Visual and visual–proprioceptive training improve gait via sensory reweighting after chronic stroke: a randomized controlled trial

Abstract

Background

Post-stroke gait recovery depends on multisensory integration, but vision- and gaze-stability training are rarely emphasized. We tested whether 12-week visual oculomotor training (VOT) or visual–proprioceptive integration (VPI) programs improve gait after chronic stroke.

Methods

In this single-center, assessor- and analyst-blinded randomized controlled trial, 45 participants were allocated to a control group (CG), VOT, or VPI. All groups received standardized gait training plus three 30-min group-specific sessions weekly for 12 weeks. Static and dynamic visual acuity were assessed at baseline and post-intervention. Pre- and post-intervention, all participants completed 10-meter walk tests, during which data were synchronously acquired using a 3D motion-capture system (Qualisys) and dual force plates (Kistler). The primary outcome was comfortable gait speed measured during the 10-meter walk tests; secondary outcomes included step-length symmetry (SLS), center-of-pressure symmetry ratio (COP-SR), other spatiotemporal, kinematic, and kinetic gait measures, and clinical scales including the Fugl–Meyer Assessment–Lower Extremity (FMA-LE) and Berg Balance Scale (BBS). Between-group effects were estimated using baseline-adjusted ANCOVA.

Results

For the primary endpoint of comfortable gait speed, VPI improved significantly more than CG (P = 0.018), whereas VOT did not differ from CG or VPI (P = 0.284 and 0.142). Within groups, gait speed increased significantly only in VPI (P = 0.019; CG: P = 0.214; VOT: P = 0.184). For the key secondary outcome of SLS, VPI improved more than CG and VOT (P < 0.001 and P = 0.008, respectively), while VOT did not differ from CG (P = 0.224). The COP-SR shifted toward symmetry in all groups, but between-group differences were not significant (all P > 0.44), whereas paretic-limb COP displacement increased significantly only in VOT and VPI (P = 0.021 and 0.009, respectively). Both interventions produced significant within-group improvements in dynamic visual acuity and lower-limb motor function (ΔlogMAR and FMA-LE; all P < 0.05 for VOT and VPI, P > 0.10 for CG). BBS scores increased significantly in the intervention groups but not in CG, whereas between-group differences in BBS change were not significant (P = 0.153). Spatiotemporal analyses showed shorter cycle time, longer non-paretic and paretic step length, reduced double-limb support, and increased paretic single-limb support in VOT and VPI (all P < 0.05 vs. baseline), with no significant changes in CG (P ≥ 0.05).

Conclusions

On top of conventional care, visual training targeting gaze stability and multisensory integration produced clinically relevant improvements in gait after chronic stroke. VPI produced the largest gains in gait speed and symmetry, supporting vision–vestibular–proprioceptive integration as a pragmatic rehabilitation target.

Trial registration: Clinical Trial Registry: ChiCTR2500114339. Registered on 10 December 2025. Retrospectively registered.

Background

Stroke remains one of the leading causes of disability and death worldwide, imposing long-term functional loss and a substantial public-health burden [1]. According to the Global Burden of Disease 2021 estimates, there were 11.9 million incident strokes, 93.8 million people living with stroke, 7.3 million stroke deaths, and 160.5 million disability-adjusted life years (DALYs) worldwide in 2021. Absolute numbers have risen markedly since 1990, largely due to population growth and ageing. Stroke ranked third for deaths and fourth for DALYs globally in 2021 [2]. In parallel, forward-looking models project a continued increase in the global age-standardized incidence of ischemic stroke through 2030 [3].

Visual impairment affects approximately 70% of people hospitalized after a stroke and includes visual-field loss and oculomotor deficits [4, 5]. These disturbances degrade gaze stability and dynamic visual acuity, restrict spatial navigation and postural orientation, and raise fall risk—making vision a pragmatic target for rehabilitation [4, 6]. Consistent with Fig. 1, lesions along the visual pathway yield characteristic field-loss patterns—from monocular blindness to homonymous hemianopia—that limit access to environmental cues and exacerbate walking disability, reinforcing the rationale for targeting visual-pathway functions in gait rehabilitation [7].

Fig. 1.

Lesion sites along the human visual pathway and the corresponding canonical visual-field defects. A Ventral schematic of the visual pathway with key structures labeled (optic nerve, optic chiasm, optic tract, lateral geniculate nucleus, optic radiations—Meyer’s loop and dorsal bundle—and primary visual cortex). Numbered markers (①–⑦) denote common lesion locations. Mini-maps illustrate the expected visual-field loss for right-sided examples (white = preserved vision; purple = deficit; circles depict each eye’s visual hemifields). Lesion–deficit pairs: ①, optic nerve → monocular blindness (ipsilateral); ②, lateral chiasmal lesion → ipsilateral nasal hemianopia; ③, midline chiasmal compression → bitemporal hemianopia; ④, optic tract → contralateral homonymous hemianopia; ⑤, temporal optic radiations (Meyer’s loop) or ventral occipital cortex → contralateral superior quadrantanopia; ⑥, parietal optic radiations or dorsal occipital cortex → contralateral inferior quadrantanopia; ⑦, unilateral primary visual cortex → contralateral homonymous hemianopia with macular sparing. Mirror patterns apply for left-sided lesions

Contemporary gait rehabilitation for people with stroke typically prioritizes strength and task practice, with limited attention to trainable elements such as multisensory integration (visual–vestibular–proprioceptive) and gaze stabilization [8]. Oculomotor deficits—reduced saccades and smooth pursuit—are common after cortical/brainstem lesions, often overlooked, and closely linked to functional limitations [9, 10]. Emerging evidence suggests that gaze stabilization and sensory reweighting interventions can significantly enhance walking recovery by improving the integration of visual, vestibular, and proprioceptive inputs [11]. Here, “sensory reweighting interventions” refers to structured tasks that intentionally challenge or alter the reliability of these inputs so that the nervous system down-weights overreliance on vision and up-weights underused vestibular/proprioceptive channels. Specifically, gaze-stabilization training strengthens the vestibulo-ocular reflex (VOR), maintaining clear vision during head motion and facilitating coordinated head–trunk–limb control [12]. Sensory reweighting paradigms then train patients to adjust reliance across modalities, enabling more effective use of vestibular and proprioceptive cues when visual information is degraded [11], thereby improving balance and gait control, particularly in unpredictable environments [13]. However, despite these promising findings, direct, head-to-head randomized comparisons between oculomotor-focused and visual–proprioceptive integration training within a unified rehabilitation framework remain limited. Moreover, many existing studies rely on clinical scales, report heterogeneous outcomes, and infrequently incorporate objective biomechanical measures [11, 14–16]. High-fidelity evaluation with 3D motion capture and synchronized force-platform recordings is therefore needed to quantify kinematic and kinetic outcomes with less bias. Force platform–based visual feedback has long been used to retrain postural control after stroke, and early trials and reviews report gains in stance symmetry and selected balance outcomes [17, 18]. Limits-of-stability testing and related training on computerized posturography systems such as NeuroCom are reliable and widely adopted [19].

