User

This study had three groups of users: eleven patients (“RobExReha-Patients”) (aged 64.4 ± 11.2 years, range 47–85) and five therapists (“RobExReha-Therapists”) (aged 38.2 ± 16.0 years, range 23–57) who evaluated the RobExReha device and an additional eleven age-matched patients (“Reference Group”) (aged 64.3 ± 9.1 years, range 49–79) were allocated to the reference group (see Table 1). The RobExReha-Patients group received, and the RobExReha-Therapists group administered training sessions using the RobExReha device; the Reference Group trained with state-of-the-art commercially available and established devices (ArmeoSpring and ArmeoPower, both Hocoma AG, CH). Both devices are typically used for arm and hand therapy in neurological or orthopedic rehabilitation and have proven to be effective [20,21,22]. Moreover, as established devices, they provide a good usability [23] and allow therapists to supervise more than one patient at a time [24]. They both combine a mechanical support of the upper extremity (either passive by a spring-loaded system (ArmeoSpring) or actuated (ArmeoPower)) with a variety of serious games presented on a standard monitor. These devices will not be further described hereinafter as they are widely distributed and known.

Inclusion criteria for patients were a subacute or chronic paresis of the upper limb due to neurological disorders, preserved language comprehension, ability to communicate, orientation in space and time, ability to sit upright for at least 45 min, and the completion of at least four training sessions with the respective therapy device (RobExReha or Armeo). Further device-related inclusion criteria for the RobExReha-Patients group were ability to reach the clip-in-position of the robotic device at 30° abduction of the shoulder, impairment on the right side, and normal or appropriately corrected vision.

We excluded patients with craniotomy, instable fractures, fixated contractions, active implants, epilepsy and severe instabilities, severe spasticity (Modified Ashworth Scale > 3, [25]) or open skin defects of the affected upper limb. Further, patients were screened by the treating neurologist for severe neuropsychological problems such as apraxia, severe neglect, severe aphasia or dementia.

Stereoscopic vision was evaluated using the Titmus-Test (Stereo Optical Co., Chicago, IL) but was not a limiting factor for study inclusion. The mean stereoscopic vision of the RobExReha-Patients group was 405 ± 983 arc sec (median: 100 arc sec; range 40–3352 arc sec). With defining the cut-off score for unimpaired stereoscopic vision at 60 arc sec [18], seven patients showed impairments in stereoscopic vision and four patients were able to see stereoscopically well. All except one patient were able to see better or equivalent to 400 arc seconds.

The five therapists administering the therapy had a working experience of 9.8 ± 8.2 years in neurorehabilitation (min-max: 1–23 y.). They were either physiotherapists (n = 2), a sport therapist (n = 1), a health scientist (n = 1) or a specially trained staff member for robotic upper limb therapy without specific health care profession (n = 1). Two of the therapists were highly trained using other robotic devices daily, and three of them had not regularly used robotic therapy. The technical affinity questionnaire (TA-EG) [26] was administered to describe their technical affinity. This questionnaire evaluates the positive and negative attitudes as well as competence and enthusiasm towards technology. The items are answered on a five-point Likert-Scale. The result is the mean value of the items with 1 indicating a low, and 5 indicating a high technical affinity. The therapists had an overall medium to high technical affinity, with an average of 3.4 (min: 3.1 max: 4.1) points (out of 5) in the TA-EG.

Table 1 Characteristics of participants

All participants gave informed written consent. Ethics approval was obtained (Ethikkommission der Bayerischen Landesärztekammer) and the study was registered in the German Clinical Trials Register (DRKS00022136).

Technology

The RobExReha robotic system should meet the demand to provide (adaptive) support to the patient, while ensuring a high level of safety. Therefore, the commercially available LBR iiwa robotic arm (KUKA AG, DE) was chosen, which is designed for safe human-robot interactions. The robotic arm was mounted on a cart in a 45° angle and interfaced with the patients’ right arm at the robot flange (see Fig. 1). The robotic setup was further interfaced with a custom-made Unity application (Unity Technologies, US) for the HoloLens (Microsoft Inc., US) that enabled patients to perform active serious gaming with robotic support for the impaired arm (see Fig. 2). The Microsoft HoloLens (Microsoft Corp., US) was chosen as the AR device, as it had already been implemented in various contexts (e.g., surgical aids, medical education, industrial engineering) [27], including in stroke [28]. The different components were linked within a separate network via ethernet cables and the HoloLens was wirelessly connected. Robot information including position of flange and axes as well as active forces for data analyses were saved with a sampling rate of 25 Hz.

