the difficulty of the next task. For example, the game scenario of the target pointing exercise, the patient is a sandwich store manager and has to prepare the client’s order in a limited time period. To do so, the patient has to reach targets (ingredients) that are displayed on the screen one by one with the handle of the robot, and bring the target object to the starting point.

Fig.2

Illustration of (A) the REAplan^® robot, (B) the assistance-as-needed and (C) the positioning of the hand in the glove and the forearm in the gutter. During RAT with REAplan^®, sensor measurements of force and position were sent to a control unit that determines assistance based on the difference between the patient’s hand position and the reference trajectory. Two types of force can assist the patient: a lateral interaction force (F_lat) and a longitudinal interaction force (F_long). F_lat helps the patient to follow the reference trajectory, and the control unit changes the stiffness coefficient, denoted K_lat (higher K_lat corresponding to more lateral assistance), based on the movement straightness. F_long helps the patient to move at a reference velocity, and the control unit changes the damping coefficient, denoted C_long (high C_long corresponding to stronger longitudinal interaction force), that is usually based on the movement smoothness (to assist the movement) (Sapin et al., 2017). Moreover, F_long can also act as a resistance force to the patient’s movement if he or she moves faster than the reference velocity. In this case, the control unit could change C_long based on the patient’s isokinetic force (C_long proportional to the patient’s isokinetic force).

The aim of the present study was to develop a robotic method to determine UL working area, UL kinematics and UL muscle strength. The assessed variables and reference values will then be used to regulate the serious game for stroke patients.

2 Methods

2.1 Subjects

A convenience sample of thirty-two healthy subjects and twenty chronic stroke patients participated in the study. Patients were recruited in the Cliniques universitaires Saint-Luc (Belgium). To be included in the study, healthy subjects had to be aged between fifty and eighty years (to match with the mean age of stroke patients), they had to be able to understand instructions and they should not present with any neurological or orthopaedic illness. For stroke patients, inclusion criteria were to have a stroke diagnosis confirmed by magnetic resonance imaging since more than 6 months, to present with UL hemiparesis, to be able to mobilise voluntarily the end-effector of the robot, to be able to understand instructions and carry out all study tests. Exclusion criteria were any other neurological or orthopaedic illness limiting UL function. The subjects’ characteristics are shown in Table 1. They were all volunteers who participated freely to the study after providing written informed consent. The study was approved by the ethics board of our Faculty of Medicine of the Université catholique de Louvain.

Table 1
Characteristics of subjects

Healthy subjects (n = 32) Stroke patients (n = 20)

Age (years), mean (SD) 59.7 (7.09) 59.3 (11.65)

Sex (male/female), n 13/19 13/7

BMI (kg/m²), mean (SD) 24.5 (3.06) 25.6 (2.91)

Time since stroke (months), mean (SD) / 33.3 (22.71)

Dominant arm (right/left), n 28/4 /

Affected arm (right/left), n / 8/12

Upper limb length (cm), mean (SD) 54.7 (3.79) 54.6 (3.47)

Fugl Meyer Assessment Upper Extremity (%), mean (SD) / 69.2 (29.50)

	Healthy subjects (n = 32)	Stroke patients (n = 20)
Age (years), mean (SD)	59.7 (7.09)	59.3 (11.65)
Sex (male/female), n	13/19	13/7
BMI (kg/m²), mean (SD)	24.5 (3.06)	25.6 (2.91)
Time since stroke (months), mean (SD)	/	33.3 (22.71)
Dominant arm (right/left), n	28/4	/
Affected arm (right/left), n	/	8/12
Upper limb length (cm), mean (SD)	54.7 (3.79)	54.6 (3.47)
Fugl Meyer Assessment Upper Extremity (%), mean (SD)	/	69.2 (29.50)

Abbreviations: BMI = Body Mass Index, SD = Standard Deviation, kg = kilogram, m² = square meter.

2.2 REAplan robot^®

The REAplan^® robot (Axinesis, Wavre, Belgium) was used in this study (Fig. 2). The REAplan^® is a distal end-effector robot that allows movements of the UL by the mobilisation of the hand in the horizontal plane. A screen is placed in front of the patient to provide a visual feedback of the position of the end-effector and the force exert by the patient on the end-effector. The robot is equipped with force and position sensors to record the position of the handle and the force exerted on the handle over time (frequency = 125 Hz).

2.3 Study design

All subjects were evaluated by clinical and robotic assessments. The dominant UL was evaluated in healthy subjects and the paretic UL in stroke patients.

