Outpatients with a chronic (time since injury >1 year) iSCI (age 41–74 years) from the University Hospital Balgrist (Zurich, Switzerland) were included in the study by a medical referral between 2013 and 2015, and the sample size was restricted by this time period (Table 1). Inclusion criteria were as follows: subjects with a chronic iSCI and a lesion level below C4, no assistive and supporting systems needed to sit in a chair, and American Spinal Injury Association Impairment Scale (AIS) C or D (i.e., sensorimotor incomplete) at time of inclusion (C = more than half of the key muscles below the neurological level have a muscle grade less than 3; D = at least half of the key muscles below neurological level have muscle grade ≥3) (17).

Exclusion criteria were as follows: head injuries leading to cognitive or visual impairment, spasticity limiting lower limb movements, medication influencing therapy, and psychiatric limitation, depressive symptoms (i.e., score >14, Beck Depression Inventory) (18) or other neurological disorders.

The purpose of the study was clear to all subjects, and written informed consent was acquired. The experimental protocol was in accordance with the Declaration of Helsinki and approved by the Canton of Zurich ethics committee (EK-24/2009, PB2016-00545).

Home-based VR-augmented training was done with the mobile prototype of the YouKicker system (YouRehab AG, Schlieren, Switzerland) for the lower limbs, which is the successor of the stationary therapy system YouKicker described previously in a clinic-based study (standard one-on-one therapist-patient sessions at the University Hospital Balgrist) (7, 19). The home-based VR-augmented therapy system presented virtual representations of the feet and legs in a first-person perspective on a laptop computer and trained the lower limbs by combining action observation and execution (Figure 1). The system used four wireless 3 degree of freedom accelerometer sensor nodes that were attached bilaterally to the dorsum of the feet and the tibia, measuring inclination angles and transmitting movements to a laptop for processing and display on the screen. The foot sensor nodes were fastened on the foot with Velcro fixed to modified elastic rubber overshoes. The overshoes were worn over the patient’s own shoes. The tibia sensor nodes were fastened on the lower leg with Velcro and adjustable straps. Sensor data were sent via Bluetooth to a laptop computer using the Unity 3D game engine (Unity Technologies, San Francisco, CA, USA).

In collaboration with therapists, clinically relevant exercises for foot and leg while seated or in a standing position were generated. Five VR scenarios were provided on the training system to train foot and leg movements in sitting and standing positions. The system contained applications to train different isolated movements as follows.

Ankle dorsal flexion (three scenarios): Footbag (sitting/standing): the subject juggles a ball between the left and right foot. An ankle dorsal flexion (approx. 35/min per leg depending on patient ability) is performed preventing foot drag. The path of the ball is pre-set. Hamster Splash (sitting): hamsters rush to the subject’s feet. With a dorsal flexion of the ankle (approx. 15/min per leg) the subject launches each hamster into a swimming pool. Higher scores are achieved when faster ankle movement are performed. This also leads to more elaborate hamster movements (somersaults, swimming patterns). Get to the Game—Activity of Daily Living (sitting/standing): a virtual person (avatar), representing the subject, walks from home via a tram station to a stadium, as fast as possible. To make the avatar move, the subject has to alternately lift his/her feet. By increasing the range of motion of the ankle and by faster ankle movements, the subject can achieve a better score due to faster and bigger footsteps.

Knee extension (one scenario): Star Kick (sitting/standing): knee extensions are performed by the subjects (approx. 12/min per leg) hitting a ball toward presented stars with foot. For every kick, the subject scores points. With this exercise, the subject trains knee extension not using direct measurement of the knee angle.

Leg ad-/abduction (one scenario): Planet Drive (sitting): Cars are driving on a roadway toward virtual feet. The lower legs are tilted sideways (approx. 4/min per leg) to avoid touching the cars.

All subjects were trained at home on the VR tasks over a period of 4 weeks, with 16–20 sessions of 30–45 min each. The following information was collected at the beginning of the study: age, gender, height, weight, etiology, and level of lesion, AIS, time since injury, medication, pain presence, and intensity. In addition, the motivation was scored by the subjects on an 11-point Numeric Rating Scale (NRS) from 0 (i.e., worst) to 10 (i.e., best) after the completion of each training session.

The training system was set up by a therapist at the subject’s home. The subject completed the first training under the supervision of the therapist and on a weekly basis the therapist visited the subject to check the training and collect data. During a typical training session, each subject completed either one block (Footbag: 3 × 2 min, Planet Drive: 2 × 2.5 min, Star Kick: 2 × 2 min, Hamster Splash: 3 × 2 min) of training and a Get to the Game session (approx. 5 min), or twice the block without a Get to the Game session. The tasks were always presented in the same order during a session using the muscle groups alternately to prevent fatigue. Around 500 repetitions of ankle movements (Footbag, Hamster Splash, and Get to the Game) and 100 knee movements (Star Kick) with each leg are performed by a typical patient during a training session. When a subject came either to the same and/or higher number of repetitions in three successive sessions (i.e., increase in difficulty) or the number of repetitions was less in three successive sessions (i.e., decrease in difficulty) during the week the therapist changed the difficulty level of the task. A higher difficulty level resulted in higher speed or number (i.e., in more repetitions) of the specific target movements.

