Walking rehabilitation in incomplete spinal cord injury: evaluating the impact of robotic exoskeleton-assisted training

Abstract

Background

The use of lower limb exoskeletons in clinical rehabilitation has expanded in recent years, offering potential benefits for walking recovery. However, current clinical evidence on their effectiveness remains inconclusive. Additionally, the way individuals adapt to these robotic devices and how this adaptation contributes to functional improvements is not yet fully understood. This study was intended to (1) investigate the safety and feasibility of the Fourier X2 exoskeleton for walking rehabilitation and (2) examine its effect on walking function following a rehabilitation program.

Methods

A randomized controlled trial was undertaken with 46 individuals who had suffered a spinal cord injury (SCI) within the last year. Participants were randomly allocated into two groups: an intervention group (IG), which received gait training using the Fourier X2 exoskeleton, and a control group (CG), which underwent conventional gait training. Each participant completed 20 gait training sessions lasting one hour. The neurological impairment ranged from C2 to L4, with participants classified under the American Spinal Injury Association Impairment Scale (AIS) C or D. The treatment program involved 20 gait training sessions, each lasting one hour, utilizing the Fourier X2 exoskeleton. Safety was assessed by tracking adverse events, while pain and fatigue levels were measured using the Visual Analogue Scale (VAS). Functional outcomes were evaluated through the Lower Extremity Motor Score (LEMS), Walking Index for Spinal Cord Injury II (WISCI-II), Spinal Cord Independence Measure III (SCIM-III), and various walking assessments, including the 10-Meter Walk Test (10MWT), 6-Minute Walk Test (6MWT), and Timed Up and Go (TUG).

Results

No major complications were observed during the study. Participants in the intervention group (IG) reported experiencing mild pain (1.7 cm, SD 1.1) and moderate fatigue (3.3 cm, SD 1.6) as measured by the Visual Analogue Scale (VAS, 0–10 cm range). Statistical analysis of WISCI-II scores showed notable progress in both the “group” effect (F = 17.82, p < 0.001) and the “group-time” interaction (F = 7.27, p = 0.03). Further post-hoc evaluations revealed that the IG achieved a significant improvement of 3.20 points (SD 2.16, p < 0.0001), whereas the control group (CG) demonstrated a minor, non-significant increase of 0.5 points (SD 1.31, p = 0.32). No other variables showed significant differences between the two groups.

Conclusions

The Fourier X2 exoskeleton is both safe and well-tolerated in clinical environments. Participants who received training with the device exhibited enhanced walking independence, as reflected in their WISCI-II scores.

Trial registration: The Chinese Clinical Trial Register (ChiCTR) includes this study under registration number ChiCTR2000041186, dated December 21, 2020.

Introduction

Spinal cord injury (SCI) often results in partial or complete paralysis of the lower limbs, significantly impairing mobility and overall quality of life. Regardless of the severity, duration, or age of onset, restoring walking ability remains a fundamental objective for individuals with SCI [1–3]. Rehabilitation therapies aimed at restoring walking function following spinal cord injury (SCI) emphasize the promotion of neural plasticity, the strengthening of preserved muscle activity, and the enhancement of compensatory mechanisms [4, 5]. High-intensity stepping training, for example, can stimulate central pattern generators (CPGs), which are neural networks responsible for coordinating rhythmic movement patterns. This stimulation can promote neuroplastic adaptations in both the spinal cord and the sensorimotor cortex, especially in individuals with incomplete SCI, given that the training is maintained consistently over time [6–9]. Additionally, repetitive, task-specific training facilitates the concurrent engagement of sensory and motor pathways, strengthening spinal networks and enhancing the execution of practiced movement [10–12]. As a result, effective walking rehabilitation must be both intensive and task-oriented.

For over 20 years, robotic technology has been utilized to enhance gait rehabilitation by providing controlled, weight-supported stepping exercises with high precision [13, 14]. However, the effectiveness of robotic rehabilitation compared to conventional therapy remains uncertain, as current evidence does not definitively establish its superiority [15–18]. A notable limitation of robotic-assisted training is its modification or omission of essential rehabilitation components, including full weight-bearing during the stance phase and crucial visual-vestibular inputs. During normal overground walking, full weight-bearing promotes activation of antigravity muscles, joint loading, and proprioceptive feedback necessary for neuroplastic adaptation. Similarly, visual and vestibular inputs contribute to postural stability, spatial orientation, and dynamic balance. However, many conventional robotic gait trainers particularly treadmill-based systems partially support body weight and constrain head and trunk movement, thereby reducing these sensory and motor stimulations. The lack of a fully natural training environment may partially account for the inconsistent correlation between therapy intensity and functional gait improvements [19]. Recently, ambulatory robotic exoskeletons have gained attention as a viable alternative, enabling task-specific, overground gait training, especially for individuals with incomplete SCI who demonstrate potential for recovery. In contrast to traditional robotic trainers, exoskeletons offer greater challenges in balance and physical exertion while delivering more natural visual and functional feedback during ambulation.