2.4 Clinical assessment

Stroke patient’s motor control was evaluated with the computerized and adaptive testing system of the Fugl Meyer Assessment Upper Extremity (Hou et al., 2012). The UL length was measured in healthy subjects in a standing posture with the UL open wide along the trunk, and the measurement taken from the lateral edge of the acromion to the radius styloid process. Isometric maximal voluntary contraction (MVC) was assessed using the MicroFet 2^® hand-held dynamometer (Hogan Health Industries, Inc. 8020 South 1300 West, West Jordan, UT, USA). UL position and dynamometer placement during the evaluation were taken according to the manufacturer’s instruction manual (Fig. 3). The subject rested on his/her back with the arm along the trunk and the elbow flexed at 90°. For the assessment of the elbow flexor muscles, the dynamometer was place at the distal-third part of the forearm. For the elbow extensor muscles, the shoulder was flexed at 90°, and the dynamometer was placed at the distal-third part of the ulna. After a training period, the subjects were asked to generate their MVC for three seconds. Three consecutive muscular contractions were performed with 30 seconds rest between each measurement. The results were expressed as the mean value of the three MVC.

Fig.3

Illustration of the UL position and the dynamometer placement during the assessment of (A) elbow flexor muscles force and (B) elbow extensor muscles force.

2.5 Robotic assessment with REAplan^® robot

The position of the subjects was standardised. They were seated on a chair with the angle between the trunk and the thighs maintained at 120° to reduce lower back stress. When necessary, they could rest their feet on a footrest to ensure greater stability. The subject’s trunk was strapped to the back of the chair to limit compensatory movements (Gilliaux et al., 2014). The subject was centred to the midline of the REAplan^® working surface where the starting position of end-effector was located. Depending on his/her ability, if necessary, the patient hand was attached with a glove to the distal end-effector or his/her forearm was fixed in a gutter to increase stability of the paretic wrist and forearm Fig. 2.

For the present study, the robot was always in a passive mode, i.e. providing no assistance. The robotic evaluation consisted of four assessment tasks: (i) the UL working area (WA) (to determine the position of targets that are reachable by the patients); (ii) the global UL isometric force and (iii) isokinetic force during a pushing forward and pulling backward movement (to determine the force that the patient was able to generate, and the level of resistance that can be given to the patient), and; (iv) the UL kinematics toward targets positioned at different distances (to evaluate the effect of amplitude and direction on kinematics indexes in order to adapt the robotic assistance). These different variables are described in more details below.

The UL WA corresponds to the area that the subject was able to reach with the REAplan^® horizontal working surface. A total of 30 targets were evenly distributed in the total WA (Fig. 4). First, the therapist assessed the subject’s passive WA by mobilising the end-effector and patient’s hand, with the patient’s UL being fully relaxed. Then, the patient’s active WA was assessed by asking the patient to mobilise the end-effector with their UL by themselves. The surface covered by the patient was painted in bleu to give him/her a visual feedback. In both cases, the maximum reach across the full surface was recorded. The patient’s WA was computed by the ratio between the surface covered by the patient (actively or passively) and the total surface of REAplan^® (%; a higher score indicates greater WA). The patient’s relative WA was calculated by the ratio between the active and passive WA. The centre of the WA corresponded to the centre of the surface covered by the patient either during active or passive mobilisations.

Fig.4

Typical trace of passive (grey) and active (black) working area (WA) from a right hemiparetic stroke patient. Typical trace of a stroke patient was presented to show the distinction between active and passive WA. However, healthy subjects covered the same area in active and in passive. The grey circles represent the 30 targets evenly placed across the WA of the REAplan^®, and the white circle represents the initial position of the end-effector.

End-effector robots allow a global movement assessment of the end-effector kinematics and the global force exerted by the UL on the end-effector. Hence, global UL force was assessed using isometric and isokinetic assessment during pushing forward (PFF) and pulling backward (PBF) movements. The isometric PFF and PBF were assessed with the end-effector fixed in front of the subject, the elbow flexed at 90°. Subjects were asked to generate their MVC three times alternating PFF and PBF, with a rest period of 15 seconds between each muscular contraction. A visual feedback was displayed to the subject on the screen by a trick meter that fills up according to the amount of force that was generated by him/her. The results were expressed as the mean value of the three MVC. The hypothesis was that the PFF and PBF assessed with the robot correlated well with the elbow extensor and flexor muscles force assessed with the dynamometer.

The maximal isokinetic PFF and PBF were assessed at a constant velocity of 18 cm/s in three directions (– 45°= 45° diagonal on the contralateral side from the assessed UL, 0°= in front of the subject and 45°= 45° diagonal on the ipsilateral side from the assessed UL). During each movement, patients were asked to perform their movement with the MVC. Each movement was performed three times consecutively, alternating PFF and PBF, with a resting period of 15 seconds between each direction. The position of the end-effector was presented on the patient’s screen. For each movement, the MVC was recorded and the MVC of the three movements was then averaged. The MVC in each direction was compared to each other.

Subjects were asked to reach targets presented in the patient’s active WA to analyse the effects of direction and amplitude on UL kinematics (Fig. 4). The movement amplitude corresponds to the distance between the starting point and the target position. Each target was reached 3 times consecutively. The position of the end-effector was presented on the patient’s screen (ratio 1:1). The kinematic parameters were mean velocity, straightness (ratio between movement amplitude and path length covered by the subject; ratios closer to 1 indicate more rectilinear paths) and smoothness (ratio between the mean speed and peak speed; higher ratio indicates smoother movements) (Gilliaux et al., 2014). Kinematic indexes were computed for each movement and the average of three measurements were then calculated.

