Personalized Exoskeleton Gait Training in Incomplete Spinal Cord Injury

Abstract

Spinal cord injury (SCI) profoundly affects motor–sensory functions, reducing mobility and quality of life. Robotic exoskeletons offer a promising solution to support gait training, improve mobility, and prevent secondary complications. Existing research predominantly focuses on complete SCI, with limited exploration of long-term effects and tailored training for incomplete SCI. This study investigates device-based outcomes of personalized exoskeleton gait training in 33 individuals with incomplete SCI, with different lesion levels: cervical, thoracic, and lumbar. Participants underwent up to 39 sessions of gait training with a commercially available lower limb exoskeleton. Session parameters, including duration, intensity, and modality, were tailored to each individual’s clinical needs as determined by a medical team. Analysis focused on endurance, performance on the device, and patient-reported outcomes related to walking fluidity, safety, and satisfaction. Results showed overall improvement in endurance and performance, with the most significant gains observed in participants with thoracic-level injuries. All participants reported increased perceived safety, walking fluidity, and high satisfaction with the training. These findings support the potential of personalized exoskeleton training to enhance outcomes and experiences for individuals with incomplete SCI. The difference in improvement between lesion levels highlights the need for customized approaches to address the diverse clinical conditions within this population.

1. Introduction

Spinal cord injury (SCI) remains one of the most debilitating conditions, impacting individuals’ motor and sensory functions below the level of injury [1,2,3]. Among the challenges posed by SCI, the total or partial loss of gait function represents a significant obstacle to individuals’ mobility, independence, and overall quality of life [4,5,6].

In recent years, the advent of robotic exoskeletons has emerged as a promising intervention for overcoming these limitations [7,8,9,10,11,12,13,14,15,16,17,18]. Exoskeletons can provide varying levels and types of assistance, enabling users to perform repetitive gait cycles overground, which are essential for promoting neuroplasticity and motor recovery [19,20,21]. These devices offer a range of benefits, including facilitating gait training, improving mobility, and reducing secondary complications like pain and spasticity [9,11,22,23,24,25]. Moreover, the use of exoskeletons in rehabilitation has been associated with psychological benefits, such as improved mood and motivation, which are critical for sustained engagement in therapy [26,27].

Existing research on exoskeleton gait training for individuals with SCI has predominantly focused on demonstrating safety [10,13,14,26,28,29] and assessing improvements in endurance [12,15,16], often within relatively small study cohorts, primarily comprising individuals with complete lesions [13,16,18,30]. Moreover, the limited-duration protocols employed in most of these studies fail to capture how participants progressively adapt to using the device over time. In clinical practice, exoskeleton training is a long, dynamic, and individualized process, where therapists continuously adjust assistance levels and modalities to match the participant’s evolving capabilities. This therapist-guided approach is particularly critical for individuals with incomplete SCI, who present with a broader spectrum of retained functions and higher potential for gait recovery [4]. The inherent variability within this population [31] requires a personalized approach to rehabilitation, where training duration and assistance settings are tailored to each individual’s abilities. Injury-specific factors—such as trunk control and upper and lower limb function—vary depending on whether the spinal cord injury is at the cervical, thoracic, or lumbar level. These factors significantly influence how individuals adapt to exoskeleton-assisted gait and progressive reductions in assistance. Despite their relevance for personalized training, these factors remain underexplored in current research.

While a study has investigated the effects of exoskeleton training on a broader population, including both complete and incomplete SCI participants, over an 8-week training period [28], it did not thoroughly characterize the relationship between training progression and lesion, particularly in the incomplete SCI population. Moreover, the clinical relevance of its findings may be limited by the use of a fixed training duration, which did not account for individual progress, potentially leading to unnecessary sessions for some participants and insufficient training for others.

In response to these gaps, our study aims to investigate the on-device effects of exoskeleton gait training in a cohort of 33 individuals with incomplete SCI, encompassing different lesion levels—cervical (C1–C8), thoracic (T1–T12), and lumbar (L1–L5)—who underwent personalized training from 2013 and 2020 in the spinal unit of the Santa Corona Hospital. Our investigation focused on endurance and performance on the device, as well as on patient-reported outcomes related to walking fluidity, safety, and satisfaction. By analyzing how therapist-guided adjustments to assistance type and modality influence progression within the device, we aim to provide a clinically relevant perspective on exoskeleton-assisted gait rehabilitation. Additionally, by examining subgroup responses based on lesion level, we aim to contribute to a more individualized and effective approach to exoskeleton training for individuals with incomplete SCI.

