Balance, also referred to as postural control, is the ability to maintain the body’s center of gravity within the base of support by adapting to and counteracting changes in gravity. Balance is a crucial factor in functional movement during daily activities and has a strong correlation with walking ability.^1,2 Hemiplegic patients due to stroke experience a loss of central nervous system control, resulting in sensory loss and muscle weakness. These impairments limit their ability to align body segments and control posture, significantly increasing the risk of falls.³ Such balance impairments are characterized by asymmetric weight distribution and impaired weight-shifting ability, leading to increased body sway and difficulty in independent walking.^4,5

Recently, sensory integration training using visual feedback and auditory feedback training has gained considerable attention as an approach to restore walking ability affected by balance disorders.⁶ Postural control and balance are influenced by various intrinsic feedback factors, including somatosensory, visual, and vestibular senses. However, when intrinsic feedback does not function properly, such as in stroke patients, information can be provided through therapists or biofeedback devices. This extrinsic feedback refers to the process of obtaining information from the surrounding environment or others.^7,8

Training with visual feedback is a method that promotes motor learning by quickly and accurately correcting movement errors using visual information. In a study by Pak and Lee⁹, weight-bearing training was conducted using visual feedback for stroke patients, resulting in increased muscle activation on the affected side and improved balance. Similarly, a subsequent study by Lee et al.¹⁰ showed that training with visual feedback had a positive impact on balance and walking in stroke patients. Another form of feedback, auditory feedback, was applied in this study through rhythmic auditory stimulation, which provides auditory stimuli at consistent rhythms to synchronize and activate the brain’s motor and perceptual areas.¹¹ Research on rhythmic auditory stimulation has shown not only improvements in stride length and walking speed but also increased walking symmetry.^12,13 Additionally, studies by Ford et al.¹⁴ and Roerdink et al.¹⁵ reported that treadmill training with rhythmic auditory stimulation improved arm and leg coordination, leading to increased pelvic and thoracic rotation. These results indicate that walking training with rhythmic auditory stimulation has highly positive effects on walking ability.

As such, rehabilitation using visual and auditory feedback has been widely studied not only for sensory integration training but also for motivating movement in stroke patients. Recently, studies have reported that robot-assisted gait training with visual feedback can provide strong feedback effects, enhancing motivation and adaptability.¹⁶ Robot-assisted gait training using visual feedback typically involves task execution on a screen in front of the patient, facilitating movement learning through problem-solving during gait training, based on motor learning theories.¹⁷

These visual and auditory stimuli create interest through diverse task execution, provide appropriate motivation, and promote repetitive learning as a form of exercise.¹⁸ Based on this evidence, it is anticipated that the combination of visual feedback and rhythmic auditory stimulation in robot-assisted gait training will yield synergistic effects. However, previous studies have not directly compared visual feedback with auditory stimulation, and studies on similar topics have had very short training periods, making it difficult to properly compare therapeutic effects. Therefore, this study aims to suggest an effective direction for robot rehabilitation for stroke patients by comparing robot-assisted gait training using visual feedback and auditory stimulation.

1. Subjects

1) Participants

The subjects of this study were selected as 40 patients who were admitted to Chung-Nam national university hospital in Daejeon. In addition, the subjects were not experienced in robot-assisted gait training, and the stroke patients were composed of patients who had no over-lapping diseases within the past 6 months after the onset of stroke to minimize the possibility of natural restoration. To minimize the selection bias, the following selection criteria were applied randomly to the two groups.

The inclusion criteria were as follows: (1) ability to walk 10 minutes (with or without assistive device); (2) impairment of balance ability (maximum berg balance scale score 45); (3) cognitive abilities enabling communication (minimum MMSE score 24); (5) medically stable and free of major cardiovascular or other medical conditions. The general characteristics of the participants are shown in Figure 1.

Fig. 1. Enrollment of stroke patients.

2) Experimental procedure

The subjects of this study were 40 patients selected from those admitted to Chung Nam national university hospital in Daejeon for physical therapy, based on specific inclusion criteria. Before initiating experiment, balance and gait abilities of subjects were assessed. And they were divided into the VRGT and the ARGT. The robotic device using gait training applied the Lokomat Pro (Hocoma AG, Zurich, Switzerland), an exo-skeletal type robotic device. The VRGT and ARGT underwent robot-assisted gait training three times per week, 30 minutes per session, for 6 weeks. Conventional physical therapy consisted of 30 minute per session for 6 weeks. To enhance reliability between trainers and evaluators, separate physical therapists, each with over three years of experience, were designated for training and evaluation tasks.