The feasibility of robotic exoskeletons in rehabilitation has been assessed in individuals with both acute and chronic SCI, focusing on their safety, usability, and overall patient comfort [19–22]. Robotic exoskeletons facilitate safe ambulation while minimizing energy demands compared to passive orthoses in individuals with thoracic-level spinal cord injuries [23, 24]. Despite continuous advancements in exoskeleton technology and durability, clinical guidelines and established indications for their application have not evolved at the same rate. Although certain studies have demonstrated enhancements in functional and spatiotemporal gait parameters, such as walking speed, the overall evidence supporting the efficacy of robotic exoskeleton-based rehabilitation remains insufficient and inconclusive. This is partly due to variations in study designs, including differences in robot types, control mechanisms, treatment duration, and session frequency, which complicate the ability to draw reliable clinical conclusions [25]. Most studies to date have involved small, heterogeneous patient samples with inconsistent protocols [26]. The largest study on robotic exoskeletons, conducted using devices manufactured by Ekso Bionics, evaluated gait training across diverse subpopulations of SCI patients [24, 27]. However, critical questions remain unanswered, particularly regarding whether robotic exoskeleton training improves walking ability and which device-specific factors contribute to these outcomes.

This study presents findings from a prospective, randomized, comparative trial assessing the clinical effectiveness of Fourier X2, a novel lower-limb robotic exoskeleton. Although Fourier X2 exhibits similarities with existing exoskeletons, its distinct features necessitate further exploration. The Fourier X2 is an overground, battery-powered exoskeleton that provides active actuation at the hip and knee joints, allowing users to perform repetitive, task-specific walking movements on flat surfaces. It includes adjustable ergonomic cuffs, real-time feedback sensors, and a control tablet for setting assistance levels. These features aim to deliver a more natural walking experience with greater engagement of balance and postural control compared to treadmill-based systems. This research seeks to expand the existing evidence on the effects of robotic exoskeleton-assisted gait training in individuals with incomplete SCI, addressing the fundamental question of whether such training leads to meaningful improvements in walking function [28, 29]. The primary objectives of this study were: (1) to evaluate the safety and adherence of robotic-assisted gait training with the Fourier X2 exoskeleton (the study was initially designed using the X1 model, but the final protocol incorporated the upgraded X2 version with improved mechanical and electronic performance), and (2) to compare improvements in walking function between a Fourier X2-based training program and conventional overground gait therapy in non-chronic SCI patients. It was hypothesized that exoskeleton-assisted training would result in greater functional walking gains than traditional rehabilitation methods. This study aimed to evaluate the safety, feasibility, and clinical effects of the Fourier X2 exoskeleton in individuals with incomplete spinal cord injury (SCI) within the first-year post-injury. Unlike previous protocols, this research focuses on a recently updated exoskeleton model and examines its potential contribution to walking independence, thereby addressing current gaps in evidence regarding device-specific rehabilitation outcomes.

Results

A total of 48 individuals with incomplete SCI were assessed for eligibility during recruitment. Participants were evenly allocated into two groups, with 24 assigned to the intervention group and 24 to the control group through a 1:1 randomization process. However, two participants dropped out due to reasons unrelated to the study one developed a urinary tract infection, while the other was discharged before completing the intervention. Specifically, one participant was lost to follow-up in each group, as also illustrated in Fig. 1. As a result, 46 participants successfully completed the study and were included in the final analysis. Table 1 outlines the demographic and clinical characteristics of the sample. The majority had sustained their injuries within the past 6 months, and both groups demonstrated similar baseline distributions in terms of age, gender, time since injury, injury level, AIS classification, LEMS, and WISCI-II scores (Table 1).

Fig. 1.

CONSORT flow diagram

Table 1.

Patients´ baseline characteristics

Characteristic	Intervention group (IG)	Control group (CG)	Levene’s test value (p-value)
Age	43.6 (14.78)	53.2 (15.93)	F = 0.34 (p = 0.81)
Sex
Male	15	16	F = 0.16 (p = 1.65)
Female	8	7
Injury level
C2–C8	2	4	F = 0.67 (p = 0.32)
T1–T6	9	3
T7–L1	6	10
L2–L4	5	6
AIS
C	16	9	F = 1.02 (p = 0.45)
D	7	14
Time since injury (months)	6.72 (1.8)	7.62 (2.1)	F = 2.17 (p = 0.16)
LEMS (0–50)	31.78 (13.87)	38.83 (14.31)	F = 1.31 (p = 0.21)
WISCI‑II (0–20)	12.46 (4.58)	11.9 (3.51)	F = 0.04 (p = 0.47)

Variables are expressed as mean (SD). (n), number of participants able to finish the study, AIS American Spinal Injury Association Impairment Scale, LEMS Lower Extremity Motor Scale, WISCI-II Walking Index for Spinal Cord Injury II. *p ≥ 0.05

Safety, feasibility, pain, and fatigue

No incidents of falls were documented during the study, indicating a safe training environment. However, three participants from the IG developed mild skin redness in the tibial contact area during their first training session, most likely due to prolonged pressure from the exoskeleton straps. This minor issue was promptly addressed by adding extra padding to the affected areas, leading to the redness subsiding within 24 h without further complications. Additionally, seven IG participants reported experiencing mild soreness in their neck and shoulder muscles approximately 24 h after completing their training session. This discomfort was likely associated with the prolonged use of walking aids, which may have placed additional strain on these muscle groups. No adverse events or discomfort were reported in the control group (CG). Regarding overall post-session discomfort levels, IG participants rated their pain at an average of 1.7 cm (SD 1.1) and their fatigue at 3.3 cm (SD 1.6) on the Visual Analogue Scale (VAS, ranging from 0 to 10 cm), suggesting that while mild discomfort was present, it remained within tolerable limits.