2. Materials and Methods

2.1. Participants

A retrospective analysis was conducted on 41 individuals with incomplete SCI who underwent robot-assisted gait training with the Ekso exoskeleton (Ekso Bionics^TM, version 1.1, Ekso Bionics, Richmond, CA, USA) at the Spinal Cord Unit of the Santa Corona Hospital (Pietra Ligure, SV, Italy) between 2013 and 2020.

Inclusion criteria for the training required participants to meet specific conditions: incomplete SCI with American Spinal Injury Association (ASIA) Impairment Scale B, C, or D; adequate range of motion in the hips (135° flexion to 20° extension), knees (130° flexion to 0° extension), and ankles (10° flexion to 10° extension); sufficient upper limb strength to operate external aids; intact skin condition; the capability to endure standing for a minimum of 30 min. Exclusion criteria comprised individuals with a spasticity level equal to or less than 3 on the Ashworth scale [32], ongoing fractures, diagnosed osteoporosis, cognitive (Mini Mental Score below 26) or communication impairments, height below 1.50 m or above 2 m, or weight exceeding 100 kg.

For this investigation, only participants who completed at least 7 training sessions with the exoskeleton and provided all clinical and demographic information were included in the analysis. Consequently, 8 subjects were excluded: 4 due to insufficient training sessions, and 4 due to missing data. The study thus encompassed 33 subjects (refer to

Table 1. Population characteristics divided by lesion level (n = 33 incomplete SCI participants).

Lesion Level	n. Participants (% of Total)	Age Median (IQR)	Male (n)	Female (n)	ASIA B (n)	ASIA C (n)	ASIA D (n)
Cervical (C1–8)	9 (27.27%)	54 (41.2–60.3)	3	6	1	7	1
Thoracic (T1–12)	14 (42.42%)	54 (38.4–62.0)	5	9	1	11	2
Lumbar (L1–5)	10 (30.30%)	41.5 (32.6–56.8)	4	6	0	9	1

for information on the subjects included in the investigation).

The study was conducted in accordance with the Declaration of Helsinki (2013) and approved by the local ethical committee (Comitato Etico Territoriale—Liguria: 353/2024—DB id 14082). Informed consent was obtained from all subjects involved in the study to analyze the data for research purposes and publish the results in an anonymous form.

2.2. Ekso Exoskeleton

The Ekso exoskeleton (Ekso Bionics^TM, version 1.1, Ekso Bionics, Richmond, CA, USA, Figure 1a), utilized for the robotic exoskeleton-assisted gait training, is a wearable bionic suit powered by batteries, commonly employed in clinical environments as a gait training tool for individuals with moderate to severe neurological conditions [33,34,35].

The exoskeleton features powered joints at the hip and knee, alongside semi-rigid passive ankle joints. It is also equipped with 15 position sensors, adjustable hip, upper and lower legs components, and it is mandatorily accompanied by external balancing aids, such as walkers and crutches.

The device provides the option to select between bilateral or unilateral assistance modes and allows the specification of how each leg is powered and controlled, with three assistance setting options available: fixed assistance, adaptive assistance, and no assistance.

Under the bilateral assistance mode (BILAT), both legs receive stance support and swing assistance (either adaptive or fixed). Conversely, the unilateral assistance mode offers stance support and swing assistance solely to a specified limb (left limb assisted [L.ASS] or right limb assisted [R.ASS]), granting freedom of movement to the opposite leg.

Within the assistance settings options:

• The “fixed assistance” (FIXED) setting allows for the adjustment of the amount of swing assistance provided during the swing phase. At 100% (MAX), the user receives full assistance, enabling walking without voluntary contribution, whereas at 0%, no assistance is provided.
• The “adaptive assistance” (ADAPT) setting dynamically adjusts the amount of swing assistance based on the performance during walking.
• The “no assistance” (FREE) option offers no assistance while still compensating for the inertia of the device. With this setting, the user can walk freely without feeling the weight of the exoskeleton.