Building on this established paradigm, our visual–proprioceptive integration protocol emphasizes precise, target-directed center-of-mass or center-of-pressure shifts in eight randomized directions under stable visual fixation, with accuracy-based progression, to drive sensory reweighting beyond conventional continuous sway or predictable weight-shift routines [20]. Against this background, we hypothesized that, beyond matched-dose conventional rehabilitation and standardized gait training, two complementary visual-based programs would improve walking: an oculomotor visual training (VOT) program to enhance gaze stability and oculomotor control—consistent with randomized trials of gaze-stability exercises in stroke—and a visual–proprioceptive integration (VPI) program to explicitly engage multisensory reweighting and thereby produce larger gains, particularly in gait symmetry and paretic weight shift. Together, these mechanisms were expected to increase comfortable gait speed by improving dynamic visual acuity, stabilizing head–eye–trunk coordination, recalibrating sensory weights toward more symmetric propulsion, and reducing double support, which in aggregate enhances step-to-step efficiency during walking.

Accordingly, we conducted a 12-week randomized controlled trial to compare oculomotor-focused versus visual–proprioceptive integration training as adjuncts to usual gait rehabilitation.

Methods

Study design and participants

This single-center randomized controlled trial enrolled 45 eligible people with stroke from Tianjin Occupational Diseases Precaution and Therapeutic Hospital, with recruitment and intervention conducted between November 2024 and July 2025. Participants were randomly allocated (1:1:1) to the control group (CG), the VOT group, or the VPI group. The target sample size was determined a priori based on the primary outcome of comfortable gait speed (m·s⁻¹). Using G*Power 3.1, we based the a priori sample size calculation on an ANCOVA framework for the primary outcome (post-intervention comfortable gait speed), with three groups and baseline values entered as a covariate. Assuming a medium effect size (f = 0.25) for the overall group effect, α = 0.05, and power (1 − β) = 0.80, the required total sample size was 36 participants. Allowing for approximately 10–15% attrition, we therefore set the target enrolment at 45 participants. The intervention period was 12 weeks. No interim analyses or formal stopping guidelines were prespecified; the trial was planned to stop once the target sample had completed the intervention. All groups received usual care combined with standardized gait training; in addition, each group completed three 30-min sessions per week, with CG receiving time- and attention-matched activities and VOT/VPI receiving group-specific training. Reporting followed the CONSORT 2010 statement. This randomized controlled trial was retrospectively registered in the Chinese Clinical Trial Registry (ChiCTR2500114339) on 10 December 2025.

Eligibility criteria

Inclusion criteria. Diagnosis of ischemic or hemorrhagic stroke; 9–26 months post-onset with a stable clinical condition; able to safely complete a 10-m walk and moderate-intensity standing/walking training with minimal assistance; refractive errors permitted, with all assessments and training performed under best correction; best-corrected visual acuity (BCVA) ≤ 0.70 logMAR; able to follow verbal commands; provision of written informed consent.

Exclusion criteria. Moderate-to-severe visual dysfunction that persists after best correction and compromises safe ambulation; uncontrolled vestibular disorders; clinically meaningful foot drop (peak ankle dorsiflexion < 0° during mid-swing on 3D gait analysis); significant cervical spine or vertebral artery disease; uncontrolled epilepsy or a history of photosensitive epilepsy; cardiopulmonary or circulatory conditions that contraindicate moderate-intensity walking exercise; severe cognitive or communication impairment precluding participation; receipt of specialized visual or vestibular rehabilitation within the previous 3 months; poor adherence or refusal to attend follow-up.

The protocol was approved by the Institutional Review Board of Tianjin University of Sport (TJUS2024-023). All procedures adhered to relevant guidelines and regulations, and all participants provided written informed consent. The full trial protocol is available from the corresponding author on reasonable request. Baseline demographics and visual assessments are summarized in Table 1.

Table 1.

Baseline characteristics of participants

Parameter		CG (n = 15)	VOT (n = 15)	VPI (n = 15)
Sex, male/female		10/5	9/6	9/6
Stroke subtype, ischemic/hemorrhagic		11/4	11/4	10/5
Paretic side, left/right		5/10	6/9	5/10
Age, y		62.57 ± 5.02 (50–67)	63.71 ± 4.98 (52–71)	61.49 ± 5.26 (49–70)
Time since stroke, mo		17.32 ± 5.28 (9–24)	17.14 ± 5.66 (9–26)	18.34 ± 5.21 (10–26)
Use of an AFO, n		3	2	3
Comfortable gait speed, m·s⁻¹		0.36 ± 0.11 (0.22–0.53)	0.37 ± 0.14 (0.19–0.58)	0.37 ± 0.09 (0.24–0.51)
FMA-LE, score		24.95 ± 3.22 (20–29)	25.62 ± 3.55 (22–30)	24.39 ± 4.28 (18–31)
BBS, score		41.92 ± 5.32 (34–48)	41.74 ± 6.24 (33–48)	42.57 ± 6.77 (32–49)
Visual assessments	Dominant eye: BCVA (logMAR)	0.100 ± 0.028 (0.06–0.14)	0.098 ± 0.049 (0.05–0.15)	0.103 ± 0.035 (0.05–0.16)
	Non-dominant eye: BCVA (logMAR)	0.175 ± 0.036 (0.12–0.23)	0.176 ± 0.037 (0.12–0.23)	0.178 ± 0.061 (0.09–0.27)
ΔlogMAR (DVA − SVA)		0.29 ± 0.10 (0.14–0.44)	0.28 ± 0.07 (0.18–0.39)	0.28 ± 0.13 (0.12–0.42)

Randomization, allocation concealment, and blinding

A computer-generated randomization list was prepared by an independent biostatistician with no role in recruitment or outcome assessment. Participants were randomized in a 1:1:1 ratio using permuted blocks with variable sizes of 3 and 6. Randomization was stratified by baseline comfortable gait speed, by visual phenotype classified as visual-field–dominant or non–visual-field–dominant impairment, and by stroke subtype, ischemic or hemorrhagic, to balance prognostic factors that may influence baseline gait and sensory-integration profiles. After completion of baseline assessments and confirmation of eligibility, a study coordinator independent of the clinical team accessed group assignments sequentially from a pre-sealed allocation list to maintain concealment and prevent predictability.

