The participants were recruited through medical chart review and regular visits between April and May 2017. Each patient in this study received standard neurorehabilitation care during their hospitalization. Patients were selected according to the following criteria: inclusion criteria were: (1) written informed consent, (2) aged above 18 years, (3) a first-ever stroke in the past 24 h until 3 months, reflecting the early rehabilitation phase (5) (4) experience upper limb impairments due to stroke (Nine Hole Peg Test > 19 s with at least one side) (28), (5) at least able to perform ante-flexion with their upper arm and extent one or more fingers against gravity (3 ≤ Medical Research Council scale < 5) (29), and (6) the participants had to be able to understand the instructions and assessments in German. Exclusion criteria were: (1) severe cognitive impairments (Montreal Cognitive assessment: MoCa < 14) (30), (2) severe apraxia (Apraxia Screen of TULIA < 9) (31), (3) aphasia (Language screening test < 15) (32), (4) severe self-reported pain, or (5) other severe orthopedic problems of the upper limb impairing participation.

The intervention consisted of nine training sessions of 30 min, spread out over a period of 3 weeks, this means three training sessions per week. An intake session was planned with eligible patients. Upon consent, the baseline measures of dexterity, grip strength, and motor impairment of the upper limb were collected. This was, if possible, followed by the first training and evaluation of the usability of the system. The level of active participation and time spent on the intervention were measured during each training session. The second, fourth, fifth, seventh, and eighth session contained only training. In the third, sixth, and ninth session, dexterity and grip strength were measured followed by the training and evaluation of the usability. The reassessment of the motor impairment of the upper limb and evaluation of the open-end questions were at the ninth session. The present study aimed to follow the procedure schedule as shown in Table 1.

Each training session was performed on a desktop computer at the occupational ward of the neurorehabilitation. The participants (if not in a wheelchair) were sitting on a chair with a rectangular pillow on their lap, so that the elbows could rest on the pillow (see Figure 1).

The LMC was placed on a table in front of the participant between the body and the computer screen. During each session, the principal investigator sat next to the participants, providing (if needed) online feedback via verbal, visual, and/or physical instructions.

The LMC incorporates three infrared emitters and two charge-coupled device cameras for capturing the motion of both hands, wrists, and forearms (15, 26). The light of the infrared emitters reflects back from the surfaces of the hands, so no markers are needed (33). Weichert et al. (34) reported that the LMC has an accuracy of 0.2 and 1.2 mm in a static and dynamic setup, respectively.

In February and March 2017, all free access games in the Leap Motion App Store© (https://apps.leapmotion.com/?sign_up=true) were evaluated to determine if the games contained key components of dexterity movements: alternating finger, pincer grasp, fine pointing, and palmar extension/flexion. Five games were selected: Dots Trial, Cut the Rope, Playground, and American Sign Language Digits. Playground contains two games named Blocks and Flower. See Figure 2 for an impression of the games.

Each therapy session contained about 6 min per game, and the sequence of the five games was randomized by the function “randperm” using MATLAB R2013b (Mathworks, Natick, MA, USA). Each game is played with both hands starting with the non-affected hand.

A HP EliteDesk 800 (Intel Core i5) desktop computer was used, with a 19-inch computer screen with an aspect ratio 4:3 (1,280 × 768).

All outcome measurements were collected and/or rated by a single trained assessor (SF) to optimize standardization of outcome measurement.

Descriptive statistics were used to calculate the group mean (M), SD, and SEM of clinical and demographic variables.

The distributions of the outcome measures (SUS, PRPS, NHPT, DextQ-24, Jamar, and FM-UE) were examined for normality of distribution to select either the parametric one-way repeated measure ANOVA or the non-parametric Friedman’s ANOVA. Wilcoxon Signed Rank Test was used in addition to the non-parametric Friedman’s ANOVA. The compliance rate is calculated by: compliance rate = TT/PT × 100%. A compliance rate of 80% was defined as good (44). The level of active participation was evaluated by the principal investigator per game. The PRPS scores per training was the average PRPS score of all the games played in that session. The answers of the open-end questions 1, 2, 3, and 6 were grouped in positive, neutral, and negative answers, so that the answers could be organized and analyzed. For the remaining open-end questions, a descriptive summary of the answers is given.

Statistical analyses were performed using the SPSS statistical software system (IBM SPSS Statistics for Windows, Version 22.0. Armonk, NY, USA: IBM Corp.), and a confidence level of 95% was used, so that level of significance was set at p = 0.05, two-tailed.

During the recruitment period, 64 stroke patients admitted to the neurorehabilitation ward were screened for eligibility. Fifteen patients who were potentially eligible were selected. Two patients did not accept informed consent and were excluded. Thirteen patients started the 3 weeks LMC dexterity training program. Clinical and demographic characteristics are presented in Table 2.