Effect on LEMS

A two-way ANOVA demonstrated a statistically significant effect of the "time" factor on LEMS scores (F = 17.3, p < 0.001), indicating notable improvements over the study period. However, neither the "group" factor (F = 0.13, p = 0.73) nor the "time-group" interaction (F = 0.4, p = 0.2) showed meaningful differences, suggesting that both groups experienced similar trends in recovery. Further post-hoc pairwise comparisons, adjusted using the Bonferroni method to control for multiple testing, revealed a significant increase in LEMS scores following the intervention. Specifically, participants in the IG exhibited an average improvement of 5.15 points (SD 4.63; p = 0.02, η² = 0.41), while those in the CG showed a comparable gain of 4.60 points (SD 2.78; p = 0.03, η² = 0.36; see Table 2). Despite these improvements within each group, the difference between IG and CG remained statistically insignificant (p = 0.38, η² = 0.04), indicating that both approaches led to similar levels of motor recovery.

Table 2.

Lower extremity motor score (LEMS) assessment

Intervention Mean (SD)			Pairwise comparison (P-value, η2)
	Baseline	Post-intervention	Change	Baseline vs. Post intervention	IG vs CG
IG	30.38 (13.67)	35.53 (11.52)	5.15 (4.63)	0.02*, η2 = 0.41	0.378, η2 = 0.04
CG	32.76 (7.63)	37.36 (6.97)	4.60 (2.78)	0.03*, η2 = 0.36

CG control group, IG intervention group; *significant differences (p < 0.05). η2: calculated effect size

Effects on functional scales

Functional performance was assessed at both baseline and post-intervention, with effect sizes presented in Table 3. A two-way ANOVA identified significant improvements over time in key mobility measures, including the 10MWT (F = 13.78, p < 0.001), TUG test (F = 24.38, p < 0.001), and 6MWT (F = 13.74, p < 0.001). However, no significant differences were observed for the "group" factor (10MWT: F = 0.23, p = 0.57; TUG: F = 0.03, p = 0.83; 6MWT: F = 0.31, p = 0.59) or the "time-group" interaction (10MWT: F = 0.83, p = 0.31; TUG: F = 2.61, p = 0.06; 6MWT: F = 0.71, p = 0.62), suggesting similar trends in both groups. Walking speed showed a significant increase, with the IG improving by 0.34 m/s (SD 0.19) (p = 0.04) and the CG by 0.21 m/s (SD 0.18) (p = 0.03). Performance on the TUG test also improved in both groups, with the IG reducing completion time by 12.6 s (SD 6.72) (p < 0.001) and the CG by 6.2 s (SD 8.31) (p = 0.03). Similarly, endurance, measured by the 6MWT, showed significant gains, with the IG covering an additional 57.39 m (SD 39.7) (p = 0.03) and the CG increasing by 45.89 m (SD 31.59) (p < 0.05). Despite these positive changes within each group, no significant differences were found between the IG and CG in any of the functional measures (p > 0.05), indicating comparable outcomes regardless of group allocation.

Table 3.

Functional outcomes following exoskeleton training or convectional training, and comparison among times and interventions

Intervention mean (SD)				Pairwise comparison (P-value, η2)
Outcome measure	Group	Baseline mean (SD)	Post-intervention mean (SD)	Change mean (SD)	Baseline vs. post-intervention (p, η2)	IG vs. CG (p, η2)
10MWT (m/s)	IG	0.25 (0.45)	0.59 (0.64)	0.34 (0.19)	0.04*, η2 = 0.48	0.74, η2 = 0.03
10MWT (m/s)	CG	0.34 (0.18)	0.55 (0.36)	0.21 (0.18)	0.03*, η2 = 0.42	–
TUG (s)	IG	43.42 (17.56)	30.82 (14.84)	− 12.62 (6.72)	0.01*, η2 = 0.73	0.82, η2 = 0.37
TUG (s)	CG	36.34 (26.45)	30.14 (18.15)	− 6.2 (8.31)	0.03*, η2 = 0.54	–
6MWT (m)	IG	121.54 (74.03)	178.93 (113.73)	57.39 (39.7)	0.03*, η2 = 0.62	0.66, η2 = 0.02
6MWT (m)	CG	117.73 (64.93)	163.72 (133.52)	45.89 (31.59)	0.04*, η2 = 0.35	–
WISCI-II (0–20)	IG	7.34 (2.64)	12.24 (3.85)	4.9 (1.21)	0.02*, η2 = 0.56	0.01*, η2 = 0.43
WISCI-II (0–20)	CG	10.56 (4.72)	11.92 (5.31)	1.39 (0.59)	0.34, η2 = 0.06	–
SCIM-III (0–100)	IG	72.54 (5.89)	76.14 (9.83)	3.6 (3.94)	0.04*, η2 = 0.47	0.76, η2 = 0.02
SCIM-III (0–100)	CG	71.82 (4.23)	77.45 (6.01)	5.63 (1.78)	0.03*, η2 = 0.34	–

CG control group, IG intervention group, *significant differences (p < 0.05). η²: calculated effect size. 10MWT 10 Meters Walking Test, TUG Test Up and Go, 6MWT 6 Minutes Walking Test, WISCI‑II Walking Index Spinal Cord Injury‑II, SCIM‑III Spinal Cord Independence Measurement‑III

Evaluation of WISCI-II scores revealed a significant influence of the "group" factor (F = 17.82, p < 0.001) and the "group-time" interaction (F = 7.27, p = 0.03). Post-hoc analysis highlighted a substantial improvement of 4.9 points (SD 1.21) in the intervention group (IG) (p < 0.001), while the CG exhibited only a slight, non-significant increase of 1.39 points (SD 0.59; p = 0.34). A detailed breakdown of WISCI-II outcomes is provided in Table 4.

Table 4.