Given the nature of the interventions, participants and treating therapists were not blinded. Outcome assessors and data analysts were blinded; assessments were scheduled at times and in facilities separate from training sessions, and participants were instructed not to disclose their allocation. If inadvertent unmasking occurred, a second blinded assessor completed that visit’s evaluation.

Continuous variables are presented as mean ± SD (range); categorical variables are presented as counts (n). All visual testing was performed under best optical correction. BCVA and DVA were measured at 4 m using an electronic ETDRS Sloan letter chart and scored in logMAR with letter-by-letter scoring (0.02 logMAR per letter, approximately 0.10 logMAR per line). ΔlogMAR: dynamic minus static visual acuity (logMAR). Abbreviations: CG, control group; VOT, visual oculomotor training; VPI, visual–proprioceptive integration; BCVA, best-corrected visual acuity; DVA, dynamic visual acuity; SVA, static visual acuity; logMAR, logarithm of the minimum angle of resolution; y, year; mo, month; AFO: ankle-foot orthosis; FMA-LE: Fugl–Meyer Assessment—Lower Extremity; BBS: Berg Balance Scale.

Study setting and instrumentation

All procedures were conducted in the Sports Biomechanics Laboratory of Tianjin University of Sport. All assessments were conducted in a standardized laboratory environment with controlled lighting and temperature. An overhead track–mounted fall-arrest harness and an emergency-stop system were available for safety.

Gait data were collected with a Qualisys three-dimensional motion-capture system (8 cameras, 300 Hz; Qualisys AB, Sweden) synchronized with two Kistler force plates (1,000 Hz; Kistler Instrumente AG, Switzerland) embedded along a 10-m straight walkway. Cross-modal synchronization was achieved via hardware triggering.

Visual and dynamic visual acuity assessment

Participants were tested in a quiet room with constant illumination. Under best optical correction, they were seated 4.0 m from a back-lit electronic ETDRS Sloan letter chart. Monocular testing was performed first with the non-tested eye occluded in a randomized left/right order (the dominant eye was documented), followed by binocular testing. Testing began at a line the participant could easily recognize and proceeded to smaller lines. Each line contained five letters; participants had 3–5 s to respond per letter [21]. Testing on a given line was terminated when ≥ 3 letters were read incorrectly. Letter-by-letter scoring was used (0.02 logMAR per correctly identified letter), and SVA was recorded in logMAR units (lower values indicate better acuity) [22].

DVA was assessed at the same 4.0 m distance using the identical optotypes and chart layout to permit direct computation of ΔlogMAR. With the head pitched approximately 30° forward, participants performed active, horizontally directed head oscillations with an amplitude of approximately 10–15° to each side, paced by a metronome set to 120 beats per minute, in line with previous clinical DVA protocols that use rapid (2 Hz) head movements to preferentially engage the vestibulo-ocular reflex [23]. Letter-by-letter scoring yielded DVA (logMAR), and ΔlogMAR was calculated as logMAR(DVA) − logMAR(SVA), with larger values indicating poorer dynamic acuity [24]. Each condition comprised two trials of continuous head oscillation; the initial direction was randomized, and the mean of the two trials was used for analysis. DVA was assessed at baseline and post-intervention.

Training protocols

Across all three groups, participants attended three 30-min sessions per week for 12 weeks. Each session comprised 24 min of group-specific task-oriented training and 6 min of seated rest. The total training volume and work-to-rest ratio were matched across groups.

CG. To control for therapist attention while avoiding exposure to the putative active ingredients of the visual-based interventions, therapists provided only standardized prompts related to safety and cadence. They deliberately withheld strategy-specific feedback, such as gaze-stabilization cues or explicit weight-shift coaching. Control-group activities consisted of in-place marching and weight-shift tasks on firm ground; unstable surfaces and any visual–vestibular or optic-flow stimuli were not used.

VOT. With best optical correction, participants performed horizontal and vertical saccades and smooth-pursuit tracking in standing. Within each 30-min session, approximately 24 min were devoted to active eye-movement practice and about 6 min to brief extra-ocular relaxation breaks, so that the overall work–rest schedule and total duration were identical to CG.

VPI. Participants performed visual feedback–guided center-of-mass/pressure shifts using the NeuroCom Balance Manager. Standing with both feet aligned to platform footprints and fixating the front screen, they shifted toward eight target directions (0°, 45°, 90°, 135°, 180°, 225°, 270°, 315°) in randomized order to superpose the on-screen cursor with the target region (2 × 12-min bouts; 6-min inter-bout rest). Stepping was not allowed and the trunk was kept upright. In the VPI group, progression criteria required ≥ 80% target-hit accuracy at the current difficulty to advance to a more challenging level; otherwise, the current level was maintained. The training setups for the VOT and VPI interventions are illustrated in Fig. 2.

Fig. 2.

Training setups for the visual oculomotor (VOT) and visual–proprioceptive integration (VPI) interventions. a VOT: the participant stands on a level walkway approximately 2 m in front of a monitor and performs gaze-stabilization tasks guided by a moving visual target, with the therapist providing manual guarding from behind. b VPI: the participant stands on the force platform of the balance system and uses real-time visual feedback to shift the center of pressure toward multiple target directions while being guarded by the therapist

Across groups, workload during standing training was titrated to a Borg rating of perceived exertion of 11–13 (light–moderate) and a session-averaged heart-rate reserve of 40%–60% [25]; all training was performed in standing with continuous therapist guarding. Adverse events were monitored and recorded.

Clinical assessments

Lower-limb motor impairment was evaluated at baseline and post-intervention using the FMA-LE (0–34 points), and functional balance was assessed with the BBS (0–56 points) [26, 27]. Both scales were administered by trained physical therapists who were blinded to group allocation.