The Ekso exoskeleton also provides three step-initiation programs:

• FirstStep. The stepping action is controlled by the physical therapist through a button push.
• ActiveStep. The user initiates the stepping action via an interface situated on the balancing aids.
• ProStep. Each step is prompted by a forward-lateral movement of the user’s hip.

2.3. Exoskeleton-Assisted Gait Training

Prior to beginning the training, an Ekso-certified therapist adjusted the exoskeleton to match the participant’s anthropometric measures: the hip width and the lengths of the upper and lower legs (Configuration phase, Figure 1b).

After this adjustment, there was a 5-min familiarization phase (Figure 1b). During this phase, the participant became accustomed to the device under simplified conditions, using the walker and the FirstStep program with MAX BILAT assistance.

Following the familiarization phase, the multi-day training (Figure 1b) with the exoskeleton started. At the beginning of each training session, an Ekso-certified therapist adjusted the device and training settings based on individual performance. Adjusting the device settings included tuning the step length, step height, swing speed, and the forward-lateral shift needed to start the steps using the ProStep program. Training settings involved selecting the modality (BILAT, L. ASS or R. ASS) and type of assistance (MAX, FIXED, ADAPT, or FREE) according to each participant’s rehabilitative goals and level of impairment. Bilateral assistance was used when both limbs required comparable support to maintain symmetrical gait, while unilateral assistance was applied when one limb was more impaired, allowing focused support to restore gait balance. Assistance type was selected according to motor capacity: MAX provided full torque for participants with minimal voluntary control; FIXED delivered a constant level of support for those requiring consistent movement guidance; ADAPT adjusted torque in response to deviations from target gait parameters, promoting active engagement; and FREE, typically used unilaterally on the less impaired side, removed active support to encourage independent limb control. For all participants, initial training sessions involved using a walker, progressing to crutches as the patient’s proficiency with the exoskeleton improved. The duration of each session ranged from 15 to 50 min, depending on observed fatigue and endurance. The number of sessions varied based on the individual rehabilitative needs of each participant.

2.4. Outcomes Measures

The impact of the exoskeleton-assisted training was evaluated using a combination of objective endurance and performance metrics, as well as subjective self-assessment measures (Figure 2).

Assessment of walking endurance and performance with the exoskeleton occurred at three time points to track the participants’ progress throughout the training: at baseline (T1), at the sixth session (T6), and upon completion of the multisession gait-assisted training (TE), which, though varying among participants, corresponded to the session wherein participants achieved the rehabilitation goal established by clinicians (range 7–40 sessions). Data from the exoskeleton’s onboard computer allowed for the evaluation of endurance with the following metrics: the total number of steps taken (number of steps, NS), total walking time with the device (walking time, WT, minutes), and total time spent standing up with the device (up time, UT, minutes).

Evaluation of walking performance with the exoskeleton was based on two parameters: the average distance walked in 6 min (M6m, meters) and the average time employed to walk 10 m (10MTt, minutes). Both metrics were inspired by the 6-min walk test [36] and the 10-m walk test [37], two widely used clinical tests, although we used different conditions and rules. More specifically M6m was calculated as

$$\mathrm{M}6\mathrm{m}=\left({\displaystyle \frac{\mathrm{N}\mathrm{S}\mathrm{ }·\mathrm{S}\mathrm{L}}{\mathrm{W}\mathrm{T}}}\right)\mathrm{ }·\mathrm{ }6\mathrm{ }\mathrm{m}\mathrm{i}\mathrm{n}$$

(1)

with SL indicating the step length in meters set by the physician and imposed by the exoskeleton during walking.

Similarly, 10MTt was calculated as

$$10\mathrm{M}\mathrm{T}\mathrm{t}=\mathrm{ }\left({\displaystyle \frac{\mathrm{W}\mathrm{T}}{\mathrm{N}\mathrm{S}\mathrm{ }·\mathrm{S}\mathrm{L}}}\right)\mathrm{ }·\mathrm{ }10\mathrm{ }\mathrm{m}$$

(2)

We also recorded self-assessment measures to incorporate the perspectives and needs of the participants. Participants provided feedback on their perception of safety and walking fluidity when using the device at the beginning (T1) and conclusion (TE) of the multisession gait-assisted training. At the end of the training (TE), they also rated their overall satisfaction with the treatment. All self-assessment measures utilized a numeric rating scale (NRS) ranging from 0 to 10.

