This retrospective study enrolled 1263 patients with unilateral lower limb hemiplegia secondary to acute stroke who underwent rehabilitation in the Department of Rehabilitation Medicine, Shanghai Tianyou Hospital, China, between January 2024 and May 2025. Based on the rehabilitation modality, patients were divided into a repetitive transcranial magnetic stimulation (rTMS) group, a lower limb rehabilitation robot (UGO) group, and a repetitive transcranial magnetic stimulation combined with lower limb rehabilitation robot (rTMS+UGO) group.

Inclusion criteria for patient selection were as follows: (1) diagnosis consistent with established stroke criteria [15], with unilateral limb hemiplegia and confirmed by neuroimaging; (2) clinically stable, with a disease course between 2 weeks and 1 year; (3) hemiplegic clinical phenotype with lower limb Brunnstrom stage I–V; (4) modified ashworth scale (MAS) score $\le$2, with patients willing to cooperate with treatment; and (5) absence of other acute diseases or severe complications.

However, individuals were excluded from the study cohort if they had (1) progressive stroke or subarachnoid hemorrhage; (2) severe systemic diseases within the past year, such as organ failure (heart, lung, liver, kidney), active infection, diabetes, or inadequately controlled hypertension; (3) severe hip and knee osteoarthritis, lower limb joint diseases, rheumatoid arthritis, ischemic femoral head necrosis, or other conditions affecting body movement or posture control; (4) history of primary or secondary epilepsy; (5) severe cognitive dysfunction; (6) history of metal implants such as pacemakers, cardiac or intracranial stents, or other implanted devices; (7) long-term use of therapeutics that affect neuro-excitability (such as antidepressants and antiepileptic drugs); and (8) history of undergone knee surgery or severe ligament injury.

Following these predetermined inclusion-exclusion criteria, a total of 158 patients were included in this retrospective study (Fig. 1).

Fig. 1.

A flow chart of the study participant selection. rTMS, repetitive transcranial magnetic stimulation; UGO, lower limb rehabilitation robot; rTMS+UGO, repetitive transcranial magnetic stimulation combined with lower limb rehabilitation robot; MAS, modified ashworth scale.

According to the treatment received, study participants (n = 158) were divided into three groups: rTMS, lower limb rehabilitation robot (UGO), and rTMS+UGO groups. All patients received routine comprehensive rehabilitation which included: (1) Muscle stretching training: using an ankle correction plate to stretch the affected lower limbs; (2) Strength and muscle endurance training of knee flexor and extensor muscles; (3) Knee control training: knee extension control training in supine position and knee flexion control training in prone position within a 0°–30° range; (4) Single-leg weight-bearing, balance training, wall squat exercises, gait training, stairs climbing practice; (5) Proprioception training: improving sensory input using Rood therapy.

All these interventions were performed once daily, 6 days a week, over 4 weeks.

Experimental-group-specific interventions were performed as follows:

(a) rTMS group: In addition to standard rehabilitation, the rTMS group was treated with navigated rTMS once daily. A repetitive transcranial magnetic stimulator (OSF-4/T, Wuhan Oseff Medical Company, Wuhan, China) was used. We used a monopulse TMS with an “8” shaped coil to stimulate the M1 leg region. Resting motor threshold (RMT) was determined by gradually reducing the intensity of stimulation until a motor evoked potential of at least 50 µV amplitude was elicited in the relaxed tibialis anterior muscle on the unaffected side in at least 5 out of 10 trials [16]. During treatment, the patient sat wearing a positioning cap, and the “8” shaped coil was placed over the lower limb motor cortex representative area on the healthy hemisphere. The stimulation parameters were as follows: frequency 1 Hz, intensity 80% RMT, stimulation time 15 seconds, inter-training interval 10 seconds, and a total of 1395 pulses per session. Treatment was administered once a day, 5 days a week, for 4 consecutive weeks.

(b) UGO group: Based on routine rehabilitation, patients in the UGO group received gait training with the UGO lower limb rehabilitation robot (Hangzhou Chengtian Technology Development Co., Ltd., Hangzhou, China). The robot is equipped with a fixed belt for the patient’s waist that can provide 0–50 kg of dynamic body-weight support while allowing free pelvic movement. Before training, the therapist transferred the patient to a robot chair, measured the length of both thighs, calves, and lower limbs, and fitted a weight-support vest and connecting straps. The robot was adjusted to the sitting position, and the length of the thigh and calf were adjusted to match the patient so that the hip and knee joints aligned with the robot’s corresponding joints. The patient’s torso, thigh, and feet were then fixed to the device.