Data acquisition and processing

Participants wore standardized athletic apparel and running shoes. All gait outcomes reported herein—including comfortable gait speed, step length, cycle time, hip/knee/ankle kinematics, and plantar center-of-pressure (COP) trajectories used to derive COP-based metrics, including COP displacement and the COP symmetry ratio (COP-SR)—were acquired in the Sports Biomechanics Laboratory using a Qualisys 3D motion-capture system synchronized with Kistler force plates. Twenty-six retroreflective markers were affixed to the pelvis and lower limbs (Fig. 3). Walking trials were performed along a straight, level 10-m walkway at each participant’s comfortable speed. To limit acceleration and deceleration, data were collected over the middle 6 m (2–8 m) of the 10-m walkway. Each participant performed three test trials, resulting in at least five valid gait cycles. A gait cycle was defined as the interval from initial contact of the non-paretic limb to the subsequent initial contact of the same limb. To minimize order effects, test sequences were randomized. For statistical analyses, baseline values were included as covariates. The primary endpoint was comfortable gait speed. Key secondary endpoints were step-length symmetry (SLS) and the COP-SR.

Fig. 3.

Reflective marker set and anatomical landmarks (anterior and posterior views). Markers were placed bilaterally as shown. Abbreviations (panel labels → anatomical terms): L, left; R, right; IAS → ASIS, anterior superior iliac spine; IPS → PSIS, posterior superior iliac spine; FTC, lateral-thigh marker at the proximal one-third of the femur (used in lieu of the greater trochanter); FLE, lateral femoral epicondyle; FME, medial femoral epicondyle; FAX, fibular head (apex of the proximal fibula); TTC, tibial tuberosity; FAL, lateral malleolus; TAM, medial malleolus; FCC, calcaneus; FM1, dorsal aspect of the first metatarsal head; FM2, dorsal aspect of the second metatarsal head; FM5, dorsal aspect of the fifth metatarsal head

SLS and COP-SR were computed as:

Here, SL_paretic and COP displacement_paretic denote the step length and COP displacement of the paretic side, and SL_nonparetic and COP displacement_nonparetic those of the non-paretic side; values approaching 100% indicate greater symmetry.

After acquisition, raw Qualisys and Kistler data were processed in MATLAB (R2023a). Marker trajectories were first baseline-corrected, then filtered with a zero-lag 4th-order Butterworth low-pass filter at 6 Hz to suppress high-frequency noise [28]. Outliers were detected using interquartile-range and Hampel procedures with manual verification, and Daubechies wavelet denoising was optionally applied without altering event boundaries. Gait events (initial contact and toe-off) were identified using a vertical ground-reaction-force threshold of 20 N. Joint-angle time series were normalized to 0–200% (two gait cycles); side-specific ensemble averages (paretic and non-paretic) were generated for analysis. Gait speed was entered as a covariate for SPM-ANCOVA and for sensitivity analyses. For each test session, at least two stable cycles were averaged; all processing code and parameters were version-controlled.

Kinetic variables were normalized to body mass; COP trajectories were geometrically normalized to foot length/width. Following normalization, distributional checks and outlier diagnostics were performed; if residual body-size effects were evident, sex, height, body mass, or leg length were added as covariates. Results are reported using the normalized metrics.

Statistical analysis

All analyses were performed using standard statistical software and SPM1D (MATLAB). Two-sided tests were used with a significance level of α = 0.05.

The primary outcome was comfortable gait speed; key secondary outcomes were SLS and COP-SR. Post-intervention between-group differences in these outcomes were evaluated using ANCOVA with group as a fixed factor and the corresponding baseline value as covariate, with HC3 heteroscedasticity-robust standard errors to address potential variance heterogeneity. When the omnibus group effect was significant, prespecified pairwise contrasts were tested. For the primary and key secondary outcomes, we report adjusted post-intervention means, adjusted between-group mean differences with 95% confidence intervals, P values, ANCOVA F statistics for the omnibus group effect, contrast t statistics, and effect sizes (partial η² for omnibus effects and Hedges’ g for pairwise contrasts). Multiplicity across these planned comparisons was controlled using the Holm–Bonferroni procedure.

For other scalar secondary outcomes, including dynamic visual acuity (ΔlogMAR), FMA-LE, and BBS, within-group pre–post changes were analyzed with paired t tests, and between-group differences in post-intervention scores were examined using ANCOVA with the corresponding baseline values as covariates.

Time-normalized joint-angle and COP trajectories were analyzed using one-dimensional SPM1D. Significant clusters in SPM{t} (paired comparisons) and SPM{F} (SPM-ANCOVA with group as fixed factor and baseline waveforms as covariate) fields were identified using random field theory at α = 0.05 with cluster-wise inference.

Results

A total of 45 participants were randomized to CG (n = 15), VOT (n = 15), and VPI (n = 15), and all completed the post-intervention assessment and were included in the primary analysis. The trial was stopped when the preplanned sample size had completed the 12-week intervention; no early stopping occurred. Baseline demographic and clinical characteristics were comparable across groups, with no significant between-group differences in dynamic visual acuity, FMA-LE, or BBS (all P > 0.05; Table 1).

Primary and key secondary outcomes

Comfortable gait speed (primary outcome)

As shown in Fig. 4a, baseline comfortable gait speed was comparable across groups (0.36–0.37 m·s⁻¹). Following the intervention, gait speed increased modestly in the CG and VOT groups and more clearly in the VPI group. Only the VPI group showed a statistically significant within-group gain (0.37 to 0.46 m·s⁻¹; t(14) = 2.65, P = 0.019, Cohen’s d = 0.68), whereas the small increases in CG (0.36 to 0.38 m·s⁻¹; t(14) = 1.30, P = 0.214, Cohen’s d = 0.34) and VOT (0.37 to 0.41 m·s⁻¹; t(14) = 1.40, P = 0.184, Cohen’s d = 0.36) did not reach significance. For the primary endpoint, the omnibus ANCOVA for post-intervention gait speed revealed a significant overall group effect (F(2,41) = 3.50, P = 0.040). As summarized in Table 2, VPI produced a significantly greater adjusted improvement than CG (mean difference + 0.08 m·s⁻¹, 95% CI 0.02–0.14, P = 0.018, Hedges’ g = 0.87), whereas VOT did not differ significantly from CG (+ 0.03 m·s⁻¹, 95% CI − 0.03–0.09, P = 0.284, g = 0.39) or from VPI (+ 0.05 m·s⁻¹, 95% CI − 0.02–0.12, P = 0.142, g = 0.54). Thus, within the context of standardized usual care, VPI was the only intervention that yielded a statistically and clinically meaningful enhancement of comfortable walking speed.