2.5. Statistical Analysis

The effects of multi-day exoskeleton-assisted gait training were assessed across the entire cohort, with participants categorized by lesion level: cervical (C1–C8), thoracic (T1–T12), and lumbar (L1–L5).

To test the hypothesis that repeated training would lead to improvements in walking endurance (NS, WT, and UT) and performance (M6m and 100MTt), a mixed-design ANOVA was conducted with “Training session” (T1, T6 and TE) as the within-subject factor and “Lesion level” (cervical, thoracic and lumbar) as the between-subject factor. Assumptions of normality and sphericity were assessed using the Shapiro–Wilk test and Mauchly’s test of sphericity, respectively. Both assumptions were satisfied for all performance and endurance measures.

Self-assessment measures did not meet the normality assumption, as evaluated by the Shapiro–Wilk test, and were therefore analyzed using non-parametric tests. To assess whether training improved perceived safety and walking fluidity within each lesion level group, a Friedman ANOVA was conducted to compare scores between T1 and TE. To evaluate whether satisfaction with exoskeleton-assisted gait training differed by lesion level, satisfaction scores were compared among the three groups using the Kruskal–Wallis ANOVA.

All statistical analyses were conducted within the Jamovi environment (Jamovi software version 0.9.2.8) with statistical significance set at p < 0.05. A Bonferroni correction was applied to correct for multiple comparisons in all post hoc analyses. Post hoc analyses of the Friedman ANOVA test were conducted using the Durbin–Conover test, while those of the Kruskal-Wallis ANOVA test were conducted using the Dwass–Steel–Critchlow–Fligner test. Effect sizes and confidence intervals were reported alongside p-values for all relevant comparisons.

3. Results

Before presenting the main outcomes, we examined the structure and progression of the exoskeleton-assisted gait training across lesion levels. While the number of training sessions did not differ significantly between groups (median (IQR): cervical lesion 13 [10.98–15.90], thoracic lesion 16.5 [11.89–23.96], and lumbar lesion 12 [8.72–20.47]), notable differences were observed in how assistance was adapted over time.

To capture this progression, we report in

Table 2. Percentages (%) of assistance type-modality within each lesion group at T1, T6, and TE.

					Modality of Assistance
Lesion Level	Session	Type of Assistance			Left Leg				Right Leg
		BILAT (%)	L.ASS (%)	R.ASS (%)	MAX (%)	ADAPT (%)	FIXED (%)	FREE (%)	MAX (%)	ADAPT (%)	FIXED (%)	FREE (%)
Cervical	T1	100.0	0.0	0.0	66.7	33.3	0.0	0.0	66.7	33.3	0.0	0.0
	T6	77.8	0.0	22.2	44.4	33.3	0.0	22.2	44.4	44.4	11.1	0.0
	TE	77.8	0.0	22.2	22.2	44.4	11.1	22.2	22.2	44.4	33.3	0.0
Thoracic	T1	100.0	0.0	0.0	68.8	31.2	0.0	0.0	68.8	31.2	0.0	0.0
	T6	93.8	0.0	6.2	25.0	43.8	25.0	6.2	25.0	50.0	25.0	0.0
	TE	81.2	6.2	12.5	12.5	50.0	25.0	12.5	12.5	50.0	31.2	6.2
Lumbar	T1	100.0	0.0	0.0	35.7	50.0	14.3	0.0	35.7	50.0	14.3	0.0
	T6	85.7	7.1	7.1	21.4	35.7	35.7	7.1	14.3	42.9	35.7	7.1
	TE	85.7	7.1	7.1	14.3	42.9	35.7	7.1	7.1	50.0	35.7	7.1

the percentages of assistance type (BILAT, L.ASS, R.ASS) and modality (MAX, ADAPT, FIXED, FREE) within each lesion-level group at T1, T6, and TE.

At T1, all participants began with bilateral assistance and predominantly MAX support. However, lumbar participants showed greater initial variability, with higher use of ADAPT and FIXED modes, suggesting better baseline motor control.

By T6, participants across all groups began transitioning to more personalized settings. Bilateral assistance decreased, particularly in the lumbar and cervical groups, with a shift toward unilateral assistance modes. The use of ADAPT and FIXED modalities increased, reflecting more active engagement in gait. In some cases, the FREE setting was introduced—typically unilaterally—to encourage greater voluntary control of the less impaired limb while maintaining support on the more affected side.