2.6 Statistical analysis

Statistical analyses were performed using the SigmaStat 3.5 software (WPCubed GmbH, Munich, Germany). The normality and equality of variances were checked by the Shapiro-Wilk and Brown-Forsythe statistical tests respectively, and the significance level was fixed at 0.05.

The relationship between the active WA of healthy subjects and their UL length was analysed to validate the WA assessment task. None of the statistical tests comparing WA and UL length were made in stroke patients because their WA can also depend on the UL paresis, contracture or pain. The relationships between PFF and elbow extensor muscles isometric force, and PBF and elbow flexor muscles isometric force were analysed for all participants (healthy subjects and stroke patients) to validate the isometric force assessment task through a large range of force measures. The isokinetic force generated in three directions were analysed in each group separately (healthy subjects and stroke patients). The relationships between the kinematic indexes, movement amplitude (corresponding to the distance between the starting point and the target) and movement direction were also analysed in each group separately.

To analyse the different relationships, Pearson correlations were used when data were normally distributed, and the coefficients were classified as follows: 0–0.25 = very low, 0.26–0.49 = low, 0.5–0.69 = moderate, 0.7–0.89 = high and 0.9–1.00 =very high (Souza et al., 2014). If data were not normally, the Spearman correlation was used, with the coefficients rated as: 0–0.30 = little to no correlation, 0.30–0.5 = fair, 0.5–0.70 = moderate and 0.70–1 = high correlation (Golubović & Slavković, 2014). To compare the force measured with the REAplan^® and the Microfet^®, a paired t-test was used. To compare the force generated by healthy subjects and stroke patients, a t-test was used. To evaluate the effect of the direction of mobilisation during the isokinetic force assessment, a one-way _RMANOVA was used in each group separately. If this test showed an interaction, it was further analysed with a post-hoc analysis (Tukey Test).

3 Results

All subjects completed the session with full collaboration and none experienced adverse events or made any complaints. Results of healthy subjects and stroke patients are shown in Tables 2 and 3.

Table 2
Results of the upper limb motor assessments tasks

Healthy subjects Stroke patient

n = 32 n = 20

Mean (SD) Mean (SD)

Working area

Active working area, % 71.0 (9.05) 51.7 (16.54)

Passive working area, % / 58.2 (12.35)

Relative working area, % / 88.8 (20.50)

Isometric force

Pulling backward/flexor muscles

Microfet 2^®, N 150.6 (36.95) 77.8 (50.06)

REAplan^®, N 152.5 (29.65) 94.8 (59.52)

Pushing forward/extensor muscles

Microfet 2^®, N 110.2 (30.50) 73.5 (51.31)

REAplan^®, N 138.6 (34.38) 83.4 (52.68)

Isokinetic force

Pulling backward

– 45°, N 70.8 (28.73) 33.8 (22.01)

0°, N 159.7 (24.07) 64.1 (56.42)

45°, N 80.4 (42.95) 32.0 (23.95)

Pushing forward

– 45°, N 60.7 (26.57) 31.7 (19.64)

0°, N 151.4 (29.38) 61.9 (57.87)

45°, N 90.0 (36.26) 44.1 (29.16)

Kinematics

Straightness 0.96 (0.02) 0.93 (0.03)

Velocity, cm/s 17.2 (3.08) 12.5 (2.75)

Smoothness 0.57 (0.03) 0.48 (0.03)

	Healthy subjects	Stroke patient
Working area
Active working area, %	71.0 (9.05)	51.7 (16.54)
Passive working area, %	/	58.2 (12.35)
Relative working area, %	/	88.8 (20.50)
Isometric force
Pulling backward/flexor muscles
Microfet 2^®, N	150.6 (36.95)	77.8 (50.06)
REAplan^®, N	152.5 (29.65)	94.8 (59.52)
Pushing forward/extensor muscles
Microfet 2^®, N	110.2 (30.50)	73.5 (51.31)
REAplan^®, N	138.6 (34.38)	83.4 (52.68)
Isokinetic force
Pulling backward
– 45°, N	70.8 (28.73)	33.8 (22.01)
0°, N	159.7 (24.07)	64.1 (56.42)
45°, N	80.4 (42.95)	32.0 (23.95)
Pushing forward
– 45°, N	60.7 (26.57)	31.7 (19.64)
0°, N	151.4 (29.38)	61.9 (57.87)
45°, N	90.0 (36.26)	44.1 (29.16)
Kinematics
Straightness	0.96 (0.02)	0.93 (0.03)
Velocity, cm/s	17.2 (3.08)	12.5 (2.75)
Smoothness	0.57 (0.03)	0.48 (0.03)

Abbreviations: SD = Standard deviation; N = Newton; – 45°= 45° diagonal on the contralateral side from the assessed upper limb; 0°= in front of the subject; 45°= 45° diagonal on the ipsilateral side from the assessed upper limb.