Fig. 4.

Pre–post changes and between-group differences in motor impairment, balance, sensory integration, and spatiotemporal gait metrics (a–i). a Comfortable gait speed. b Dynamic visual-acuity loss (ΔlogMAR; lower values indicate better dynamic acuity). c Fugl–Meyer Assessment—Lower Extremity (FMA-LE). d Unaffected (non-paretic) limb step length (USL). e Affected (paretic) limb step length (ASL). f Step-length symmetry (SLS). g Gait-cycle duration. h Phase composition of the gait cycle: double-limb support (DLS, black), affected (paretic) single-limb support (A-SS, blue), and unaffected (non-paretic) single-limb support (U-SS, red). i Berg Balance Scale (BBS). Bars depict mean ± SD for all scalar outcomes; in panel h, stacked bars depict mean phase proportions. Within-group pre–post comparisons were performed using paired t tests, and post-intervention between-group contrasts were obtained from ANCOVA adjusted for the corresponding baseline value, with multiplicity controlled by the Holm–Bonferroni procedure within outcome domains. Symbols: * P < 0.05; ** P < 0.01 (within-group pre–post change). # P < 0.05 (between-group difference at post-test). Abbreviations: CG, control group; VOT, visual oculomotor training; VPI, visual–proprioceptive integration

Table 2.

Between-group comparisons for the primary endpoint and key secondary endpoints

Parameter	Group contrast	ANCOVA: adjusted mean difference (former − latter)	95% CI (lower, upper)	P value	Hedges’ g (effect size)
Gait speed (m·s⁻¹)	VPI vs. CG	+ 0.08	[0.02, 0.14]	0.018	0.87
	VOT vs. CG	+ 0.03	[− 0.03, 0.09]	0.284	0.39
	VPI vs. VOT	+ 0.05	[− 0.02, 0.12]	0.142	0.54
SLS (%)	VPI vs. CG	−28.30	[− 42.40, − 14.2]	< 0.001	−1.46
	VOT vs. CG	−8.70	[− 22.90, 5.60]	0.224	−0.44
	VPI vs. VOT	−19.70	[− 33.70, − 5.60]	0.008	−1.02
COP-SR (%)	VPI vs. CG	−8.10	[− 29.60, 13.40]	0.446	−0.27
	VOT vs. CG	−5.30	[− 26.80, 16.20]	0.617	−0.18
	VPI vs. VOT	−2.80	[− 22.70, 17.10]	0.775	−0.10

Step-length symmetry (key secondary outcome)

As illustrated in Fig. 4f, baseline SLS was markedly biased toward the non-paretic limb (substantially above 100%), with comparable asymmetry across the three groups. Following the intervention, both VOT and VPI produced clear reductions in SLS toward 100% (i.e., more symmetrical step lengths), whereas the CG showed only a modest, non-significant improvement (P = 0.058). According to the ANCOVA for post-intervention SLS (adjusted for baseline; Table 2), VPI yielded a markedly greater adjusted improvement in SLS than CG (adjusted difference − 28.3%, 95% CI − 42.4 to − 14.2, P < 0.001, Hedges’ g = − 1.46) and also outperformed VOT (− 19.7%, 95% CI − 33.7 to − 5.6, P = 0.008, g = − 1.02), whereas VOT did not differ significantly from CG (− 8.7%, 95% CI − 22.9 to 5.6, P = 0.224). These findings indicate that VPI markedly improved step-length symmetry and outperformed both conventional gait training alone and VOT in normalizing step-length distribution between limbs.

COP symmetry ratio (key secondary outcome)

As shown in Fig. 6f, baseline COP-SR was skewed toward the non-paretic side (values > 100%), with similar degrees of asymmetry across groups. After the intervention, COP-SR in all three groups shifted toward 100%, indicating more balanced anterior–posterior COP excursions between limbs; the magnitude of improvement tended to be larger in the VOT and VPI groups, while the CG also showed some reduction in asymmetry. However, ANCOVA revealed that adjusted between-group differences in COP-SR were not statistically significant (VPI vs. CG: −8.1%, 95% CI − 29.6 to 13.4; VOT vs. CG: −5.3%, 95% CI − 26.8 to 16.2; VPI vs. VOT: −2.8%, 95% CI − 22.7 to 17.1; all P > 0.40; Table 2). Thus, COP-based symmetry improved to a comparable extent across groups.

Fig. 6.

Plantar load distribution and center-of-pressure (COP) control during gait, and effects of visual-based interventions (a–f). a “Butterfly” COP trajectories overlaid on a plantar-pressure probability map from a representative participant, visualizing the heel-to-forefoot roll-over and mediolateral excursions across multiple steps. b Three-dimensional plantar-pressure topography in quiet stance showing the typical multi-peak pattern at the posterolateral heel and metatarsal heads. c Two-dimensional plantar-pressure heat maps for left and right feet during walking; dashed lines indicate the inferred gait line. d COP gait lines plotted separately for the left and right sides to compare curvature, length, and deviation from the midline. e Stance-phase COP displacement (m) on the unaffected and affected limbs, pre- and post-intervention by group. f COP symmetry ratio, pre- and post-intervention by group. Bars depict mean ± SD. Within-group pre–post differences were assessed with paired tests (symbols: * P < 0.05, ** P < 0.01). Post-test between-group contrasts were analyzed separately using ANCOVA adjusted for baseline within the plantar-pressure/COP outcome domain

Each group contained n = 15 participants. Between-group contrasts were estimated with ANCOVA (factor = Group; covariate = the corresponding baseline value) using heteroscedasticity-robust standard errors (HC3); two-sided α = 0.05. P values were adjusted within outcome domains by the Holm–Bonferroni procedure. The sign convention for the adjusted mean difference is “former minus latter”; for SLS and COP-SR, values closer to 100% indicate better symmetry, so negative contrasts mean the former group is more symmetric. Hedges’ g follows the same sign as the contrast, with larger absolute values indicating stronger effects. Comfortable gait speed (m·s⁻¹) was the sole primary endpoint; SLS (%) and COP-SR (%) were prespecified key secondary endpoints.