(1) Visual feedback with Robot-assisted Gait Training group (VRGT)

Robot-assisted gait training with visual feedback was conducted using a program integrated into the Lokomat Pro system, based on various task-based training exercises displayed on the front screen. The visual feedback used in this study was provided by a program called Augmented Feedback, which enhances motivation for movement through task performance in diverse virtual environments. This program involves activities such as catching animals or navigating through different scenarios displayed on the screen (Figure 2, 3).

Fig. 2. Visual Feedback (Augmented Feedback).

Fig. 3. Robot-assisted gait training (Lokomat Pro).

(2) Auditory stimulation with Robot-assisted Gait Training group (ARGT)

In this study, Robot-assisted gait training with auditory stimulation was conducted through regular auditory cues provided by a metronome. The intervention for walking speed involved determining each patient’s comfortable walking speed and adjusting the tempo of the metronome accordingly to match their steps per minute. For the safety of the participants, all training was supervised by a physical therapist with more than three years of experience in robot-assisted gait therapy, with one assistant assigned to each patient during the experiment. The experiment began with a one-minute adaptation period to familiarize the patient with the metronome tempo, followed by a 30-minute gait training session synchronized to the metronome. The training program lasted for six weeks, with sessions conducted three times a week, each lasting 30 minutes.

2. Experimental methods

1) Measurement

(1) Medical Research Council (MRC)

The MRC was applied to evaluate lower extremity muscle strength. The MRC is divided into six grades: Normal (5), Good (4), Fair (3), Poor (2), Trace (1), Zero (0). Hip flexion, extension, abduction, knee flexion, extension, ankle dorsiflexion, plantar flexion of affected side lower extremity was assessed and has a total score of 30 points. The mean value was recorded.

(2) The Berg Balance Scale (BBS)

In this study, the BBS was used as a tool for balance assessment. The BBS are 14 different items that evaluates the degree of the balance and fall risk in stroke patients. The evaluation items are for dynamic and static balance. And it takes about 15 minutes. The evaluators were therapists with more than three years of clinical experience and conducted evaluation before and after intervention.

(3) The Timed Up and Go Test (TUG)

The TUG is a representative test that can evaluate dynamic balance. Therefore, in this study, TUG was conducted and a total of three attempts were made and the average value was recorded.

(4) 10 Meter Walking Test (10MWT)

The 10 Meter Walking Test (10MWT) was used to evaluate walking ability and walking speed. The 10MWT was conducted and a total of three attempts were made and the average value was recorded.

(5) Fugl-Myer Assessment (FMA)

In this study, the FMA, a representative motor function assessment, was evaluated. FMA evaluated lower extremity motor function, excluding upper extremity function, and has a total score of 34 points.

(6) Modified Barthel Index (MBI)

The MBI evaluates whether patients with brain damage can perform activities of daily living, and serves as the basis for functional judgment of sequelae. Therefore, in this study, the MBI was conducted to determine the impact on daily life through functional recovery.

2) Data analysis

The statistical analysis of this study was performed using SPSS Version 25 (SPSS Inc., Chicago, IL, USA). The general characteristics of the subjects were tested for normality using descriptive statistics (Shapiro-Wilk Test). Paired t-tests were used to evaluate differences before and after treatment. Also, Independent t-tests were performed to determine the significance of differences between groups. The level of significance was set at p<0.05.

1. General and medical characteristics of subjects

The general and medical characteristics of all subjects in the VRGT and ARGT were all homogenous (Table 1).

Table 1

Clinical characteristics of participants (n=24)

Variables	VRGT (n= 12)	ARGT (n= 12)	p
Age (year)	55.6± 10.4	56.7± 4.4	0.752
Height (cm)	167.4± 6.8	166.6± 5.4	0.542
Weight (kg)	67.9± 5.2	66.9± 7.1	0.954
Delay(months)	7.3± 1.2	7.5± 1.7	0.723
MMSE-K	28.3± 1.1	28.2± 1.7	0.912
Gender (male/female)	7/5	9/3	0.581
Hemiplegic side (left/right)	9/3	5/7	0.642
Mechanism (Haemorrhage/ Ischaemia)	3/9	8/4	0.105

VRGT: Visual feedback robot-assisted gait training group, ARGT: Auditory stimulation robot-assisted gait training group, MMSE-K: Mini-Mental State Examination-Korean.