WISCI II outcomes and technical aid used in functional gait test (10MWT, TUG, 6MinWT) at baseline and post-intervention time

Group	Participant	Baseline assessment				Post-intervention assessment
		Use of devices	Braces	Physical assistance	WISCI II score	Use of devices	Braces	Physical assistance	WISCI II score
IG	1	Walker*	No	No	12	Two crutches*	No	No	15
	2	Parallel*	No	One person	14	Two crutches*	No	No	16
	3	Two crutches*	No	No	15	Two crutches*	No	No	16
	4	Parallel	Yes	One person	4	Parallel	No	No	6
	5	Walker*	Yes	No	9	Two crutches*	No	No	12
	6	Walker	Yes	One person	5	Walker	No	No	10
	7	Parallel	No	One person	5	Parallel	No	No	6
	8	Two crutches	Yes	No	15	Two crutches	No	No	17
	9	Walker*	Yes	Yes	8	Two crutches*	Yes	No	10
	10	Parallel	Yes	One person	5	Parallel	Yes	No	5
	11	Two crutches	Yes	No	14	Two crutches	No	No	17
	12	Walker	Yes	One person	5	Walker	Yes	No	10
	13	Walker*	No	One person	9	Two crutches*	No	No	15
	14	Walker*	Yes	No	7	Two crutches*	Yes	No	12
	15	Parallel	Yes	One person	6	Parallel	Yes	No	9
	16	Two crutches*	Yes	No	14	One crutch*	Yes	No	17
	17	Parallel*	Yes	One person	6	Walker*	Yes	No	12
	18	Walker	Yes	One person	7	Walker	Yes	One person	7
	19	Two crutches*	No	No	12	One crutch*	No	No	17
	20	Walker *	Yes	No	8	Two crutches*	Yes	No	12
	21	Parallel*	Yes	One person	6	Walker*	Yes	No	12
	22	Walker*	Yes	No	8	Two crutches*	Yes	No	12
	23	Two crutches*	No	No	11	One crutch*	No	Yes	16
CG	1	Walker	Yes	One person	7	Walker	Yes	One person	7
	2	Walker	Yes	One person	8	Walker	Yes	One person	8
	3	Two crutches	Yes	No	14	Two crutches	Yes	No	15
	4	Walker	Yes	No	10	Walker	Yes	No	10
	5	Parallel	Yes	One person	6	Parallel	Yes	One person	8
	6	Two crutches	No	No	15	Two crutches	No	No	15
	7	Parallel	Yes	One person	7	Parallel	Yes	One person	7
	8	Two crutches	No	No	12	Two crutches	No	No	13
	9	Walker	Yes	One person	10	Walker	Yes	One person	12
	10	Two crutches	No	No	15	Two crutches	No	No	15
	11	Parallel	Yes	One person	6	Parallel	Yes	One person	6
	12	Two crutches	No	No	14	Two crutches	No	No	16
	13	Two crutches*	Yes	No	12	One crutch*	No	No	14
	14	Two crutches*	No	No	15	One crutch*	No	No	18
	15	Parallel	Yes	One person	5	Parallel	Yes	One person	7
	16	Walker	Yes	No	12	Walker	Yes	No	13
	17	Two crutches	No	No	14	Two crutches	No	No	15
	18	Two crutches	Yes	No	13	Two crutches	No	No	14
	19	Parallel	Yes	One person	6	Parallel	Yes	No	8
	20	Parallel	Yes	One person	5	Parallel	Yes	One person	6
	21	Walker	No	One person	13	Walker	No	No	14
	22	Walker	No	No	11	Walker	No	No	12
	23	Two crutches*	Yes	No	13	One crutch*	No	No	16

CG control group, IG intervention group; *change of assistive device from pre- to post-intervention, 10MWT 10 Meters Walking Test, TUG Test Up and Go, 6MWT 6 Minutes Walking Test, WISCI-II Walking Index Spinal Cord Injury- II

For SCIM-III scores (Table 3), a significant effect was detected for the "time" factor (F = 10.3, p = 0.03), whereas neither the "group" factor (F = 0.24, p = 0.67) nor the "time-group" interaction (F = 0.42, p = 0.83) reached statistical significance. Post-hoc comparisons demonstrated notable gains in functional independence, with the IG showing an increase of 3.6 points (SD 3.94) (p = 0.04) and the CG improving by 5.63 points (SD 1.78) (p = 0.03). However, the difference between the two groups after the intervention remained statistically insignificant (p = 0.76).

Discussion

This study explored the effectiveness of robotic-assisted gait training for individuals with incomplete spinal cord injury (SCI) within a year of onset. Conducted as a prospective, randomized clinical trial with blinded evaluators, the research focused on two key aspects: (1) determining the safety and adherence of training with the Fourier X2 exoskeleton and (2) comparing its impact on walking function against conventional rehabilitation methods.

The exoskeleton training was well tolerated, with no severe adverse effects reported. A few participants experienced minor skin irritations, but none discontinued the study due to device use. Both groups showed post-treatment improvements in clinical and functional measures; however, participants who trained with the robotic exoskeleton demonstrated a greater enhancement in WISCI-II scores. This highlights the potential of ambulatory robotic exoskeletons and offers valuable insights into their underlying mechanisms, as explored in the following discussion.

The Fourier X2 exoskeleton introduces several advancements over its predecessor, the Fourier X1, which has demonstrated feasibility in providing walking assistance for individuals with stroke who retain some walking ability [30]. Like other exoskeletons, the Fourier X2 features active actuation of joint flexion/extension while restricting movement in other degrees of freedom. However, it also incorporates unique design elements, such as active ankle joint actuation, which may enhance gait symmetry and stability. Additionally, its lightweight and modular design allows for greater adaptability to different body types and rehabilitation needs. The Fourier X2’s design characteristics including active hip and knee actuation, real-time feedback sensors, and an overground configuration that encourages active balance engagement may have contributed to improved coordination and reduced reliance on external assistance in the intervention group. Although these device-specific effects were not directly quantified, they provide a plausible explanation for the greater WISCI-II improvements observed following exoskeleton-assisted training.