Other clinical secondary outcomes

As illustrated in Fig. 4b and c, and 4i, both VOT and VPI produced significant improvements in dynamic visual acuity (ΔlogMAR) and lower-limb motor function (FMA-LE), whereas changes in CG were small and non-significant (P < 0.05 for VOT and VPI; P > 0.10 for CG). ANCOVA for post-intervention ΔlogMAR and FMA-LE (adjusted for baseline) indicated that only VPI showed a significantly greater reduction in ΔlogMAR and a greater gain in FMA-LE compared with CG, while VOT did not differ significantly from CG. BBS scores increased over time in all three groups, with significant within-group improvements in VOT and VPI but not CG. However, between-group differences in BBS change did not reach statistical significance. No training-related serious adverse events occurred in any group. A few participants in the VOT and VPI groups reported mild, transient visual fatigue during or immediately after training sessions, which resolved spontaneously and did not require modification of the protocol.

Spatiotemporal and kinematic gait outcomes

Spatiotemporal gait parameters

As shown in the spatiotemporal panels of Fig. 4, both visually augmented interventions produced coherent changes toward a faster and more symmetrical gait pattern. Cycle time decreased significantly in VOT and VPI (VOT: t(14) = − 2.68, P = 0.018, d = − 0.69; VPI: t(14) = − 4.50, P < 0.001, d = − 1.16), whereas the reduction in CG was smaller and did not reach significance (t(14) = − 2.10, P = 0.054, d = − 0.54). Consistently, double-support duration was reduced and paretic single-limb support was increased in both VOT and VPI (double support: VOT: t(14) = − 3.09, P = 0.008, d = − 0.80; VPI: t(14) = − 4.50, P < 0.001, d = − 1.16; paretic single-limb support: VOT: t(14) = 4.20, P < 0.001, d = 1.08; VPI: t(14) = 4.60, P < 0.001, d = 1.19), whereas changes in CG were smaller and only approached significance for paretic single support (t(14) = 2.11, P = 0.053, d = 0.55). Both experimental groups also showed significant increases in non-paretic and paretic step length (non-paretic: VOT: t(14) = 2.18, P = 0.047, d = 0.56; VPI: t(14) = 3.79, P = 0.002, d = 0.98; paretic: VOT: t(14) = 2.41, P = 0.030, d = 0.62; VPI: t(14) = 3.44, P = 0.004, d = 0.89), whereas step lengths in CG did not change significantly (all P > 0.10). These spatiotemporal adaptations paralleled the improvements in gait speed and step-length symmetry, particularly in the VPI group, indicating a shift toward a faster and more symmetrical gait pattern.

Sagittal-plane joint kinematics

As illustrated in Fig. 5, sagittal-plane hip, knee, and ankle kinematics over two consecutive gait cycles (0–200% gait cycle) were analyzed using SPM1D to compare pre- and post-intervention waveforms for the affected and unaffected limbs. In the CG, no major time clusters survived correction, indicating minimal change in joint-angle trajectories. By contrast, both VOT and VPI produced distinct clusters of change at key gait phases: in late stance, hip extension increased and ankle plantarflexion during push-off was enhanced on the paretic side; in early stance, knee flexion during loading response became more pronounced; and during swing, knee flexion and ankle dorsiflexion increased, consistent with improved foot clearance. Importantly, these changes were accompanied by a visible convergence of affected- and unaffected-limb trajectories, indicating that interlimb asymmetries in joint motion were reduced after both visually augmented interventions, with VPI showing more extensive and sustained clusters of change than VOT. These SPM1D findings should be regarded as exploratory, but are consistent with the clinical improvements in gait speed and step-length symmetry, suggesting that VPI in particular promoted a more symmetrical and physiologically normal pattern of lower-limb joint kinematics during walking. No significant between-group time clusters were detected in the SPM1D ANCOVA.

Fig. 5.

Sagittal-plane flexion–extension kinematics at baseline and after the 12-week intervention across two time-normalized gait cycles (a–i). Ensemble-mean joint-angle trajectories (solid lines) with between-stride dispersion (shaded bands) are shown for the hip (a–c), knee (d–f), and ankle (g–i). Columns correspond to groups: CG (a, d, g), VOT (b, e, h), and VPI (c, f, i). In each subfigure, baseline and post-intervention waveforms are plotted for the unaffected (blue) and affected (red) limbs, as indicated in the legend. Time is normalized to 0–200% of the gait cycle (two consecutive cycles, each defined from non-paretic heel strike to the subsequent heel strike of the same limb). Positive values denote flexion for the hip and knee and dorsiflexion for the ankle. Waveforms summarize two valid gait cycles per participant and limb. Time-series differences between groups at post-test were evaluated using a one-way SPM1D ANCOVA (factor: group; covariates: baseline waveform metrics and comfortable gait speed)

Plantar-pressure and COP kinetic outcomes

As illustrated in Fig. 6a–d, plantar-pressure probability maps and COP trajectories revealed characteristic kinetic asymmetries after stroke. In the representative example, the paretic limb showed a rearfoot-dominant loading pattern with reduced first-metatarsal/hallux contribution, a laterally deviated and more curved gait line with micro-oscillations during early stance, and limited convergence of the COP trajectory at forefoot push-off, whereas the non-paretic limb exhibited a straighter, more centralized progression path with a clearly defined forefoot push-off zone. At the group level (Fig. 6e), paretic COP displacement increased significantly in both VOT (0.18 to 0.23 m; t(14) = 2.60, P = 0.021, Cohen’s d = 0.67) and VPI (0.18 to 0.24 m; t(14) = 3.03, P = 0.009, d = 0.78), but not in CG, indicating that both visual-based interventions enhanced roll-over and forward progression over the paretic limb. As mentioned above, COP symmetry ratio shifted toward 100% in all groups without significant adjusted between-group differences. These plantar-pressure and COP changes suggest that both VOT and VPI facilitated more effective paretic loading and center-of-mass progression, consistent with the improvements in gait speed and step-length symmetry, with larger gains in the VPI group.

Discussion

Summary of main findings

In this single-center randomized controlled trial in individuals with chronic stroke receiving standardized usual care and gait training, both visual oculomotor training and visual–proprioceptive integration produced additional improvements in gait performance, lower-limb motor function, and functional balance compared with baseline. Numerical improvements in BBS scores were observed in all three groups, with statistically significant within-group gains only in the two visual-based interventions, although between-group differences in balance change did not reach statistical significance.