The robot was activated to bring the patient to standing, followed by 1–2 minutes of walking under body-weight support. After examining all the equipment, the therapist selected a passive or active mode based on the patient’s functional status. Passive mode was applied when the patient was unable to overcome gravity to actively extend the knee or when significant fatigue or poor attention prevented high-quality active movement. If the patient could generate certain active movement, the mode was switched to active as soon as they were able to push the leg under weight support. Subsequently, the robot was lowered so that the feet touched the ground, and walking training began. Each training session lasted 20 minutes, 5 times per week, for 4 weeks. They were monitored continuously; if substantial discomfort was reported, training was stopped immediately, the patient’s condition was evaluated, the reason was analyzed, and the training plan was adjusted accordingly.

(c) The UGO+rTMS group: In this group, patients were treated with rTMS once daily in addition to routine rehabilitation, applying the same procedure as in the rTMS group. After a 30-minute interval, UGO training was conducted for 20 minutes using the same protocols as in the UGO group. Training was conducted 5 times per week, for a total of 4 weeks.

All patients were evaluated by an independent therapist, blinded to group allocation and did not participate in the treatment delivery, at baseline, two weeks after the start of treatment, and again at four weeks. The primary evaluation indicators were as follows.

(a) Brunnstrom scale: The Brunnstrom scale was used to evaluate muscle tension, joint movement, and the ability to perform independent movement. The comprehensive score ranges from stage I to VI [17].

• Stage I: No voluntary movement is observed.

• Stage II: Only minimal, involuntary or synergistic movements appear.

• Stage III: Coordinated flexion of the hip, knee, and ankle can be elicited in both sitting and standing positions.

• Stage IV: In the sitting position, the patient can flex the knee more than 90° so that the foot slides under the chair, and can perform ankle dorsiflexion without lifting heel off the ground.

• Stage V: In the leg-strengthening position, the affected limb can first flex the knee and then extend the hip; with the knee extended and body weight shifted to the unaffected leg, ankle dorsiflexion can be performed.

• Stage VI: In the standing position, the hip can be abducted beyond the range achievable by lifting the ipsilateral pelvis; in sitting, the knee can be extended while the lower limb performs internal and external rotation, allowing combined inversion and eversion movements of the foot.

(b) Fugl-Meyer Assessment for Lower Extremity (FMA-LE) score: The FMA-LE score was used to evaluate the recovery of lower limb motor function [18]. The scale has a total of 34 points, with higher scores indicating better lower limb motor function.

(c) Modified Barthel Index (MBI): The MBI was used to evaluate the patient’s independence in daily life activities [19]. The total score ranges from 0 to 100 points, with higher scores indicating greater functional independence.

(d) Berg balance function scale: The Berg balance scales were used to evaluate the balance performance of patients [20]. The maximum possible score was 56 points, with higher scores reflecting better balance ability.

Knee hyperextension angle measurement: Knee hyperextension was recorded using a disc goniometer. A normal range is considered to be 0° to 5°. A value exceeding 5° was defined as hyperextension, with a larger angle indicating more severe deformity [21].

Surface electromyography: The average electromyographic (AEMG) of the vastus medialis, rectus femoris, hamstrings, and gastrocnemius of both the affected and unaffected sides was measured using an electromyography evoked potential system (NDI-094, Suzhou Haishen Medical Device Associates Co., Ltd., Suzhou, China) at room temperature. Patients were seated with both hips and knees flexed and the ankles fixed in a neutral 90° position. After routine skin preparation with alcohol, electrodes were placed over each target muscle (vastus medialis, rectus femoris, hamstring, and gastrocnemius). Surface electromyography signals were acquired with the NDI-094 instrument. The conductive area of each electrode was 0.5 cm in diameter, with an inter-electrode distance of 2 cm, and the electrodes were aligned parallel to the muscle.

During testing, surface electromyography signals were recorded while patients stood up without assistance and maintained a stable standing posture for 5 seconds, sitting down independently, performing maximum bilateral knee extension and flexion efforts, and maintaining each for five seconds. Patients were instructed to alternate between relaxed states and maximal voluntary isometric contraction (MVIC), and surface electromyography signals were recorded in both conditions. Each operation was repeated three times, and the AEMG during MVIC was calculated from these repetitions, with the average of the three trials used for analysis.