The Fourier X2 also differs in its strapping system, which uses ergonomic, adjustable cuffs with quick-release mechanisms for improved comfort and ease of use. Its actuator technology and sensor integration provide precise, real-time feedback, enabling more personalized and effective rehabilitation sessions. The device’s control interface is user-friendly, featuring a touchscreen tablet that connects via Bluetooth, allowing therapists to adjust assistance levels and monitor progress seamlessly. These upgrades represent engineering and ergonomic improvements aimed at enhancing user comfort and adaptability, but no direct clinical evidence currently confirms superior functional outcomes for the X2 compared to the X1.

While there is ongoing debate about the clinical effectiveness of robotic-assisted rehabilitation compared to conventional methods, this study does not aim to resolve this broader question or evaluate whether the specific design differences of the Fourier X2 lead to superior outcomes. Instead, it seeks to contribute to the growing body of evidence on the impact of robotic exoskeletons in similar patient populations, particularly individuals with incomplete SCI. By examining the effects of the Fourier X2 in this context, we hope to advance understanding of how such technologies can improve walking rehabilitation outcomes and inform future clinical applications [31].

This research highlights the practicality of integrating the Fourier X2 exoskeleton into walking therapy for individuals with incomplete SCI. The study specifically targeted participants within 12 months post-injury, a critical period when neurological recovery is more dynamic, potentially maximizing therapeutic benefits. Notably, the majority were within 6 months of injury, reinforcing the notion that early exoskeleton-based rehabilitation may yield superior outcomes. Although no subgroup analysis was performed, participants who initiated training within the first 6 months post-injury appeared to show slightly greater improvements in walking independence compared to those beyond 6 months. However, this observation should be interpreted cautiously, as the sample size was insufficient to determine statistical significance. However, the ideal timing, duration, and treatment protocols for such interventions remain undefined. Recent studies on exoskeleton-assisted gait training in acute SCI have primarily focused on safety and feasibility rather than neurological recovery [21].

With the exception of two individuals, all participants completed the full 15-session training program, and the withdrawals were unrelated to the intervention. Our findings on feasibility and adverse events are consistent with previous research [21, 27, 32, 33]. No joint swelling was observed throughout the training, aligning with earlier studies [21, 27, 33], and only three participants experienced minor skin irritation in contact areas, which resolved quickly between sessions. These results are comparable to those reported for Ekso and ReWalk exoskeletons [21, 27, 32] and show better outcomes than some other studies [6]. No falls occurred, similar to findings in prior research [21], though some participants reported neck and shoulder discomfort, likely due to the physical effort required when using walking aids. Despite this, none requested modifications to their external support, and all demonstrated improved walking ability with the exoskeleton. Although some participants may have been capable of progressing to less supportive aids, such as switching from a walker to crutches, we prioritized maintaining their initial support to maximize step count during therapy. Reducing support could have introduced greater physical strain, potentially leading to fatigue and limiting overall step volume. Interestingly, 14 of the 23 participants in the intervention group were able to transition to less supportive walking aids during or after the training period, compared to only three participants in the control group. Although this variable was not analyzed statistically and no standardized criteria exist for such transitions, this observation suggests that the exoskeleton may have facilitated improvements in balance and lower-limb control that allowed for reduced reliance on external support. Additionally, there is no established guideline on the optimal timing for transitioning between assistive devices in exoskeleton-based rehabilitation. Based on these considerations, we conclude that the Fourier X2 exoskeleton is a safe and viable option for walking therapy.

Pain and fatigue levels during robotic gait training were measured using the Visual Analog Scale (VAS) [21, 34], a validated tool for assessing subjective symptoms and perceptions [35]. Participants reported mild pain (below 3 cm on the VAS) and moderate fatigue (below 4 cm) during sessions. These findings align with previous research [21, 27, 36, 37], although some studies utilized the Borg Scale’s Rate of Perceived Exertion instead.

Consistent with earlier research [21, 27, 32–34, 38], both the intervention (IG) and control groups (CG) showed significant improvements in LEMS and functional gait assessments, including the 10MWT, TUG, and 6MWT. These findings suggest that exoskeleton-assisted walking therapy is as effective as conventional rehabilitation when considering these measures alone. However, the IG demonstrated a significantly greater improvement in WISCI-II scores, indicating enhanced walking ability. While WISCI-II is generally considered less sensitive to gait function changes than the 10MWT and TUG in individuals with incomplete SCI [39], its superior improvement in the IG suggests that robotic exoskeletons may provide added benefits in walking rehabilitation. This selective improvement in WISCI-II, without parallel changes in other functional outcomes or LEMS, may be attributed to the task-specific nature of the exoskeleton-assisted training. The Fourier X2 program emphasized repetitive stepping, balance control, and interlimb coordination—skills directly assessed by the WISCI-II rather than isolated lower-limb muscle strength measured by LEMS. Consequently, participants likely improved in walking independence through neuromotor adaptation and improved balance confidence rather than strength gains alone. This explanation aligns with previous studies reporting that gait-focused robotic training can enhance functional walking performance even when motor recovery scores remain unchanged [27, 32].