Table 3

Correlations between the clinical assessment and the robotic assessment

Robotic assessment	Clinical measures	Correlation coefficient
Active working area (healthy subjects)	Upper limb length	ρ= 0.52
Pulling backward force (healthy subjects	Elbow flexor muscles force	ρ= 0.73
and stroke patients)
Pushing forward force (healthy subjects	Elbow extensor muscles force	ρ= 0.68
and stroke patients)
Averaged velocity (healthy subjects)	Movement amplitude	r= 0.90
Averaged velocity (stroke patients)	Movement amplitude	r= 0.82
Movement straightness (healthy subjects)	Movement amplitude	ρ= 0.86
Movement straightness (stroke patients)	Movement amplitude	ρ= 0.90
Movement smoothness (healthy subjects)	Movement amplitude	ρ= 0.11
Movement smoothness (stroke patients)	Movement amplitude	ρ= 0.01

3.1 UL working area

Healthy subjects active WA corresponded on average 71±9% of the total working space of the REAplan^®. This means that healthy subjects could reach around 20 of the 30 available targets. The active WA of healthy subjects showed a moderate correlation with their UL length (ρ= 0.52, p = 0.003) (Fig. 5).

Fig.5

The relationship between the active working area (WA) and the upper limb length in healthy subjects.

Stroke patients’ active average WA corresponded to 52±16% of the total working space of the REAplan^®. This means that patients could reach around 15 out of the 30 targets. A typical right hemiparetic trace of the passive and active WA from an example stroke patient is shown in Fig. 4. Their active WA was significantly smaller than their passive WA (p = 0.018), and the relative WA (ratio between active and passive WA) of stroke patients was 89%. Healthy subjects’ and stroke patients’ WA was lateralized to the assessed side. The average distance between the midline of the REAplan^® working surface and the centre of the WA was 4.6±2.3 cm (healthy subjects) and 4.1±4.7 cm (stroke patients).

3.2 Isometric force

Among healthy subjects, maximum PFF measured on the REAplan^® robot ranged from 41 N to 188 N, with an average of 138.6±34.38 N. PBF ranged from 66 N to 181 N, with an average of 152.5±29.65 N. Among stroke patients, maximum PFF ranged from 11 N to 187 N, with an average of 83.4±52.68 N, and PBF ranged from 28 N to 187 N, with an average of 94.8±59.52 N.

PBF for all participants was highly correlated with elbow flexor muscles force (ρ= 0.73; p < 0.001), and PFF was moderately correlated with elbow extensor muscles force (ρ= 0.68; p < 0;001) (Fig. 6). Moreover, for stroke patients, PBF and PFF were highly correlated with Fugl Meyer Assessment Upper Extremity (respectively ρ= 0.81 and ρ= 0.79; p < 0.001). Healthy subjects generated a greater force compare to stroke patients for PFF (p < 0.001) and PBF (p = 0.007). In both groups, PBF was significantly greater than PFF (p < 0.001).

Fig.6

The relationship between isometric force measured by the REAplan^® robot and by the MicroFet 2^® dynamometer in healthy subjects and stroke patients. (A) presents the muscle force during a pulling backward contraction (PBF) and (B) presents the muscle force during a pushing forward contraction (PFF).

3.3 Isokinetic force

Figure 7 presents the isokinetic assessment results of healthy subjects and stroke patients. Isokinetic force was significantly higher for healthy subjects compare to stroke patients for PBF and BFF in each direction (p < 0.001), and movement direction influenced the isokinetic forces. Healthy subjects reached their maximal PFF and PBF when moving in front of them (0°) (p < 0.001). For the PFF, healthy subjects exerted a greater force when they moved to the assessed side (45°) than when they moved to the contralateral side (– 45°) (p < 0.001). The ratio between the force generated in front and in diagonals was analysed. For healthy subjects, the forces generated at 45° were on average 59% (PFF) and 50% (PBF) of the forces generated at 0°, and the forces generated at – 45° were on average 40% (PFF) and 49% (PBF) of the forces generated at 0°. For stroke patients, the forces generated at 45° were on average 71% (PFF) and 50% (PBF) of the forces generated at 0°, and the forces generated at – 45° were on average 51% (PFF) and 53% (PBF) of the forces generated at 0°.

Fig.7

Muscle isokinetic force generation during (A) pulling backward (PBF) and (B) pushing forward (PFF) (B) movements in three directions: in front of the patient (0°), diagonal on the side of the assessed upper limb (45°) and diagonal on the opposite side of the assessed upper limb (– 45°). The full lines represent a significant difference between directions for healthy subjects. ^**p < 0.001.