By the final session (TE), these trends became more pronounced. Bilateral assistance further declined, and the proportion of participants using ADAPT, FIXED, or FREE settings increased across all groups. The lumbar group showed the most substantial shift toward reduced assistance, consistent with their higher functional capacity.

3.1. Exoskeleton-Assisted Gait Endurance

The investigation into the effects of exoskeleton-assisted gait training revealed a gradual and consistent improvement in endurance on the device across training sessions (Figure 3).

A significant main effect of “Training session” was observed in all lesion levels: WT: F(2,58) = 60.1, p < 0.001, η²_p = 0.675, 95% CI [0.378, 0.630]; UT: F(2,58) = 30.1, p < 0.001, η²_p = 0.510, 95% CI [0.210, 0.483]; NS: F(2,60) = 71.0, p < 0.001, η²_p = 0.703, 95% CI [0.416, 0.655]. Post hoc comparisons showed that walking time and the number of steps increased significantly across sessions (post hoc WT: T1–T6, p < 0.001; T6–TE, p = 0.024; post hoc NS: T1–T6, p < 0.001; T6–TE, p = 0.004). Uptime also increased significantly until the sixth session (post hoc UT: T1–T6, p < 0.001), then remained stable through the end of training. No significant differences were found between lesion levels, and no interaction effects were observed.

3.2. Exoskeleton-Assisted Gait Performance

In addition to gains in endurance, participants exhibited significant improvements in functional gait performance throughout the training period (Figure 4).

All participants progressively increased their distances in the M6m, with a significant main effect of “Training session” (F(2,50) = 12.02, p < 0.001, η²_p = 0.325, 95% CI [0.080, 0.349]). Post hoc comparisons revealed steady gains between time points (post hoc T1–T6: p = 0.02; T1–TE: p = 0.001; T6–TE: p = 0.03). In parallel, 10MTt performance improved significantly, reflected in a decrease in completion time over the course of training (training session effect: F(2,50) = 12.06, p < 0.001, η²_p = 0.325, 95% CI [0.080, 0.349]). Post hoc analyses confirmed a significant reduction by T6, which was sustained through the final evaluation (post hoc T1–T6: p = 0.006; T1–TE: p = 0.002).

Importantly, a significant interaction between “Training session” and “Lesion level” emerged for both metrics, with moderate-to-large effect sizes (M6m: F(4,50) = 2.84, p = 0.03, η²_p = 0.185, 95% CI [0.018, 0.144]; 10MTt: F(4,50) = 3.90, p = 0.008, η²_p = 0.238, 95% CI [0.027, 0.169]). Post hoc analysis showed that participants with thoracic lesions exhibited the most substantial improvements: they consistently reduced their 10MTt times across sessions (Post hoc T1–T6: p = 0.02; T1–TE: p = 0.002; T6–TE: p = 0.04) and significantly increased their M6m distance from T1 to TE (Post hoc T1-TE p = 0.006). In contrast, no meaningful performance changes were observed over time among participants with cervical or lumbar lesions.

3.3. Safety, Walking Fluidity, and Satisfaction Perception with Exoskeleton

At the end of the first training session (T1), participants across all lesion levels, cervical, thoracic, and lumbar, reported low levels of walking fluidity and perceived safety while using the exoskeleton (

Table 3. Self-assessment measures of walking fluidity, perceived safety, and satisfaction (NRS, [0–10]) reported by incomplete SCI participants divided by lesion level (cervical, thoracic, and lumbar). Results are presented as median (IQR).

Lesion Level	Walking Fluidity (0–10)		Safety (0–10)		Satisfaction (0–10)
	T1	TE	T1	TE
Cervical	4.0 (3.8–5.0)	7.0 (6.6–7.9)	9.0 (8.5–9.4)	5.0 (4.6–5.6)	8.0 (6.9–8.2)
Thoracic	5.0 (3.8–6.1)	7.5 (6.8–8.5)	9.0 (8.3–9.6)	5.5 (4.0–6.8)	8.0 (7.3–9.0)
Lumbar	5.0 (3.8–5.6)	7.5 (6.5–7.9)	9.0 (8.4–9.8)	5.5 (3.7–6.1)	8.0 (6.9–8.3)

). However, these perceptions improved significantly by the end of the training program (Training session T1-TE: p < 0.001), with no significant differences observed between groups. Similarly, by the conclusion of the training, participants from all lesion level categories reported high levels of satisfaction with the exoskeleton experience.