Eight of the thirteen patients could complete the whole training program. One participant experienced a new stroke between the seventh and eighth session, and therefore, follow-up measures were excluded from the main analysis. Three other patients were discharged earlier. And finally, one older patient, a 62-year old male, stopped the intervention during the second training session due to lack of motivation. Importantly, there were no severe intervention-related adverse events like severe shoulder pain or severe fatigue.

Overall analyses revealed few missing data in the outcome measures: SUS (3.1%) and DextQ-24 (0.65%). Average imputation was used to correct these gaps. Furthermore, one participant was not able to perform the NHPT with his affected hand and is not included in the NHPT analysis. The other outcome measures had no missing data.

Compliance rate was high (87%) and while the only previous LMC study did not report data on compliance rates (15), we consider this high feasibility of VBT with LMC. Active participation level scores were in line with Iosa et al. (15). The slight decrease in PRPS scores after the seventh training session could result from lack of progression in game level difficulty, as all attainable levels were attained after 2 weeks. Such progression in difficulty next to variance and challenge in games is needed to increase the efficacy and compliance. Another reason could be that the LMC is at some point perhaps not sensitive enough anymore to the quality of movement and misses crucial information on upper limb motor recovery (48). For example, the LMC registers the movements as “performed well,” while the real-world movements are still not optimal or contain compensation strategies (49). This could affect the level of challenge in the VBT. Nevertheless, the level of active participation remained very good throughout the training program, but it should be kept in mind that the VBT with LMC was performed in the presence of a therapist during hospitalization and therefore possibly attributing to a high participation level of the patients.

Other studies have examined VBT delivered in the home environment of chronic stroke patients (50, 51). Piron et al. used a tele-rehabilitation approach using two computers, a camera, and a magnetic transmitter/receiver to deliver remote upper limb training in chronic stroke patients with favorable results. All participants completed the intervention, but exact compliance rates were not reported (50). Standen et al. implemented upper limb training with a virtual glove with LEDs and an infrared camera, which proved feasible. However, considerable variation was found both in terms of duration of use (1.46–70% of the recommended duration) and the number of days used (10–100%) (51). In our study, we found high compliance rates and good participation, showing that the LMC system, relatively cheap and simple compared to these systems, is feasible already in the early rehabilitation phase in an in-patient setting. However, home-based implementation of the LMC in the early rehabilitation phase needs further study to determine if compliance rate and active participation would also be acceptable.

From the viewpoint of the therapist, we found that participants were not always completely capable to perform the VBT independently. Visual, verbal, and physical instructions were sometimes needed to support participants. Most therapist instructions were related to depth perception in the VR environment. This is indeed already described previously that stroke patients may experience difficulties in perceiving objects in a 3D environment (52). The gameplay instructions were not embedded in the games and personal progress was not recorded in the software. This prompted the need for a therapist to switch between games, record the personal progress, and intervene when there were technical problems with the LMC. Proposed improvements from patient-reported suggestions are in line with these limitations.

Although this pilot study was not designed and powered to evaluate efficacy, we found improvements in the objective and subjective dexterity outcomes (31.5% in NHPT, 16.3% in DextQ-24. resp.). This is in line with previous studies (17, 50, 51). The improvements in dexterity may result from the training being intensive, highly repetitive, and task-specific (5). We also found significant improvements (11.3%) in grip strength. There is no resistance of real-world objects involved in the LMC training so this is somewhat unexpected and may be due to spontaneous neurological recovery in the early rehabilitation phase (1, 7, 49). Iosa et al. (15) also found improvements in grasp force after VBT with LMC. Possibly VBT could improve grip strength through the increased number of repetitions of pinch and grip movements but this needs further investigation.

The neurological recovery of motor function of the upper limb was measured with the FM-UE, for which we did not find significant changes. The FM-UE is however a general evaluation of pathological synergies in the upper extremity and not a specific dexterity measure and may fail to detect small improvements in fine manual dexterity and grip strength. In addition, the FM-UE may not be sensitive to the quality of the movement (e.g., if a patient was able to completely perform a movement-item at baseline and performed the same movement-item much smoother at ninth session, it was both scored with a 2, “can be performed”). Furthermore, the baseline scores of our participants were already quite high, leaving less room for improvement.

The present study was subject to limitations, such as the pre-experimental, this means one group pre-test/post-test, design, small sample size, and the limited duration of the intervention. In addition, although the DextQ-24 was validated in patients with Parkinson’s disease (38), it was not formally validated in sub-acute stroke patients.