3.4 UL kinematics

Figure 8 presents the kinematic results of healthy subjects and stroke patients. Healthy subjects had significantly better kinematic indexes than stroke patients (p < 0.001). The UL kinematic were dependent on target position. Average velocity was positively and highly correlated with movement amplitude (r= 0.82–90; p < 0.001). For healthy subjects, average velocity to reach remote targets (21.8 cm/s) represented 220% of the velocity needed to reach near targets (9.9 cm/s). For stroke patients, average velocity to reach remote targets (18.0 cm/s) represented 300% of the velocity needed to reach near targets (6.0 cm/s). The same observation was made between movement straightness and amplitude. Straightness was positively and highly correlated with movement amplitude (ρ= 0.86–90; p < 0.001). For healthy subjects and stroke patients, straightness was improved by 7.5% and 17.8% respectively, when they reached remote targets compared to near targets. On the contrary, movement amplitude had no relevant influence on the smoothness indexes (ρ= – 0.46–47), and movement direction did not influence any kinematic indexes (ρ= 0.01–0.11).

Fig.8

The relationships between (A) mean velocity and movement amplitude and (B) mean movement straightness and movement amplitude. Linear regression for healthy subjects: velocity = 8.71+(0.25*movement amplitude); straightness = 0.932+(0.0012*movement amplitude) and stroke patients: velocity = 6.01+(0.20*movement amplitude); straightness = 0.839+(0.0022*movement amplitude).

4 Discussion

Serious games implemented on end-effector robots can be configured to match to individual patient’s performances. The assessed variables and reference values, presented in this study, can be included into a serious game in order to position the targets to reach, to provide suitable robotic assistance or resistance, and to determine an appropriate level of isometric force that the patient has to develop to successfully complete the task and move to the next exercise.

Many serious games start with an initial test to evaluate the patients’ UL range of motion (Burke et al., 2009; Cameirao et al., 2010). In the study of Burke et al., patients had to virtually roll eight balls with their UL from the centre of a screen in eight directions as far as possible (four cardinal directions and four intermediate directions). Data were then used to determine the position of game elements (Burke et al., 2009). In the present study, UL working area assessment task is an initial test to determine reachable areas where targets can be placed.

The distinction between active and passive WA is useful for determining the target position and configuring the activation of robotic assistance. In the present study, stroke patients were able to reach an average of 15 targets allowing the induction of game variability and adaptability. For instance, at the beginning of the patient’s rehabilitation, all targets can be placed in the active WA. Later, and according to the patients’ performance improvements, more targets can be placed in the passive WA, with for example, 50% of targets in the active WA and 50% of targets in the passive WA, creating more challenge to the patient. Finally, to provide an even greater challenge, all targets could be placed in the passive WA. Concerning robotic assistance, most articles use robot assistance based on the patients’ performances (Bishop & Stein, 2013; Conroy et al., 2011; Timmermans, Seelen, Willmann, & Kingma, 2009). Through the distinction between active and passive WA, serious games could additionally adapt the level of assistance depending on which WA the patient is working in. For instance, no assistance or a resistance could be provided in the active WA, whereas some assistance could be provided in the passive WA.

The force measured with the REAplan^® and the isometric dynamometer MicroFet 2^® was moderately to highly correlated. Participants were able to develop a higher force on the robotic device compare to the dynamometer. Indeed, the robotic device allows a global upper limb force assessment during a pushing forward and pulling backward movements. The recorded force was primarily determined by the elbow extensor or elbow flexor muscles respectively. But, muscles from the shoulder and the wrist could also influence the recorded force. So, the force represents a global upper limb force. As for the dynamometer assessment, the recorded force was an analytical force assessment of the elbow flexor or extensor muscle force. The force measured with the REAplan^® was highly correlated with Fugl Meyer Assessment Upper Extremity. Moreover, the force measured with the REAplan^® was higher for healthy subjects compare to stroke patients, and the PBF was higher than the PFF. These results support the validity of the force assessment using an end-effector robot. This assessment task can be included in a serious game in order to adapt the level of force the patient needs to generate in the game. The game can integrate scenarios involving PFF, for example, where the patient has to apply force on the end-effector inside the game to fill a bottle of water. If the patient is able to generate a high force, the game will configure a high force threshold to succeed in the exercise.

The REAplan^® measured a global UL force where shoulder and wrist muscles can be activated. This global force assessment was considered closer to real-life than the analytical measure. During some activities of daily leaving, patients generate a global reaching or pulling force. For instance, when a patient opens or closes a door, he or she activates muscles from all the UL. Therefore, it appears relevant to assess patient’s overall UL muscle forces during pushing and pulling movements in order to adapt the game difficulty during the serious game.

The results showed that isokinetic force depended on the direction of movement. A ratio between the forces generated in each direction for both groups was established. These ratios can be used by the serious game in order to estimate the isokinetic force that the patient should be able to generate in the different directions of assessment based on a single direction. These estimations will be used to configure the resistance force of the robot in subsequent tasks.