The greater improvement in the IG suggests a lower level of impairment compared to the CG following exoskeleton training. Since the WISCI-II accounts for the level of physical assistance, walking aids, and braces required during ambulation, the greater improvement observed in the IG likely reflects reduced reliance on external support and enhanced walking independence rather than substantial gains in muscle strength. This clarification better highlights the potential of robotic gait therapy to promote functional walking autonomy in individuals with incomplete SCI [40]. The connection between impairment severity, as indicated by LEMS, and walking ability measured by WISCI-II is widely recognized in individuals with SCI [41]. No significant differences in LEMS were identified between the groups in this study, suggesting that exoskeleton-assisted training may contribute to enhanced balance control during ambulation. Given that robotic exoskeletons do not inherently provide balance support, the use of external assistive devices is essential, requiring individuals to engage in active postural stabilization. Unlike passive orthoses, which are associated with high energy expenditure, reduced therapy duration, and an increased risk of upper limb injuries, exoskeletons facilitate controlled mobility while promoting balance-related adaptations. The primary finding of this study is that robotic exoskeletons enhance walking ability in individuals with incomplete SCI, as reflected by greater WISCI-II improvements in the intervention group. These gains may be partly influenced by improved dynamic balance and coordination rather than by compensatory mechanisms alone, as the present study did not directly assess neuromuscular or kinematic changes. Accordingly, compensatory and balance-related strategies should be interpreted as potential contributing factors rather than confirmed outcomes. However, additional variables such as upper limb movement, postural control, and electromyographic (EMG) activity were not assessed, necessitating further research to substantiate this hypothesis. The functional improvements observed after only 5 weeks of training are likely related to early neural and motor-control adaptations rather than to structural muscular changes. Repetitive, task-specific stepping may enhance interlimb coordination, proprioceptive feedback, and central pattern activation, leading to improved balance and gait symmetry. Such short-term neural adaptations have been reported in previous studies of robotic and treadmill-based gait training and may explain the significant gains in walking independence achieved by participants in the intervention group.

Although participants in both the IG and CG demonstrated improvements in walking speed (10MWT) and distance (6MWT), those in the IG required less external assistance, as reflected by WISCI-II scores. The increase in walking speed observed in the intervention group (0.3 m/s) is close to the threshold commonly regarded as clinically meaningful for individuals with incomplete SCI, which ranges from approximately 0.3–0.4 m/s in previous literature. In contrast, the 0.21 m/s improvement observed in the control group likely represents a minimal detectable change without substantial functional relevance. Similarly, the 6MWT gains of over 50 m in the intervention group can be interpreted as functionally significant, suggesting enhanced endurance and community ambulation potential. These findings reinforce that exoskeleton-assisted training may translate into practical improvements in walking ability beyond statistical differences. Despite these benefits, the use of exoskeletons presents several challenges, including the time required for donning, setup, and removal, the high cost of the device, the necessity for trained personnel, and the constraints of limited battery life. A comprehensive cost–benefit analysis of this technology falls beyond the scope of this study. Participants with neurological injury levels ranging from C2 to L4 were included, based on the assumption that both upper motor neuron (UMN) and lower motor neuron (LMN) impairments could benefit from exoskeleton-assisted training.

Several critical aspects of exoskeleton-assisted therapy remain undefined, including the ideal dosage and duration of training, the appropriate criteria for progressing between different walking aids (e.g., from parallel bars to walkers to crutches), and the potential benefits of incorporating advanced gait activities. A notable strength of this study is its emphasis on individuals in the early stages of SCI recovery (within the first-year post-injury), providing valuable insights into the optimal application and clinical indications of exoskeleton-based rehabilitation.

Study limitations

This study has several limitations. First, we did not tailor guidance force or step initiation detection to each patient’s needs. The concept of Assist-As-Needed (AAN) [42], which adapts robotic guidance to the patient’s actual walking capacity, was proposed shortly after robotic walking assistance was introduced, based on the Challenge Point Theory [42]. However, despite extensive research, there is no consensus on how to optimize robotic guidance for therapy [43]. Determining the optimal level of assistance requires measuring the patient’s walking capacity, which depends on factors such as joint force production (estimated via EMG), central fatigue, and motivation. Some studies have proposed AAN implementations based on muscle fatigue estimates [44], model-based lower limb biomechanics [45], or user-exoskeleton force interactions [46]. However, these implementations have sometimes led to "slacking" [47]. We believe that a well-implemented AAN paradigm, accounting for factors like central fatigue, motivation, and biomechanical complexity, could improve therapeutic outcomes. Future research should focus on optimizing the device to meet patients’ functional needs [47]. Second, this study did not include gait kinematic analysis or the use of standardized gait quality scales. While functional tests such as the 10MWT, 6MWT, and WISCI-II were used to assess walking performance, these measures do not capture the biomechanical characteristics of gait. Future research incorporating motion analysis and qualitative gait assessment tools could provide more detailed insights into the motor adaptations associated with exoskeleton-assisted walking.

Third, the study was limited by a relatively small sample size, which prevented stratification by lesion level or time since injury. Given that functional recovery potential varies according to both neurological level and chronicity of injury, future trials with larger samples are needed to allow stratified analyses and improve the reliability of subgroup comparisons. Although the IG appeared younger and had a shorter duration post-injury, which may have contributed to greater improvements, these demographic differences were not statistically significant likely due to the limited sample size. Fourth, no follow-up assessments were conducted beyond the intervention period, restricting insights into the long-term effects of the training. Fifth, logistical and time constraints influenced both the recruitment period and the number of training sessions. Sixth, blinding the intervention to participants was not feasible, introducing the possibility of a placebo effect. Finally, although a crossover or combined-intervention subgroup could provide valuable within-subject comparisons between conventional and exoskeleton-based training, this design was not feasible in the current study due to safety considerations and the risk of carryover effects. Future research could explore such designs to better delineate individual responsiveness to different rehabilitation modalities.