Compared with the CG, VPI led to significantly greater gains in comfortable gait speed and step length symmetry, and also outperformed VOT with respect to step length symmetry. In contrast, VOT did not demonstrate statistically significant superiority over the control condition for these key gait outcomes. Although all three groups showed within-group improvements in the COP symmetry ratio, baseline-adjusted between-group differences in this measure did not reach statistical significance after correction for multiple comparisons. Both VOT and VPI groups demonstrated improvements in dynamic visual acuity (reduced ΔlogMAR) and FMA-LE scores, whereas changes in the CG were smaller. Taken together, these findings confirm our a priori hypothesis that both interventions would improve gait performance, with VPI yielding the larger gains in gait speed and step-length symmetry. Sensitivity analyses additionally adjusting for stroke subtype as a covariate yielded estimates and significance patterns that were consistent with the primary models, suggesting that the main conclusions are robust to stroke subtype.

Effects and potential mechanisms of visual-based interventions

Stroke survivors commonly exhibit oculomotor deficits—including abnormalities of saccades, smooth pursuit, and VOR function—which are recognized barriers to functional recovery and contribute to impaired postural control [10, 29]. The VOR is the primary mechanism that stabilizes gaze by generating head-motion–driven compensatory eye movements to preserve a stable visual scene [30]. VOR-oriented training combined with multisensory integration appears to strengthen anticipatory trunk-muscle control and augment dynamic gaze stability [31]. Such training may also increase excitability within vestibulospinal pathways, improving head–trunk stabilization and postural responses—pathways known to be compromised after stroke [32]. Thus, reinforcing VOR-related mechanisms through gaze-stability training may facilitate both upright stance and balance during gait. In the present study, improvements in dynamic visual acuity indicate enhanced VOR-mediated gaze stability in both visual-based groups. Enhancing gaze stability directly translated into improved gait function: participants were able to maintain clear vision at faster speeds, avoid vision-induced slowing, and thereby achieve greater gait velocity, accompanied by reduced double-support time, increased paretic-side single-limb support, and overall step-length elongation compared with the control condition.

Stable gaze is fundamental for the coordinated motion of the head, trunk, and pelvis during walking [29]; yet after stroke, reduced head–eye–trunk control degrades the quality of visual information and further compromises postural balance [15]. Stroke survivors frequently exhibit visual dependence—an overreliance on vision with reduced capacity to reweight toward vestibular and somatosensory inputs when visual cues are degraded or conflicting [33, 34]. After stroke, maladaptive sensory integration can limit this flexibility, promoting visual overreliance and underuse of vestibular and proprioceptive cues, which contributes to gait asymmetry [35]. An observational study in chronic stroke has shown that individuals with high visual dependence display disproportionate decrements in balance—greater sway, instability, and prolonged double-support durations—when visual information is removed or made incongruent, together with more asymmetric gait patterns [36]. The VPI program was explicitly designed to challenge such visual dependence by coupling gaze stabilization with visually guided, eight-directional weight-shift training that requires accurate shifts, particularly toward the paretic side. This form of multisensory balance practice—engaging visual, vestibular, and proprioceptive inputs—is thought to lower visual dependence and enhance stability when vision is reduced or unreliable [8, 9, 37]. In our cohort, improvements in ΔlogMAR, temporal phase proportions, and COP-related behavior suggest a shift away from excessive visual reliance toward more effective multisensory reweighting during gait. First, ΔlogMAR decreased in both intervention arms, and VPI outperformed control at post-test, indicating enhanced VOR-mediated gaze stabilization during head motion. Second, participants spent less time in double support and more time in paretic single-limb support, consistent with reduced “visual guarding” and greater reliance on vestibular–proprioceptive cues to maintain stability. Third, plantar-pressure analyses showed larger paretic-limb COP displacement together with COP-symmetry ratios drifting toward 100%, reflecting more confident paretic load acceptance and a more balanced left–right contribution to forward progression. Prior work shows that incorporating gaze-stability exercises can reduce postural sway during stance and improve dynamic balance (e.g., shorter Timed Up and Go) [29]. Consistent with this mechanism-based design, both interventions improved symmetry-related metrics, with VPI showing the largest numerical gains in step-length symmetry, paretic-side single-limb support, and COP-related measures, although COP-based between-group differences did not reach statistical significance. VOT most plausibly works by enhancing gaze stability and dynamic visual acuity during head motion, sharpening saccadic/pursuit control and visuomotor timing, and nudging sensory reweighting away from visual overreliance during head–eye–trunk coordination [38]. These mechanisms align with recent randomized trials and reviews showing that gaze-stability or oculomotor training improves balance and gait after stroke and can facilitate sensory reweighting [8, 15, 29]. Oculomotor exercises may also improve the temporal coordination of lower-limb movements; for example, virtual-reality paradigms that pair head-turning tasks with simultaneous visual and vestibular training have been shown to enhance ambulation in individuals with chronic vestibular hypofunction [39].

Three-dimensional gait analysis showed greater hip extension during late stance, increased knee flexion during loading, and enhanced ankle push-off and swing-phase dorsiflexion on the paretic side, yielding a more symmetric sagittal-plane pattern after VOT and especially VPI, consistent with improved paretic limb propulsion and swing clearance. These changes are consistent with more effective weight transfer and propulsion through the affected limb rather than compensatory reliance on the less-affected side. Plantar-pressure and COP analyses revealed larger increases in paretic-limb COP displacement in both intervention groups—particularly with VPI—indicating more complete loading and roll-over of the affected limb, whereas COP symmetry ratios tended to drift toward 100% across all groups without statistically significant between-group differences after baseline adjustment and multiplicity control. Taken together, these converging kinematic and kinetic adaptations suggest that visual-based interventions—especially when they explicitly couple gaze stabilization with weight shifting—may promote more efficient sensory reweighting and help restore a more coordinated and symmetric gait pattern after stroke.

Clinical implications and comparison with previous studies

Earlier studies have shown that various visual and vestibular-targeted interventions, including gaze-stabilization exercises and virtual reality–based protocols, can improve balance confidence, clinical balance scores, and overground walking speed in individuals with stroke or vestibular disorders [40, 41]. Our findings extend prior visual- and vestibular-targeted work on balance and gait after stroke, which was largely limited to clinical scales and lacked head-to-head comparisons of different visual-based paradigms. To our knowledge, this is the first randomized controlled trial to systematically contrast a visual oculomotor training program with a visual–proprioceptive integration paradigm, both delivered as adjuncts to usual care and gait training. The present results therefore provide more granular evidence that visual-based interventions can yield additive benefits beyond standard rehabilitation and suggest that training paradigms emphasizing multisensory integration, such as VPI, may be particularly effective for restoring gait symmetry.