2. Changes in balance and gait abilities of participants in this study

Both experimental groups showed significant differences before and after the intervention in all areas measured (p<0.05). Additionally, between-group comparisons of VRGT and ARGT showed significant differences only in MRC and FMA. BBS, TUG, 10MWT, and MBI showed no significant difference in VRGT compared ARGT (Table 2, 3).

Table 2

Changes in balance ability and muscle strength of the participants in this study (n=24)

	VRGT (n= 12)		ARGT (n= 12)		t(p)
	Pre-test	Post-test	Pre-test	Post-test
MRC (score)	15.83± 0.68	20.33± 1.02	15.08± 0.90	18.50± 0.67	2.461 (0.022*)
Difference (post-pre)	4.50± 1.65*		3.41± 0.90
t	-19.541		-13.146
p	< 0.001*		< 0.001*
BBS (score)	40.50± 1.65	50.16± 1.72	40.41± 5.21	48.16± 1.80	0.414 (0.833)
Difference (post-pre)	9.66± 1.97		7.75± 4.00
t	-16.258		-6.707
p	< 0.001*		< 0.001*
TUG (sec)	27.25± 2.31	14.50± 2.81	27.58± 5.71	16.50± 4.44	0.347 (0.733)
Difference (post-pre)	-12.75± 3.19		-11.08± 5.17
t	13.249		6.746
p	< 0.001*		< 0.001*

VRGT: Visual feedback robot-assisted gait training group, ARGT: Auditory stimulation robot-assisted gait training group, MRC: Medical Research Council, BBS: Berg Balance Scale, TUG: Timed Up and Go. *Significant difference (p<0.05) between VRGT and ARGT.

Table 3

Changes in gait ability, FMA and MBI of the participants in this study (n=24)

	VRGT (n= 12)		ARGT (n= 12)		t (p)
	Pre-test	Post-test	Pre-test	Post-test
10MWT (sec)	29.16± 2.54	17.08± 1.25	29.25± 5.70	18.91± 4.79	0.414 (0.641)
Difference (post-pre)	-12.08± 2.89		-10.33± 4.83
t	13.819		7.410
p	< 0.001*		< 0.001*
FMA (score)	23.66± 3.24	30.16± 1.57	23.50± 2.84	27.16± 1.99	4.326 (0.001*)
Difference (post-pre)	6.50± 3.64*		3.66± 3.28
t	-5.922		-3.867
p	< 0.001*		0.003
MBI (score)	59.33± 7.63	67.33± 6.69	53.50± 4.98	61.41± 7.98	0.098 (0.923)
Difference (post-pre)	8.00± 0.93		7.91± 3.00
t	-3.778		-5.588
p	0.003		< 0.001*

VRGT: Visual feedback robot-assisted gait training group, ARGT: Auditory stimulation robot-assisted gait training group, 10MWT: 10 Meter Walking Test, FMA: Fugl-Meyer Assessment, MBI: Modified Bathel Index. *Significant difference (p<0.05) between VRGT and ARGT.

This study aims to compare the effects of robot-assisted gait training using visual feedback and auditory stimulation on the balance and gait of stroke patients, proposing more effective clinical methods. Recovery of walking ability, which is a primary goal of rehabilitation, determines whether patients can lead independent lives in daily activities.¹⁹ Walking ability depends on lower limb strength, sensory integration, center of gravity shifting, and weight-bearing support. Among these, muscle strength is closely related to balance and has a strong correlation with walking ability.^20,21 In particular, ankle weakness reduces plantarflexion and dorsiflexion, hindering standing activities and walking.^22,23 This study anticipated that maintaining lower limb strength would aid the recovery of balance and walking ability. The results demonstrated significant pre-and-post intervention improvements in both the VRGT and ARGT groups, with VRGT showing greater effectiveness in enhancing lower-limb strength (p<0.05). This improvement in lower limb strength was expected to result in changes in balance ability. The balance ability of stroke patients is often characterized by asymmetrical weight shifting and an over-reliance on the non-affected side for weight-bearing.²⁴ Such asymmetry leads to greater postural sway than in healthy individuals, increasing the risk of falls. Asymmetric weight-bearing is also a major cause of reduced walking speed, resulting in inefficient and energy-consuming gait patterns compared to healthy individuals.²⁵