Isokinetic muscle strength testing: The peak torque (PT), peak torque relative to body weight (PT/W), and total work (TW) of the affected knee were evaluated using an A8 isokinetic excitation testing system (Guangzhou Yikang, Guangzhou, China). During testing, patients were seated with the length of the seat and the backrest adjusted so that the thigh lay flat on the seat, and the back was in full contact with the backrest. The backrest angle was set to a comfortable position, ensuring alignment between the dynamometer axis and the anatomical axis of knee joint rotation. The trunk was stabilized with a seat belt and shoulder harness, and the proximal thigh of the tested limb was fixed with a nylon strap. The resistance pad connected to the dynamometer lever arm was positioned 3 cm above the medial malleolus of the affected leg, with the dynamometer axis aligned horizontally with the lateral femoral condyle. Before testing, the knee joint and ankle joint were passively mobilized to minimize the risk of injury, followed by three low-resistance preheating training sessions. The isokinetic dynamometer was set to an angular velocity of 60°/s. Patients were then instructed to extend and flex the knee joint with the maximum force, performing 5 repetitions per set for 3 sets, and the average value was used for analysis.

Adverse event assessment: Throughout the treatment period and after interventions, all adverse events were systematically monitored and recorded. Adverse reactions, such as spasm, fatigue, and rTMS-related headache, were recorded in both groups of patients.

Statistical analyses were conducted using SPSS version 27.0 (IBM Corp., Armonk, NY, USA). The Shapiro-Wilk was used to assess the normality of data distribution. Measurement data conforming to a normal distribution were expressed as mean $\pm$ standard deviation (mean $\pm$ SD). Non-normally distributed continuous variables were expressed as Median (Q1, Q3), with intergroup comparisons performed using the Kruskal-Wallis H test. When the Kruskal-Wallis H test showed significant differences, post-hoc pairwise comparisons were conducted with the Mann-Whitney U test, and p-values were adjusted using the Bonferroni correction.

Categorical variables were expressed as frequencies and percentages (%). When the number of cases exceeded 5, the $\chi$² test was used for comparison between groups; otherwise, the Fisher exact test was applied, and Generalized estimating equations were used for pairwise comparison of time points within the group, and the p-value was corrected by Bonferroni method. Repeated measures analysis of variance (ANOVA) was used to evaluate measurement data collected at multiple time points across groups. If a significant time $\times$ group interaction was observed, simple effects analysis was conducted to identify the source of interaction. Within-group pairwise comparisons between time points were performed using paired t-tests with Bonferroni correction, and between-group comparisons at the same time point were performed using independent t-tests, also with Bonferroni adjustment. The corrected significance level for multiple pairwise comparisons was set at $\alpha$ = 0.0167 ($\alpha$ = 0.05/3). A p-value 0.05 was considered statistically significant.

The median age of patients was 58 (53, 64) years in the rTMS group, 61 (54, 66) years in the UGO group, and 60 (56, 60) years in the rTMS+UGO group, with no significant age difference observed among the three groups (p $>$ 0.05). The median disease duration was 4 (2, 5) months, 4 (2, 5) months, and 4 (2, 7) months, respectively, with no significant difference among the three groups (p $>$ 0.05). Furthermore, there were no significant differences in gender, stroke type, paretic side, or the prevalence of hypertension, diabetes, hyperlipidemia, and heart disease (p $>$ 0.05). Baseline characteristics of stroke patients are summarized in Table 1.

The changes in Brunnstrom staging in the three groups before and after treatment are presented in Table 2. There was no significant difference in Brunnstrom staging among the groups before or after 2 weeks of treatment. However, by week 4, the rTMS+UGO group demonstrated a significant improvement over the rTMS group (p 0.05). Furthermore, longitudinal comparisons within each group, with Bonferroni-adjusted multiple comparisons, revealed that the Brunnstrom stage improved significantly at both 2 and 4 weeks compared with baseline in all three groups. Particularly, the proportion of patients in stage II decreased, whereas the proportion in stage III and higher stages increased over time.

The lower limb FMA-LE scores for the three groups at each assessment point are summarized in Table 3. Before treatment, there were no significant differences in FMA-LE score among the groups (p $>$ 0.05). The intra-subjects analysis showed a significant main effect of time (F = 511.149, p 0.001), indicating substantial improvement in lower limb motor function after intervention. The inter-subjects analysis revealed a significant main effect of group (F = 3.897, p = 0.022). Moreover, the time $\times$ group interaction was also significant (F = 10.158, p 0.001). Additionally, at both 2 and 4 weeks, the increase in FMA-LE score in the rTMS+UGO group was significantly greater than in either the rTMS or UGO group (p 0.05).