4. Discussion

This study investigated the effects of exoskeleton-assisted gait training in individuals with incomplete SCI, with a specific focus on how structured training influenced their ability to use the device over time. We evaluated changes in endurance, walking performance, and self-perceived outcomes during exoskeleton use and explored whether these outcomes differed according to lesion level: cervical, thoracic, or lumbar.

Compared to previous research, our study offers a broader and more clinically relevant perspective [8,38]. Prior investigations have often relied on small sample sizes—typically fewer than 10 participants [11,23,39,40]—and employed short, fixed-duration training protocols. While some studies, such as Gorgey et al. [23], implemented longer programs, they were limited to case series designs and did not classify results by neurological level. Moreover, previous research has primarily emphasized general functional outcomes rather than examining how repeated training affects participants’ progressive ability to operate the device. In contrast, our study included 33 individuals with cervical, thoracic, or lumbar injuries and applied a flexible, personalized training approach. This design allowed for a detailed analysis of training progression across subgroups within a realistic clinical rehabilitation setting.

All participants, regardless of lesion level, demonstrated consistent improvements in endurance throughout the training period. Walking time, upright time, and step count per session increased steadily, indicating improved tolerance to upright activity and greater engagement with the exoskeleton. These results are comparable to or exceed those reported in previous literature. For example, Gorgey et al. [23] tested four individuals with SCI, one with an incomplete injury, over 10 to 15 sessions and observed an increase in step count from 59 to 2284 per session, along with walking time increasing from 12 to 57 min using a similar exoskeleton.

These endurance gains are particularly meaningful given the known physiological benefits of upright mobility. Increased time spent walking or standing has been associated with improved cardiovascular health, spasticity management, and prevention of complications such as pressure ulcers [9,41]. Although minimal clinically important difference (MCID) thresholds for exoskeleton-based endurance measures have not been formally established, the magnitude of the gains observed suggests clinical relevance.

With respect to walking performance, all participants exhibited improvements while using the exoskeleton, though the extent of change varied by lesion level. The most pronounced gains were seen in individuals with thoracic injuries. After converting 10-m travel time (10MTt) data to walking speed, thoracic participants improved by 0.07 m/s, exceeding the 0.06 m/s threshold generally accepted as clinically meaningful in the 10-m walk test [42]. Lumbar participants improved by 0.05 m/s, and cervical participants by 0.03 m/s. Similarly, in the meters in 6 min (M6m) parameter, thoracic participants improved by 24.3%, surpassing the 22% MCID range proposed for the 6 min walking test [43]. In comparison, lumbar participants improved by 15.8%, while cervical participants showed only a 0.99% gain. Importantly, all performance metrics were collected during exoskeleton-assisted walking; comparisons to MCID thresholds derived from overground tests are therefore approximate and serve only as contextual references.

Our observation that participants with thoracic-level injuries achieved the most significant performance improvements represents a novel contribution to the literature. Although general gains in exoskeleton performance are well-established, few studies have directly compared outcomes by injury level. In our cohort, thoracic-level participants often retained sufficient trunk and upper limb control to interact effectively with the device, particularly during tasks requiring deliberate weight shifting for step initiation. This enabled them to gradually reduce assistance levels over time—transitioning from high-support assistance settings (e.g., FIXED) to more adaptive or minimal assistance settings (e.g., ADAPT or FREE). This increased engagement likely facilitated greater functional gains.

In contrast, cervical-level participants faced more extensive motor impairments, including reduced trunk and upper limb control. These limitations made it more difficult to use the exoskeleton’s step initiation system, which relies on lateral weight shifts. As a result, most cervical participants remained in high-assistance modes throughout training. While this allowed safe gait practice, it constrained opportunities for active engagement and limited the potential for significant improvement.

Participants with lumbar-level injuries, despite having better baseline motor function, showed less dramatic improvements than those with thoracic injuries. This may be due to a ceiling effect: many of these individuals were already operating near the performance limits of the device.