Stroke patients often experience visual sensory deficits, that play a critical role in maintaining balance. Visual input provides information that the central nervous system integrates with other sensory inputs to regulate posture.²⁶ To address asymmetric weight-bearing, sensory feedback training using visual and auditory stimuli has been employed.²⁷ Pak and Lee⁹ demonstrated that balance training using visual feedback significantly improved balance. Subsequently, studies by Khallaf et al.²⁸ and Druzbicki et al.²⁹ validated the effectiveness of gait training with visual feedback. Moreover, research by Ham and Lim³⁰ integrating visual feedback with robot-assisted gait training reported positive outcomes in balance and walking recovery.

Based on this evidence, this study evaluated balance ability using BBS and TUG. The VRGT group showed significant pre and post intervention improvements in BBS and TUG scores (p<0.05). However, no significant difference was observed between the VRGT and ARGT groups (p>0.05), suggesting that while VRGT was effective in restoring balance, it was not more effective than ARGT. These results were expected to restore balance ability due to the improvement of lower extremity muscle strength, but both groups showed therapeutic effects in the BBS and TUG, which are balance ability assessments after the intervention, but this was not the case in the comparison between groups. Both temporal feedback and auditory stimulation helped patients restore balance ability through external feedback, but the results of the study showed that there were limitations in showing the superiority of the two interventions. Previous studies have reported that gait training with rhythmic auditory stimulation effectively improves walking speed and symmetry.^13,26 Further, Yoon and Lee³¹ found that treadmill training with rhythmic auditory stimulation contributed to stable balance recovery and gait improvement. These findings suggest that audiovisual stimuli are highly effective in restoring balance and gait abilities in stroke patients by improving impaired sensory input. Recently, training methods based on motor learning concepts have gained attention. Among these, task-specific training, such as treadmill training, is a representative example for recovering walking ability in stroke patients.³² However, traditional treadmill training often requires multiple therapists for support, and the physical effort involved leads to inconsistencies in therapeutic assistance. This has drawn interest toward robot-assisted gait training, which offers the advantages of early gait rehabilitation and efficient training methods.³³ Studies by Wong et al.³⁴ demonstrated that robot-assisted gait training is closely associated with improvements in dynamic balance and walking ability. Research by Bonnyaud et al.³⁵ suggested that robot-assisted gait training facilitates symmetrical gait patterns. Similarly, in this study, both experimental groups showed significant pre and post intervention differences in the 10MWT (p<0.05), confirming the effectiveness of robot-assisted gait training in restoring walking ability. However, no significant differences were observed between the VRGT and ARGT groups (p>0.05), indicating that visual feedback did not lead to greater improvements in walking ability compared to auditory stimulation. Changes in walking ability were inferred through functional assessments such as the FMA and MBI, which reflect recovery of daily living activities. Post-intervention improvements in walking ability were accompanied by significant changes in FMA and MBI scores (p<0.05). Notably, VRGT showed a significant advantage over ARGT in FMA scores, while no significant differences were observed between the groups in MBI scores (p>0.05). In the case of VRGT, which showed improvement in muscle strength in the comparison between groups, the biggest reason why there was a significant difference in FMA was that the difference between groups was reflected in the improvement in scores in items related to voluntary movement or muscle strength of the lower extremities among the sub-evaluation items. However, as shown by the results showing that the scores for the MBI, which can infer changes in daily life, did not show a significant difference, it was found that there was no significant difference in the impact on daily life between the two interventions.

This study faced limitations, including overlapping training and hospitalization periods and the inability to fully control variables such as participant selection and dropout rates due to a maximum hospitalization period of eight weeks. Furthermore, the six-week duration of the intervention may have been insufficient to induce significant changes in patients. Above all, the absence of a control group that only performed general treatment for appropriate comparison is thought to have had a significant impact on the theoretical background of the results of this study. Future studies should address these limitations by incorporating longer training periods or follow-up observations to achieve more robust findings. In addition, we hope to secure a large number of participants to provide another theoretical background through an appropriate comparison between the control group and the intervention of this study.

This study was supported by S University (2018).