Conclusion

This research reports the findings of a prospective, randomized, comparative trial investigating gait training with the Fourier X2 exoskeleton. The results support the safety and feasibility of this device for rehabilitation in individuals with SCI within the first-year post-injury. While both the IG and CG demonstrated enhancements in walking speed and endurance, the IG exhibited a greater reduction in external assistance, as indicated by WISCI-II scores. These findings suggest that robotic exoskeletons may facilitate walking improvements by promoting coordination and balance-related adaptations; however, such mechanisms were not directly assessed in this study and should be interpreted as potential contributors requiring further investigation. Further large-scale, multi-center trials are required to validate these findings and optimize therapeutic protocols.

Methods

This study adopted a prospective, randomized, comparative trial design (Fig. 1), with data collection and analysis conducted in line with CONSORT (Consolidated Standards of Reporting Trials) guidelines for non-pharmacological interventions [48]. Participants were recruited using a non-probabilistic convenience sampling method and were then randomly allocated to one of two groups: the intervention group (IG), which received exoskeleton-assisted gait training, or the control group (CG), which underwent traditional gait training. The randomization process was carried out using Randomizer.org to ensure unbiased allocation. Additionally, to maintain objectivity, the evaluator remained blinded to group assignments throughout the study.

The study protocol received ethical approval from the Neijiang Normal University Ethical Committee (NUU-IRB 202204089) and was conducted in accordance with the ethical standards outlined in the Helsinki Declaration (1975), with its most recent revision in 2020. Initially, the study was designed to assess the Fourier X1 exoskeleton; however, the final protocol incorporated an upgraded model, the Fourier X2. The enhancements in the Fourier X2 were confined to refinements in gearbox mechanics and electronic components, while its overall mechanical framework and control system remained unchanged. Consequently, both the Fourier X1 and X2 exoskeletons maintained functional equivalence from a therapeutic standpoint.

Fourier X2 exoskeleton

The Fourier X2 exoskeleton (Fourier Intelligence, China) is a state-of-the-art robotic device designed for gait rehabilitation in patients with neurological impairments, including stroke and spinal cord injury (SCI). Building on the success of its predecessor, the Fourier X1, the X2 incorporates several advancements to enhance usability, adaptability, and therapeutic outcomes. Key improvements include a more compact and lightweight design, enhanced actuator performance, and advanced control algorithms for personalized rehabilitation.

The Fourier X2 features actuators at the hip and knee joints (see Fig. 2), which provide precise and adjustable assistance during walking. The physical interface includes ergonomic, adjustable cuffs for the thighs and shanks, secured with straps and quick-release mechanisms for ease of use and patient comfort (see Fig. 2B). The exoskeleton is connected to a rigid backpack housing the battery, main control unit, and communication electronics (see Fig. 2B). A flexible pelvic structure allows adaptation to different body sizes while maintaining stability in the sagittal and frontal planes (see Fig. 2B). Joint motion is driven by electric actuators that deliver torque following pre-programmed joint trajectories synchronized with gait phases, which are detected through embedded inertial measurement units (IMUs) and joint encoders. The timing of torque application is automatically adjusted to each participant’s step cycle to ensure smooth movement patterns. At 100% guidance, the exoskeleton provides full torque necessary to move the participant’s lower limbs without voluntary muscle contribution, ensuring consistent stepping kinematics. This torque level was set at the beginning of the study and maintained constant across all sessions to standardize the training dose and avoid variability between sessions.

Fig. 2.

Fourier X2 exoskeleton; A frontal view, B lateral view, C Fourier X2 exoskeleton on a patient using two crutches; (1) backpack that containing battery, main microprocessor and the communication electronics connects, (2) attachment and fitting, (3) actuators, (4) force-sensing-resistors for measuring foot-floor contact, (5) flexible arm for adaptation to pelvis

The control interface is user-friendly, with a touchscreen tablet running custom software that connects to the exoskeleton via Bluetooth. The device offers adjustable levels of assistance, ranging from full support (100% guidance) to minimal assistance (0% guidance), allowing therapists to tailor the training intensity to the patient’s capabilities and progression. While this flexibility theoretically enables optimal patient-specific training, there are currently no standardized clinical guidelines for setting and modulating assistance levels based on patient capacity or therapeutic progress. In line with previous studies, we utilized 100% guidance to provide rigid trajectory control, ensuring consistent and repeatable movement patterns during therapy.

Participants

This clinical trial included individuals with incomplete spinal cord injury (SCI) who were within their first-year post-injury, recruited from the Rehabilitation Department at Neijiang Normal Hospital. To qualify, participants had to be between 16 and 70 years old, have an SCI level ranging from C2 to L4, and an injury classified as AIS C or D according to the American Spinal Cord Injury Association Impairment Scale. Additional criteria required them to have sustained upper limb strength (triceps brachii ≥ 3 on the MRC scale) for using assistive devices like walkers or crutches, be capable of standing, and exhibit lower limb spasticity of ≤ 2 on the Modified Ashworth Scale. Written informed consent was also mandatory. Exclusion criteria included pregnancy, any neurological disorders other than SCI, recent lower limb fractures (within the past year), severe joint contractures or arthrodesis, and pressure ulcers at potential exoskeleton contact sites. While the inclusion age range was relatively wide (16–70 years), most participants were between 33 and 59 years old, except for six individuals aged 19, 23, 25, 61, 64, and one participant aged 71 years was included after confirming their medical stability, functional suitability, and ability to complete the training protocol, as this minor deviation was ethically approved and did not affect study safety or outcomes.