Fig. 1

Left: subject using the RobExReha system with the HoloLens for the gaming application. Right: arm-robot interface used by a patient during the therapy with the RobExReha system

Fig. 2

Arm-robot interface of the RobExReha system: Picture A and B show the elbow brace, picture C shows the counterpart on the robotic flange, which consisted of two carbon fibre plates with three electromagnets attached on each. The two plates were connected via a joint at the robot flange. The plate supporting the upper arm was rotatable, while the second plate for the forearm was fixed on the flange and could therefore be controlled by the robotic movement. Picture D shows the hand module

Arm-robot interface

To connect the patient’s arm to the robot flange, a mechanical human-machine interface comprising a brace on the upper and lower arm and a counterpart on the robotic flange was developed. The brace consisted of two 3D printed shells (Fig. 2), which were available in two different sizes. The upper and lower arm shells were connected via a ferromagnetic steel plate, that featured a joint at the height of the elbow (see Fig. 2). The brace was donned on the patient’s arm by a therapist and fixed with Velcro straps.

Subsequently, the steel plate of the brace (Fig. 2A/B) could be attached magnetically to the carbon plates fixed at the flange (Fig. 2C). It was thereby ensured by a bolt and stop to prevent the angle of the elbow from exceeding 180°. The electromagnets for the connection of the brace had a nominal force of about 210 N. This was strong enough to hold the brace with the arm in position but weak enough to release it quickly by a therapist in case of emergency.

A hand module with a grip hold was attached to the carbon plate that guided the lower arm (see Fig. 2D). The position of the hand module could be easily adjusted to the length of the lower arm with a screwed connection.

Calibration procedure

To calibrate the position of the robot, the shoulder height and upper arm length were manually measured using a measuring tape in the beginning of the first session and entered in the software. At the beginning of each session, the robot moved the flange to the respective clip-in position at 30° shoulder abduction. Once the brace was clipped to the arm, the arm was weighed by the robot to enable the calibration of robotic support. Afterwards, the patient’s specific range of motion (ROM) was evaluated, which consisted of two assessments: the active ROM, in which the patient could actively move the impaired arm, and the passive ROM without pain as guided and determined by the therapist. The passive ROM typically exceeded the active ROM in patients. This differentiation enabled the implementation of two different robotic support modes: (A) The support level was set to a constant level for both ranges, or (B) the support level within the active ROM was set to a constant level while the support of the movement beyond the active ROM increased dynamically until the patient achieved the desired movement. The robot enabled six support levels that were adjustable with increments of 20%, from 0 to 100%.

Safety features

To ensure the patients’ safety, the robot’s movements were restricted to the respective patient’s passive ROM. An additional safety feature was an automated stop in case the active sum force of all axes exceeded 30 Nm. Moreover, a safety stop button for the supervising therapist was implemented.

To give feedback about compensatory movements of the trunk or shoulder girdle, an alert was implemented to notify both therapists and patients if the patient’s shoulder position (i.e., the acromion) left the anticipated position. In case this position deviated more than a predefined threshold from the initial shoulder position, a visual alert was displayed in the gaming environment. Initially, the threshold of this shoulder compensation alert was set to a range of 5 cm around the ideal shoulder position. However, the alert tended to be displayed too early and too frequent and thus interrupted the gaming experience. After increasing the threshold to 7.5 cm, the alert only appeared in cases where a relevant compensatory movement was visually evident for the supervising therapist. The patients reacted adequately by actively correcting the position of their shoulder.

Gaming scenario

The training setup included an AR-game for HoloLens. The training game was based on a 3D puzzle (see Fig. 3): A building (e.g., tower, house) and a blue hand appeared within the field of view of the patient. Subsequently, the building fell into pieces, with its contours remaining. The pieces were spread in front of the building within the patient’s specific active ROM. The designated target positions for the rebuilding process of the building were within the range of the patient’s passive ROM. The blue hand served as controller and it was moved by the position of the real hand. Once the virtual hand approached a puzzle piece, the virtual fingers grabbed the piece. Next, the respective target position of the grabbed piece was indicated and the piece could be moved to the target position, automatically being released by the hand.

Fig. 3

View through the HoloLens with the gaming scenario: a blue hand (yellow arrows) could select puzzles pieces. Once a piece was selected, the hand closed (see right figure) and the piece could be moved to the building

The difficulty of the game could be adjusted via the distance and size of the building with respect to the hand position, which led to a smaller/greater required travel distance of the patient’s hand to reach the designated spot. This, along with the robotic support level, could be also adjusted while gaming.