Kinematic assessment is recommended to evaluate UL after stroke (Murphy, Resteghini, Feys, & Lamers, 2015; Santisteban et al., 2016). In the present objective, it is therefore interesting to configure a serious game according to patient’s UL kinematics. Previous studies have already validated the assessment of UL kinematics in healthy subjects and brain-damaged patients with end-effector robots (Gilliaux et al., 2014; Gilliaux et al., 2016). In the present study, the effects of direction and amplitude on UL kinematics were analysed. The correlation between velocity and amplitude was high to very high. The increase of velocity between the nearest target and furthest targets were higher than the minimal detectable change fixed at 33.2% (Gilliaux et al., 2014). This means that healthy subjects and patients reached a higher velocity when the target was placed more distantly. However, velocity was not influenced by movement direction. Stewart et al. also studied the UL kinematic in stroke and healthy subjects during reaching movements to targets presented at three distances: 8, 16 and 24 cm and two directions – 45° and 45° (Stewart, Gordon, & Winstein, 2014). They observed a moderate to strong correlation in both groups between peak velocity and target distance (r= 0.75–0.78). The magnitude of the correlation did not differ between directions (Stewart et al., 2014). Steenbergen et al. studied the UL kinematics of the less affected hand of young adolescents suffering from CP and healthy subjects during a reach-and-grasp task (Steenbergen & Meulenbroek, 2006). Targets were placed at different distances: 60%, 100%, and 140% of the subject’s UL length. In both groups, the peak hand velocity increased proportionally with the target’s distance (Steenbergen & Meulenbroek, 2006). Based on these results, the robotic assistance adaptation should take into account that the reference velocity toward targets placed more distantly should be higher than toward targets placed closer (Sapin et al., 2017).

The robotic assistance is also based on movement straightness and smoothness. In this study, movement straightness was highly correlated with amplitude, but not with direction. The improvement as a function of distance for stroke patients was small, but higher than the minimal detectable change fixed at 9.9% (Gilliaux et al., 2014). Therefore, in the serious game, the expected straightness would be higher for more distant targets in order to appropriately adapt robot assistance to patients’ abilities. The results showed that the smoothness index evolved independently of the target distance and movement direction. Therefore, the smoothness index kinematic index should be the same value for all targets in the serious game.

This study has two limits. First, the maximum force recorded by the REAplan^® robot was limited to 200 N. Some healthy subjects were able to generate an isokinetic or isometric force higher than 200 N. In the study, these participants were excluded for isometric and isokinetic assessment tasks. It should not be a problem in the serious game, as no patient approached the 200 N threshold. In this study, the highest force generated by stoke patients was 187 N. Second, UL spasticity should also be assessed during the game in order to adapt it adequately to patient’s impairments. A previous study has shown that UL spasticity can be assessed using a robotic device (Dehem at al., 2017).

5 Conclusion

This study proposes a new UL assessment protocol of stroke patients using an end-effector rehabilitation robot. This protocol can be included in a serious game to evaluate performance indicators (UL working area, UL kinematics, UL muscle strength), and configure the game difficulty (target position, level of assistance/resistance, level of force) according to individual patients’ abilities.

Conflict of interest

Nothing to declare.

Footnotes

Acknowledgments

This work was supported by the Région Wallonne, the Fondation Motrice and the Fondation Saint-Luc. The authors would like to thank Julien Sapin, Martin Vanderwegen, Daniel Galinski, Ariane Ghiette and Jackie Ries for their help during the study. They would also like to thank Centre Neurologique William Lennox, Centre Hospitalier Valida and Cliniques universitaire Saint-Luc and all the subjects for their participation in this study.

References

Bishop,

, & Stein,

(2013). Three upper limb robotic devices for stroke rehabilitation A review and clinical perspective. NeuroRehabilitation, 33(1), 3–11. doi: 10.3233/NRE-130922

Borghese,

N. A.

, Pirovano,

, Lanzi,

P. L.

, Wuest,

, & de Bruin,

E. D.

(2013). Computational Intelligence and Game Design for Effective At-Home Stroke Rehabilitation. Games Health J, 2(2), 81–88. doi: 10.1089/g4h.2012.0073

Burke,

J. W.

, McNeill,

M. D. J.

, Charles,

D. K.

, Morrow,

P. J.

, Crosbie,

J. H.

, & McDonough,

S. M.

(2009). Optimising engagement for stroke rehabilitation using serious games. The Visual Computer, 25(12), 1085–1099. doi: 10.1007/s00371-009-0387-4

Cameirao,

M. S.

, Badia,

S. B.

, Oller,

E. D.

, & Verschure,

P. F.

(2010). Neurorehabilitation using the virtual reality based Rehabilitation Gaming System: Methodology, design, psychometrics, usability and validation. J Neuroeng Rehabil, 7(48), 1–14. doi: 10.1186/1743-0003-7-48.