All outcome measures were tested at the University Hospital Balgrist. For outcome stability for the subjects with iSCI before the intervention, two different assessment time points were chosen before treatment: 4 weeks before treatment program, i.e., pre-baseline assessments, and directly before treatment program, i.e., baseline assessments. After finishing of the training program, post-assessments were performed, and 2–3 months after treatment program, follow-up assessments were performed. The results at the four different assessment time points are presented in Table 2.

Outcome measures were included to test muscle strength, balance, and mobility, and for each field, the most conclusive outcome measure was chosen as primary outcome before testing: Lower Extremity Motor Score (LEMS) for lower limb muscle strength, Berg Balance Scale (BBS) for balance during functional activities, and the outcome Timed Up and Go (TUG), which covers functional mobility. This test correlates with performance in other mobility/walking outcomes, e.g., 10 m walking test (10MWT) and Walking Index for Spinal Cord Injury (WISCI) II, in subjects with iSCI (20). All other outcome measures were chosen as secondary outcome measures: 10MWT, 6 min walking test (6minWT), Spinal Cord Independence Measure (SCIM) III mobility, and WISCI II.

The minimal clinically important differences (MCIDs) are listed in the following section, if available for SCI. MCID is the minimal change of an assessment considering the clinical importance for the subject or clinician.

SPSS 23 (IBM, New York, NY, USA) was used for statistical analysis. First, pair-wise comparisons were performed between the pre-baseline and baseline assessments with the Wilcoxon signed-rank test. Second, the outcome measures of the averaged pre-baseline and baseline time points were compared to post-assessment and follow-up assessment time points with the Wilcoxon signed-rank test. Third, a correction for multiple comparisons was calculated with the Bonferroni correction and a significance level of P ≤ 0.017 was set.

The clinical characteristics of the subjects with iSCI (mean age 60 ± SD 10.2 years) are presented in Table 1. Two of the subjects were AIS C and 10 AIS D. Because of medical issues unrelated to the study, subject P3 withdrew after agreeing to participate in the study. None of the subjects had any pain while playing the games or after the sessions.

The motivation after training was 8.6 ± 1.3. The most attractive games as voted by the patients were the following: Hamster Splash (number of subjects: 5), Star Kick and Planet Drive (2 each), and Footbag and Get to the Game (1 each). The most difficult and challenging game for all patients was “Planet Drive.” The subjects played mainly in a sitting position (chair or wheelchair) but changed to a standing position for the last repetitions of Footbag, Star Kick, and Get to the Game.

All subjects familiarized themselves with the VR system quite quickly. Overall, the subjects evaluated the system very positively. It was reported to be user-friendly and they liked the visual features. Some of the subjects could also imagine using the system for longer than the required 4 weeks and reported a positive effect such as being able to lift their feet better, improved sensation in the legs/feet, and improved swimming and walking.

Between the time points pre-baseline and baseline, the outcome measures did not differ, which is an indication for outcome stability for the included subjects before the intervention. The values of the outcome measures are presented in Table 2 whereby one subject was unable to perform the walking assessments (P11).

At post-assessment, significant increases (P ≤ 0.017) in comparison with the averaged pre-baseline and baseline were found in the primary outcome measures such as muscle strength (LEMS, P = 0.008), balance (BBS, P = 0.008), and functional mobility (TUG, P = 0.005). In addition, 7 out of 11 subjects improved (i.e., an increase of one grade at least) in ankle dorsiflexion (L4) and four of them reached the MCID of LEMS (overall) at post-assessment. The secondary outcome measures showed with respect to walking speed/distance and mobility no significant effects: 10MWT (P = 0.169), 6minWT (P = 0.037); SCIM III mobility (P = 0.018), and WISCI II (P = 0.180). Concerning walking speed/distance, seven tested subjects met the limits for the MCID of 10MWT and one subject reached the MCID of 6minWT at post-assessment. In addition, a clinically important difference was shown in two subjects using WISCI II.

At follow-up assessment, a significant increase (P ≤ 0.017) in comparison with the averaged pre-baseline and baseline was found in functional mobility (TUG, P = 0.005). No significant changes were found in muscle strength (LEMS, P = 0.065) and balance (BBS, P = 0.28) as well as in walking speed/distance and mobility: 10MWT (P = 0.169), 6minWT (P = 0.32), SCIM III mobility (P = 0.026), and WISCI II (P = 0.317).