Study protocol

To evaluate the usability and feasibility, the groups used the devices (either directly as a patient or administering the therapy as therapist). After at least four sessions, their experience was evaluated using questionnaires (see Table 2). Additionally, during the training with the RobExReha device, all technical difficulties, user errors or safety aspects were documented (Fig. 4). Ten out of eleven patients of the RobExReha-Patients group additionally completed an extra (fifth) therapy session which was recorded using a video protocol to investigate the donning and set-up time of the RobExReha device.

Fig. 4

Schematic of the structure of this study: the RobExReha-Patients trained 4–5 times with the RobExReha device, while technical and user incidents were reported. After completion of at least four training sessions, the questionnaire evaluation was conducted. The Reference Group only participated in the questionnaire survey and reported their perception of the conventional robotic gaming therapy

User activity RobExReha-patients group

Patients in the RobExReha-Patients group received training with the RobExReha device. A session consisted of the donning and setup of the device, subsequent gaming therapy, termination of the session and doffing of the device. Amount and mode of robotic support as well as the length of the gaming sessions were adjusted by the therapist according to the patient’s needs. Figure 5 shows the User Interface to configure these parameters.

Fig. 5

User Interface of the RobExReha system for the therapist. In this screen, the therapy session could be planned and adapted: Separate gaming sessions could be added and adjusted in length (german description: “Dauer”), as well as the rest length between the sessions (“Pausezeit zwischen den Spielen”). Additionally, the requirements in terms of range of motion (“Beweglichkeit”) and the support by the robot (“Kraft”) could be adjusted via this interface by the therapist. Under the “Calc-Mode” the paradigm of robotic support (adaptive/non-adaptive) could be chosen;

RobExReha-therapists group

The therapists administered the RobExReha device in at least five sessions to a patient before answering the questionnaire. They were responsible for donning, doffing, and configuring the training session as well as supervising the patient.

Reference group

The patients in the Reference Group received rehabilitation training with the commercially available ArmeoPower or ArmeoSpring (Hocoma AG, Switzerland) as part of their inpatient schedule. Six patients in the Reference Group trained with the ArmeoPower device and five with the ArmeoSpring device, with a mean of 15 (± 25, min: 4, max: 90) completed therapy sessions. They received no further intervention regarding this study. This data acquisition was done during the developmental stage of the RobExReha device and thus before the data acquisition of the RobExReha device.

Questionnaires Table 2 Overview over the standardized questionnaires used for the usability evaluation Patients’ questionnaire

Both patient groups filled out the Quebec User Evaluation of Satisfaction with assistive Technology (QUEST) [29] and the Raw Task load index (RTLX) [30]. The QUEST was designed to measure the level of satisfaction attributable to assistive technologies. We used the device subscale score (8 items), which was scored in terms of perceived satisfaction from 0 to 5 (5 ~ highly satisfied) and had been previously applied to measure satisfaction with rehabilitation and assistive robotic devices [31]. We used the QUEST without individual weighting of results and item four was changed to “reliability and safety” in deviation from the original scale. The item “how satisfied are you with the durability of the device?” was excluded, as no meaningful evaluation after the intervention period of four to five sessions was possible. The RTLX evaluates the workload using six subjective subscales: mental, physical and temporal demand, performance, effort and frustration. Each item is rated within a 100-points range. The mean value is calculated to receive the overall task load index. We used the raw version without individual weighting of these subscales. A value of 100 suggests very high task demands or very high failure in terms of performance [32].

As the RobExReha-Patients group used the HoloLens for the gaming environment, they additionally filled out the “presence in augmented reality” (pAR) questionnaire (adapted from [33]). The pAR questionnaire evaluates the patients’ impression of the output of the HoloLens. The questionnaire was adapted by turning questions into statements that could be answered with a 5-point Likert scale (5 ~ totally agree).

Therapists’ questionnaire

The therapists completed a questionnaire consisting of the same adapted device subscale of the QUEST as the patients, the system usability scale (SUS) [34] and the short version of the User Experience Questionnaire (UEQ-short) [35] to express their perception of the device’s usability.

Statistical methods

The results were evaluated descriptively and the results of QUEST and RTLX from the RobExReha-Patients and the Reference Group were compared using Mann-Whitney U-test. Due to the exploratory character of the study, no correction for multiple comparisons was done. Analyses were done in Matlab (The MathWorks Inc., US) and SPSS (IBM, US). The alpha level was set to 0.05 for all analyses. Data is presented as either the mean (± standard deviation) or the median (min-max).