Cameirao,

M. S.

, Faria,

A. L.

, Paulino,

, Alves,

, & Bermudez,

I. B. S.

(2016). The impact of positive, negative and neutral stimuli in a virtual reality cognitive-motor rehabilitation task: A pilot study with stroke patients. J Neuroeng Rehabil, 13(70), 1–15. doi: 10.1186/s12984-016-0175-0.

Chen,

& Michael,

(2005) Serious Games: Games That Educate, Train, and Inform.

Conroy,

S. S.

, Whitall,

, Dipietro,

, Jones-Lush,

L. M.

, Zhan,

, Finley,

M. A.

,... & Bever,

C. T.

(2011). Effect of gravity on robot-assisted motor training after chronic stroke: A randomized trial. Arch Phys Med Rehabil, 92, 1754–1761. doi: 10.1016/j.apmr.2011.06.016.

da Silva Cameirao,

, Bermudez,

I. B. S.

, Duarte,

, & Verschure,

P. F.

(2011). Virtual reality based rehabilitation speeds up functional recovery of the upper extremities after stroke: A randomized controlled pilot study in the acute phase of stroke using the rehabilitation gaming system. Restor Neurol Neurosci, 29(5), 287–298. doi: 10.3233/RNN-2011-0599

Dehem,

, Gilliaux,

, Lejeune,

, Detrembleur,

, Galinski,

, Sapin,

, Vanderwegen,

& Stoquart,

(2017) Assessment of upper limb spasticity in stroke patients using the robotic device REAplan, Journal of Rehabilitation Medicine, 49, 565–571. doi: 10.2340/16501977-2248

10.

Galinski,

(2014). Conception et optimisation d ’ un robot de rééducation neuromotrice du membre supérieur avec compensation active. PhD thesis in engineering sciences, Université catholique de Louvain, Louvain-La-Neuve (UCL-IMMC).

11.

Gilliaux,

, Lejeune,

T. M.

, Detrembleur,

, Sapin,

, Dehez,

, Selves,

, & Stoquart,

(2014). Using the robotic device REAplan as a valid, reliable, and sensitive tool to quantify upper limb impairments in stroke patients. J Rehabil Med, 46, 117–125. doi: 10.2340/16501977-1245.

12.

Gilliaux,

, Lejeune,

T. M.

, Sapin,

, Dehez,

, Stoquart,

, & Detrembleur,

(2016). Age Effects on Upper Limb Kinematics Assessed by the REAplan Robot in Healthy Subjects Aged 3 to 93 Years. Ann Biomed Eng, 44(4), 1224–1233. doi: 10.1007/s10439-015-1396-2.

13.

Golubović,

, & Slavković,

(2014). Manual ability and manual dexterity in children with cerebral palsy. Hippokratia, 18(4), 310–314.

14.

Heins,

, Dehem,

, Montedoro,

, Stoquart,

, Edwards,

, Dehez,

,... & Lejeune,

(2017). Robotic-assisted serious game for motor and cognitive post-stroke rehabilitation. IEEE 5th InterNatl Conf Serious Game Appl Heal, Perth, 1–8.

15.

Hou,

W. H.

, Shih,

C. L.

, Chou,

Y. T.

, Sheu,

C. F.

, Lin,

J. H.

, Wu,

H. C.

,... & Hsieh,

C. L.

(2012). Development of a computerized adaptive testing system of the Fugl-Meyer motor scale in stroke patients. Arch Phys Med Rehabil, 93(6), 1014–1020. doi: 10.1016/j.apmr.2011.12.005.

16.

Jaillard,

, Naegele,

, Trabucco-Miguel,

, LeBas,

J. F.

, & Hommel,

(2009). Hidden dysfunctioning in subacute stroke. Stroke, 40(7), 2473–2479. doi: 10.1161/STROKEAHA.108.541144.

17.

Katz,

, Ring,

, Naveh,

, Kizony,

, Feintuch,

, & Weiss,

P. L.

(2005). Interactive virtual environment training for safe street crossing of right hemisphere stroke patients with unilateral spatial neglect. Disabil Rehabil, 27(20), 1235–1243. doi: 10.1080/09638280500076079.

18.

Langhorne,

, Bernhardt,

, & Kwakkel,

(2011). Stroke rehabilitation. The Lancet, 377(9778), 1693–1702. doi: 10.1016/s0140-6736(11)60325-5

19.

Mackay,

, & Mensah,

(2004). The atlas of heart disease and stroke. World Health Organisation, 1–112.

20.

Mehrholz,

, Pohl,

, Platz,

, Kugler,

, & Elsner,

(2015). Electromechanical and robot-assisted arm training for improving activities of daily living, arm function, and arm muscle strength after stroke. Cochrane Database Syst Rev (11), CD006876. doi: 10.1002/14651858.CD006876.pub4

21.