Device-related factors likely contributed to these differences. The exoskeleton used in this study is optimized for users with lower-limb impairments and relatively preserved trunk function, making it particularly appropriate for those with thoracic-level injuries. For highly mobile lumbar participants, the device may not have offered a sufficient challenge to drive further improvement, while for more severely impaired cervical participants, it may not have provided adequate compensatory support. These mechanical characteristics likely contributed to the performance change observed across lesion levels.

Beyond the objective improvements in endurance and performance, self-reported outcomes confirmed the value of exoskeleton training across all lesion levels [28,29,44,45]. By the end of the program, participants consistently reported better walking fluidity, increased safety, and high satisfaction with the training, regardless of their neurological level. This is notable, as it contrasts with the clear differences seen in physical performance, particularly the greater gains in the thoracic group. Surprisingly, participants with more severe impairments, such as those with cervical injuries, reported similar positive experiences. This suggests that the psychological and motivational impact of upright mobility is substantial, even when physical progress is limited. The ability to stand and walk (even with assistance) appears to enhance confidence, autonomy, and emotional well-being, making exoskeleton training beneficial beyond its physical effects.

5. Conclusions

Our study offers insight into how personalized exoskeleton-assisted gait training is applied in routine clinical settings and how individuals with incomplete SCI respond to this approach over time. By focusing specifically on performance within the device, we examined the impact of clinician-guided adjustments to assistance type, modality, and training intensity, adapted to each participant’s functional capacity and rehabilitation goals.

Although retrospective in design, our approach allowed us to include a large and diverse group of participants, representative of real-world rehabilitation practice. Participants were analyzed by lesion level to explore group-specific responses; however, some clinical variability remained within groups, reflecting the heterogeneity commonly seen in individuals with SCI. This variability should be taken into account when interpreting the results, particularly with respect to lesion-level patterns. Nonetheless, broad inclusion enabled us to observe how individuals with different lesion levels responded to a fully personalized training strategy—an aspect that has not been studied before.

Our findings suggest that individuals with thoracic-level injuries may be particularly well-suited to this form of training, likely due to a favorable balance between retained motor function and compatibility with the device’s support capabilities. In contrast, participants with cervical or lumbar injuries may require modified approaches, alternative assistance strategies, or complementary therapies to achieve comparable benefits.

In addition to objective improvements, participants reported highly positive self-assessment scores at the end of the training program. These subjective outcomes, observed across all lesion levels, point to the broader experiential value of the intervention. However, because our study collected self-assessment data only before and after training, it remains unclear how perceptions evolve during the intervention. Future work could benefit from integrating qualitative methods, such as interviews or open-ended questionnaires, to better capture users’ changing experiences and expectations across sessions.

It is important to note that all outcome measures in our study were collected during exoskeleton-assisted training. We observed measurable improvements in performance and endurance while participants were using the device. However, we did not assess participants outside the exoskeleton or after the intervention period. Therefore, we cannot determine whether these gains would be present during unassisted walking. Based on principles of motor learning and task-specific rehabilitation, it is plausible that the observed improvements in endurance and task performance could carry over to unassisted walking immediately after training—particularly in participants with thoracic-level injuries who progressed to lower assistance levels and demonstrated the greatest in-device improvements [39,45].

Over the longer term, retention of these gains may depend on whether the training induced durable neuromuscular adaptations, which is plausible given prior findings in repetitive robotic gait training [46,47]. It also remains unknown whether participants would retain their ability to use the exoskeleton at the same performance level after a period of non-use. Some performance decline could be expected without continued exposure, as reported in other robotic training contexts [8]. However, individuals with greater autonomy during training—particularly those with thoracic or lumbar lesions—may be more likely to retain device-specific motor strategies.

Understanding whether improvements are maintained—either off or on the device—is clinically important for informing post-training care. If performance diminishes without regular practice, periodic “booster” sessions may be needed to sustain gains. Conversely, if improvements are retained, therapy frequency could be reduced or shifted to other modalities. Future studies should include out-of-device assessments, long-term follow-up, and appropriate control groups to determine the durability and generalizability of performance and endurance improvements observed during exoskeleton training.

Overall, our findings support the potential of personalized exoskeleton gait training as a meaningful and adaptable intervention for individuals with incomplete SCI. Future studies should build on these results by including longer-term follow-up, combining exoskeleton use with other rehabilitation strategies or technologies, and collecting more detailed user feedback. These steps will be important to improve the clinical use of exoskeletons and maximize their therapeutic impact in neurorehabilitation.