The changes in balance function among the three groups over time are presented in Table 4. Before treatment, there was no significant difference in the Berg scores among the three groups (p $>$ 0.05). However, the Berg scores had a significant main effect of time (F = 858.145, p 0.001), indicating that balance function improved significantly across the treatment period in all groups. Furthermore, a significant main effect of group (F = 24.478, p 0.001) and a substantial time $\times$ group interaction effect (F = 7.691, p 0.001) were also found. Post hoc analysis revealed that, at both 2 and 4 weeks, the Berg scores in the rTMS+UGO group were significantly higher than in the rTMS and UGO groups alone.

As shown in Table 5, in the baseline, the knee hyperextension angle did not differ among the three groups (p $>$ 0.05). However, the pre- and post-treatment in the knee hyperextension angle showed statistically significant differences, indicating a substantial time effect (F = 495.044, p 0.001), with progressive reduction in hyperextension angle of all three groups during the treatment period. Furthermore, there was a significant between-group effect (F = 3.927, p = 0.022), and substantial time $\times$ group interaction (F = 5.759, p 0.001), indicating that the degree of improvement differed across the three groups. After 4 weeks of treatment, the knee hyperextension angle in the rTMS+UGO group was significantly lower than in the rTMS and UGO groups (p 0.05).

Activities of daily living were assessed using the MBI score (Table 6). There was no significant difference in baseline MBI scores among the three groups (p $>$ 0.05). The difference in pre- and post-treatment MBI was statistically significant (time effect: F = 727.850, p 0.001), indicating that MBI scores increased progressively in all three groups. There was a substantial between-group effect (F = 4.804, p = 0.009) and a marked time $\times$ group interaction (F = 5.096, p = 0.001), indicating differential improvement among the groups. After 4 weeks of treatment, the increase in MBI score in the rTMS+UGO group was significantly higher than in the rTMS and UGO groups (p 0.05).

Table 7 shows that, before treatment, there were no significant differences in PT, PT/W, and TW for knee extension among the three groups (p $>$ 0.05). During knee extension, PT, PT/W, and TW showed significant time effects (F = 271.591, p 0.001; F = 172.662, p 0.001; F = 642.075, p 0.001), indicating a progressive increase in these indicators over the treatment period. Similarly, substantial group effects (F = 5.778, p = 0.004; F = 3.263, p = 0.041; F = 3.899, p = 0.022) and marked time $\times$ group interactions (F = 115.771, p 0.001; F = 3.759, p = 0.007; F = 28.338, p 0.001) were also observed, indicating that the degree of improvement differed among the groups. After 2 weeks of treatment, PT and TW in the rTMS+UGO group were significantly higher than in the UGO group. After 4 weeks of treatment, PT, PT/W, and TW were significantly greater in the rTMS+UGO group than in both the rTMS and UGO groups.

Table 8 shows that, at baseline, there were no significant differences in PT, PT/W, and TW for knee joint flexion among the three groups (p $>$ 0.05). During knee flexion, PT, PT/W, and TW all showed significant time effects (F = 24.082, p 0.001; F = 23.971, p 0.001; F = 41.832, p 0.001), indicating that these parameters increase over time in all three groups. However, there was no significant group effect (p $>$ 0.05), and no significant time $\times$ group interaction for PT, PT/W, and TW (p $>$ 0.05).

The changes in AEMG of the vastus medialis, rectus femoris, hamstring, and gastrocnemius before and after treatment in the three groups are summarized in Table 9. Before treatment, there were no significant differences in AEMG values of the affected-side muscles among the three groups (p $>$ 0.05). A substantial time effect was found for all four muscles (p 0.05): AEMG values of vastus medialis and rectus femoris increased significantly over time, whereas AEMG values of hamstring and gastrocnemius muscles decreased significantly with time (p 0.05).

Furthermore, there was a marked group effect for the affected-side muscle (p 0.05) and a significant time $\times$ group interaction effect (p 0.05). After 4 weeks of treatment, AEMG values of the vastus medialis and rectus femoris were significantly higher in the rTMS+UGO group than in the rTMS and UGO groups. In contrast, AEMG values of the hamstring and gastrocnemius muscles were considerably lower than in the rTMS group and the UGO group (p 0.05).

In the rTMS group, 4 individuals experienced adverse events, including 1 with increased spasticity, 1 with fatigue, and 2 with rTMS-related headache. In the UGO group, 3 patients reported adverse reactions, including 2 with elevated spasticity (spasm aggravation) and 1 with fatigue. Furthermore, 2 patients developed adverse events in the rTMS+UGO group (1 with spasticity (spasm aggravation)) and 1 with rTMS-related headache. However, the overall rate of adverse events did not differ substantially among all three groups (p $>$ 0.05). The adverse events observed in all three groups after treatment are presented in Table 10.