Murphy,

A. M.

, Resteghini,

, Feys,

, & Lamers,

(2015). An overview of systematic reviews on upper extremity outcome measures after stroke. BMC Neurol, 15(29), 1–15. doi: 10.1186/s12883-015-0292-6.

22.

Norouzi-Gheidari,

, Archambault,

P. S.

, & Fung,

(2012). Effects of robot-assisted therapy on stroke rehabilitation in upper limbs: Systematic review and meta-analysis of the literature. The Journal of Rehabilitation Research and Development, 49(4), 479. doi: 10.1682/jrrd.2010.10.0210.

23.

Perry,

J. C.

, Andureu,

, Cavallaro,

F. I.

, Veneman,

, Carmien,

, & Keller,

(2011). Effective game use in neurorehabilitation: User- centered perspectives. Handb Res Improv Learn Motiv through Educ Games Multidiscip Approaches, 1–44.

24.

Pignolo,

(2009). Robotics in neuro-rehabilitation. J Rehabil Med, 41, 955–960.

25.

Pollock,

, Farmer,

S. E.

, Brady,

M. C.

, Langhorne,

, Mead,

G. E.

, Mehrholz,

, & van Wijck,

(2014). Interventions for improving upper limb function after stroke. Cochrane Database Syst Rev (11), CD010820. doi: 10.1002/14651858.CD010820.pub2.

26.

Santisteban,

, Teremetz,

, Bleton,

J. P.

, Baron,

J. C.

, Maier,

M. A.

, & Lindberg,

P. G.

(2016). Upper Limb Outcome Measures Used in Stroke Rehabilitation Studies: A Systematic Literature Review. PLoS One, 11(5), e0154792. doi: 10.1371/journal.pone.0154792.

27.

Sapin,

, Dehez,

, & Gilliaux,

(2017). Rehabilitation system and method. Patient No. PCT/EP2016/072676, 1–26.

28.

Souza,

L. A.

, Martins,

J. C.

, Teixeira-Salmela,

L. F.

, Lara,

E. M.

, Moura,

J. B.

, Aguiar,

L. T.

, & de Morais Faria,

C. D.

(2014). Validity and reliability of the modified sphygmomanometer test to assess strength of the lower limbs and trunk muscles after stroke. J Rehabil Med, 46(7), 620–628. doi: 10.2340/16501977-1823.

29.

Steenbergen,

, & Meulenbroek,

R. G.

(2006). Deviations in upper-limb function of the less-affected side in congenital hemiparesis. Neuropsychologia, 44(12), 2296–2307. doi: 10.1016/j.neuropsychologia.2006.05.016.

30.

Stewart,

J. C.

, Gordon,

, & Winstein,

C. J.

(2014). Control of reach extent with the paretic and nonparetic arms after unilateral sensorimotor stroke II: Planning and adjustments to control movement distance. Exp Brain Res, 232(11), 3431–3443. doi: 10.1007/s00221-014-4025-7.

31.

Timmermans,

A. A.

, Seelen,

H. A.

, Willmann,

R. D.

, & Kingma,

(2009). Technology-assisted training of arm-hand skills in stroke: Concepts on reacquisition of motor control and therapist guidelines for rehabilitation technology design. J Neuroeng Rehabil, 6, 1. doi: 10.1186/1743-0003-6-1.

32.

Winstein,

C. J.

, Stein,

, Arena,

, Bates,

, Cherney,

L. R.

, Cramer,

S. C.

,... & Outcomes,

(2016). Guidelines for Adult Stroke Rehabilitation and Recovery: A Guideline for Healthcare Professionals From the American Heart Association/American Stroke Association. Stroke, 47(6), e98–e169. doi: 10.1161/STR.0000000000000098.

	Healthy subjects	Stroke patient
	n = 32	n = 20
	Mean (SD)	Mean (SD)
Working area
Active working area, %	71.0 (9.05)	51.7 (16.54)
Passive working area, %	/	58.2 (12.35)
Relative working area, %	/	88.8 (20.50)
Isometric force
Pulling backward/flexor muscles
Microfet 2^®, N	150.6 (36.95)	77.8 (50.06)
REAplan^®, N	152.5 (29.65)	94.8 (59.52)
Pushing forward/extensor muscles
Microfet 2^®, N	110.2 (30.50)	73.5 (51.31)
REAplan^®, N	138.6 (34.38)	83.4 (52.68)
Isokinetic force
Pulling backward
– 45°, N	70.8 (28.73)	33.8 (22.01)
0°, N	159.7 (24.07)	64.1 (56.42)
45°, N	80.4 (42.95)	32.0 (23.95)
Pushing forward
– 45°, N	60.7 (26.57)	31.7 (19.64)
0°, N	151.4 (29.38)	61.9 (57.87)
45°, N	90.0 (36.26)	44.1 (29.16)
Kinematics
Straightness	0.96 (0.02)	0.93 (0.03)
Velocity, cm/s	17.2 (3.08)	12.5 (2.75)
Smoothness	0.57 (0.03)	0.48 (0.03)