Eleven subjects rated their motor function using the adapted PGIC after the last day of treatment rated on a 7-point scale (Table 3). Seven subjects rated motor function as markedly improved (much improved or very much improved).

Improvements were found in 64% of the subjects in ankle dorsiflexion (L4), and a general increase in lower limb muscle strength (LEMS) was achieved. In addition, an MCID by one-third of the subjects was reached. These changes may have been of relevance in enabling the observed gains in walking speed, stability, and mobility (reflected by improvements in TUG and BBS). However, it has been argued that changes in the LEMS, i.e., muscle strength, do not necessarily relate to changes in SCIM and WISCI (31). Reliance on walking aids and/or personal assistance evaluated by SCIM mobility and WISCI II seem not to be supported by the actual training. Along the same line, only two subjects showed clinically important differences using WISCI II and no significant overall effect.

Furthermore, longer walking distances included in the outcome measures SCIM mobility and 6minWT (only one subject reached MCID) might not be specifically trained by the performed VR tasks. Nevertheless, the TUG assesses various aspects of mobility which are correlated with performance in for example the 10MWT (one-third of the tested subjects met the limits for the MCID) in subjects with iSCI (23). In healthy elderly people, the TUG shows excellent correlation with the BBS and gait speed (20) and is moderately related to executive function (32). The BBS correlates well with other mobility measures and muscle strength (33, 34) and is important for postural control (35).

In accordance with the objective assessments, 64% of the subjects rated their motor functions as markedly improved (i.e., PGIC ≥ 4) when they were asked after the treatment: “since beginning of this program, how would you describe the change (if any) in activity limitations and motor function related to your condition?” The other tested subjects reported at least minimal improvements (i.e., PGIC 2 or 3). This is in opposition to a report claiming that evaluated functional improvements were not actually perceived as such by subjects with a SCI (36).

Despite the task specificity of the training, i.e., isolated single-joint movements, subjects improved on global functional scores. In other words, this type of training with task- and deficit-specific variation enabled a transfer of gained function to other tasks (e.g., functional mobility). In line with our results, a study with lower limb strength training (12 weeks, 30 sessions) also showed a transfer of specific training to general improvements. They found attenuated neuromuscular impairments and improved locomotion in subjects with iSCI (37). Another study found that muscle strength training of the lower limbs generates improved results in walking-related outcome measures compared to robot-assisted gait training in subjects with iSCI with limited ambulatory function (38). Hence, these findings support our results that functional improvements, e.g., mobility, may also be feasible with no task-specific training.

The central idea of motor learning is repeating exercises, feedback about movement execution and motivation (12). When for example the ankle movements are calculated per leg per training session, the subjects executed around 300 movements each. This high number of repetitions, leading to increased internal stimulation of higher motor brain areas, might have induced plastic changes. It has been shown that improvements induced by motor training in healthy controls correlates with an increase in cortical volume in humans (39–43) and monkeys (44). The effects were stronger when a new task was learned and the changes occurred rather rapidly (within weeks) (39, 44). As in the current study, the action processing system is activated by visuo-motor tasks and its downstream motor areas are involved in movement execution, thus effectively promoting functional recovery (45). For subjects with chronic iSCI, there seems to be similar effects. Our group could recently show training-induced functional improvements (e.g., balance) accompanied by structural brain plasticity after intense VR-augmented training, i.e., isolated single-joint movements, for the lower limbs with a high number of repetitions (46). However, few clinical studies adequately address the issue of dosing and timing of lower and upper limb rehabilitation after SCI (47). Wirz et al. (10) demonstrated that longer trainings led to improved mobility in subjects with iSCI.

As mentioned earlier, beside the amount of training, active participation and motivation is a key element for achieving functional improvements (11). The motivation after training was quite high (8.6 on an 11-point NRS from 0 to 10), and no pain or spasticity was reported by the subjects playing the VR tasks. Playing games interactively with a high motivation and the combination of observation and motor imagery, i.e., being a part of the immersive environment, enhance the activation of sensorimotor networks (46, 48).

Interestingly, the clinic-based training, i.e., training at the hospital with one-to-one therapist–patient support, showed greater improvements of balance and walking capacity in the absolute assessment values as the currently tested home-based system (7). The setup in both studies was similar but differed in therapist presence during the training sessions. Therefore, human therapist presence and feedback may have an additional effect on subject’s motivation and should be further investigated.

The study described here is limited by the small and heterogeneous group of 12 subjects with iSCI (AIS C/D with various lesion levels), the uncontrolled study design and non-blinded assessment. While the pre-baseline did at least help to ensure that each subject was at a stable level before the beginning of the intervention, it is recommended to conduct a larger randomized controlled study to verify or refute the conclusions of this study.