Intervention

The intervention group (IG) participated in a structured robotic gait training program spanning 15 sessions over 5 weeks, with three non-consecutive sessions per week. Each session lasted about an hour, divided into different phases: 20 min for putting on and removing the exoskeleton, 30 min of walking exercises, a 5-min break, and another 5 min for data recording. Training was conducted indoors on a flat, stable surface in the physiotherapy room. Participants used external support such as parallel bars, a walker, or crutches based on their functional capacity and preference. A physiotherapist researcher provided hands-on safety assistance, while an engineer operated the exoskeleton system. During sessions, participants received feedback on sustaining their walking pace and, afterward, on the distance walked.

All participants continued receiving a standard multidisciplinary rehabilitation program throughout the study, which included daily physiotherapy sessions focusing on stretching, passive and active-assisted range of motion exercises, upper-limb strengthening, balance training, and functional mobility activities. Occupational therapy and patient education were also provided as part of their inpatient routine.

The control group (CG) followed a traditional gait training program comprising 15 sessions of 30 min each, distributed similarly to the IG (three sessions per week for 5 weeks). The CG protocol included manual gait re-education with physiotherapist assistance, analytical mobilization of lower limbs, strengthening exercises targeting hip and knee extensors, and progressive walking practice within parallel bars or using walking aids when appropriate. Unlike the IG, participants in the CG did not use or test the Fourier X2 exoskeleton at any stage of the study. Training was conducted by the same rehabilitation team under equivalent supervision and safety monitoring conditions. To minimize bias, both groups trained separately, and the evaluator remained blinded to group allocation throughout the study. All Participants in both groups continued their usual medications and rehabilitation programs without modification, meaning that walking training was incorporated for both the IG and CG as an extra component of the study.

Safety and outcome measures

Throughout the study, safety was rigorously monitored, and any adverse events, particularly those involving skin lesions, falls, or pain, were documented and analyzed. Skin assessments were performed before and after each session using the pressure ulcer classification system from the US National Pressure Ulcer Advisory Panel and the European Pressure Ulcer Advisory Panel [49]. After each session, pain and fatigue were assessed using the Visual Analogue Scale (VAS, 0–10 cm) [35]. The evaluation of feasibility and adherence was based on the number of participants who completed the treatment and their reported experiences of pain and fatigue.

A physiotherapist uninvolved in the treatment measured clinical and functional outcomes at baseline and post-intervention. The severity and classification of spinal cord injury (SCI) were established through a neurological examination following the International Standards for Neurological Classification of Spinal Cord Injury (ISNCSCI) [50]. To assess motor function, the Lower Extremity Motor Score (LEMS) [41] was applied, emphasizing lower limb strength rather than the combined motor ability of the upper and lower extremities [40, 51]. The evaluation of participants' performance included the 10-Meter Walking Test (10MWT) [52], Timed Up and Go (TUG) [53], 6-Minute Walking Test (6MWT) [52], Walking Index for SCI (WISCI-II) [54], and Spinal Cord Independence Measure-III (SCIM-III) [55].

Statistical analysis

Statistical analyses were carried out using IBM SPSS v22 (IBM Corporation, Armonk, NY). The sample size calculation was grounded in the Lower Extremity Motor Score (LEMS) as the primary outcome, drawing on pilot study results that showed a mean difference of 8.2 and a standard deviation of 10 [56]. LEMS was selected for this calculation because it is an objective, validated measure of lower-limb motor recovery that strongly correlates with walking ability in individuals with incomplete SCI. At the time of study design, limited published data were available for walking-specific outcomes such as WISCI-II in robotic exoskeleton interventions; therefore, LEMS provided a reliable and conservative variable for estimating the expected variability and effect size. To ensure a Type I error (α) of 0.05 and a power of 80%, each group included 24 participants. Homogeneity in age, gender, time since injury, WISCI-II, and LEMS were assessed through Levene’s test for equality of variances. A two-way repeated-measures ANOVA was applied to assess changes in motor and functional gait outcomes, with “time” (baseline and post-intervention) and “group” (IG and CG) serving as factors. The Greenhouse–Geisser correction was implemented to compensate for non-sphericity, and the Bonferroni post-hoc test was utilized to pinpoint differences between time points and groups. Participants who did not finish the study were excluded from the analysis due to missing post-intervention data. Given the small sample size and the exploratory design, missing data imputation was not performed to avoid potential bias; thus, analyses were conducted on a per-protocol basis. Statistical significance established at p ≤ 0.05.

Author contributions

FAK prepared the initial draft of the manuscript, offered language assistance, and conducted a critical revision of the content. YAK co-initiated the study, critically reviewed the manuscript, provided language support, and proofread the article. GX performed all statistical analysis, data interpretation, and final approval of the manuscript. XD provided additional language support.

Funding

None.

Data availability

The datasets generated and/or analyzed during the current study are not publicly available due [to maintain the privacy of the patients participating in the study] but are available from the corresponding author on reasonable request.

Declarations

Ethics approval and consent to participate

The research involving human subjects was evaluated and sanctioned by the Institutional Review Board of Neijiang Normal University under approval number NUU-IRB 202204089. All procedures adhered to the ethical standards established in the 1964 Declaration of Helsinki and its later amendments. The study was conducted in compliance with applicable guidelines and regulatory requirements. Written informed consent was acquired from all participants and their legal guardians.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.