These findings also have direct implications for clinical practice. VOT is inexpensive, requires minimal equipment, and can be implemented in a wide range of settings as a feasible option to address oculomotor dysfunction and improve dynamic visual acuity, especially in patients who initially tolerate only modest head motion or present with prominent visual complaints [42, 43]. In contrast, VPI demands access to a force platform or computerized posturography system but produced larger and more consistent improvements in gait speed, step-length symmetry, paretic-side single-limb support, and modest gains in functional balance, together with more favorable COP-related measures in our study. These BBS improvements, together with the biomechanical changes, suggest that targeting sensory integration may support improvements in balance capacity in daily-life tasks. For stroke survivors with marked gait asymmetry or insufficient loading of the paretic limb, particularly in the subacute and chronic phases, incorporating VPI into standard gait training may therefore represent a rational strategy to explicitly promote sensory reweighting and symmetrical weight bearing. Practically, clinicians might consider using brief bouts of VOT to enhance gaze stability and head-movement tolerance, followed by VPI sessions that employ multi-directional COP feedback to drive more symmetrical, paretic-side–oriented weight shifts during standing and walking.

Finally, our results suggest that visual-based interventions can be integrated into individualized rehabilitation prescriptions. Patients differ in the severity and nature of their visual, vestibular, and somatosensory impairments, as well as in their baseline gait patterns and degree of visual dependence [8]. Profiling these characteristics may help clinicians decide whether a patient is more likely to benefit from oculomotor-focused training, multisensory weight-shift based training, or a tailored combination of both. Future work incorporating formal assessments of sensory weighting and visual dependence could facilitate such patient stratification and guide the development of personalized, mechanism-based visual rehabilitation strategies for gait recovery after stroke.

Limitations and future directions

Retrospective trial registration. A key limitation of this study is that the randomized controlled trial was registered in a public trial registry only after participant enrolment had begun. The primary and secondary outcomes and the main analysis plan were defined a priori and strictly followed during data collection and analysis, and no changes were made to these pre-specified elements after trial registration. Nevertheless, retrospective registration may still introduce a risk of selective reporting and should be considered when interpreting the findings. Future clinical trials of similar interventions should be prospectively registered before enrolment of the first participant to maximise transparency and research integrity.

Sample size and single-center scope. As a single-center trial with a relatively modest sample, the present findings may be vulnerable to type II error for some outcomes and may not fully capture the heterogeneity of the broader stroke population. In addition, the specific rehabilitation environment may limit generalizability to other clinical settings. Replication in larger, multicenter cohorts that include more diverse and cross-cultural samples is needed to enhance external validity and to verify the robustness of the observed effects.

Short follow-up and uncertain durability/transfer. Outcome assessments were restricted to the immediate post-intervention period, so the durability of the observed improvements and their transfer to everyday activities remain uncertain. While VPI produced meaningful gains in gait speed and symmetry, we did not evaluate whether these changes are maintained over months or translate into reduced fall risk, increased community ambulation, or improved participation-level outcomes. Future studies should incorporate longer-term follow-up and ecologically valid functional measures to determine the persistence and real-world impact of visual-based gait interventions. In addition, we did not administer dedicated experimental measures of visual dependence or sensory reweighting (e.g., Sensory Organization Testing or rod-and-frame paradigms). Accordingly, our mechanistic interpretations rest on converging proxies—ΔlogMAR, phase proportions, and COP metrics—and should be confirmed with direct assessments in future trials.

Unexplored dose–response relationship. We used a fixed training dosage and progression scheme for both VOT and VPI. Consequently, the optimal frequency, intensity, and duration—as well as the most effective sequencing or combination of these two approaches—remain unknown. It is possible that alternative dosing schedules, longer intervention periods, or individualized progression criteria could yield larger or more sustained benefits. Future research should therefore explore dose–response relationships and compare different dosing regimens, including varying the relative emphasis on oculomotor versus multisensory COP-based training, to define the most efficient and effective training combination for different stroke subgroups, which together underscore the need for larger, longer-term trials of visual-based gait rehabilitation.

Conclusions

Within standardized usual care, this trial demonstrates the added value of vision-oriented rehabilitation for gait recovery after chronic stroke. Both VOT and VPI produced within-group improvements in overall gait pattern—reflected by longer step length, more symmetrical spatiotemporal parameters, and enhanced phase coordination across the hip, knee, and ankle. Statistically significant within-group gains in gait speed were observed only in the VPI group, whereas VOT primarily improved step length and symmetry-related spatiotemporal metrics. In both interventions, these changes were accompanied by COP-based indices consistent with a more balanced plantar load distribution. Improvements on key gait endpoints were generally larger with VPI, supporting multisensory integration as an effective target to correct gait asymmetry. Overall, these findings provide empirical support for a rehabilitation paradigm that “stabilizes vision to stabilize walking,” highlighting visual–vestibular–proprioceptive integration as a critical therapeutic target. This approach complements conventional strength- and task-oriented therapy and informs mechanism-based, individualized prescriptions for gait rehabilitation.

Abbreviations

AFO

Ankle-foot orthosis

ANCOVA

Analysis of covariance

ASL

Affected (paretic) limb step length

A-SS

Affected (paretic) single-limb support

BCVA

Best-corrected visual acuity

BBS

Berg Balance Scale

CG

Control group

CI

Confidence interval

COP

Center of pressure

COP-SR

Center-of-pressure symmetry ratio

DALYs

Disability-adjusted life years

DLS

Double-limb support

ETDRS

Early Treatment Diabetic Retinopathy Study

FMA-LE

Fugl–Meyer Assessment—Lower Extremity

HC3

Heteroscedasticity-consistent standard error estimator (type HC3)

logMAR

Logarithm of the minimum angle of resolution

RPE

Rating of perceived exertion

SD

Standard deviation

SLS

Step-length symmetry

SPM1D

One-dimensional Statistical Parametric Mapping

SVA

Static visual acuity

U-SS

Unaffected (non-paretic) single-limb support

USL

Unaffected (non-paretic) limb step length

VOR

Vestibulo-ocular reflex

VOT

Visual oculomotor training

VPI

Visual–proprioceptive integration

ΔlogMAR

Difference between dynamic and static visual acuity (logMAR[DVA] − logMAR[SVA])

Author contributions

HL conceived and designed the study, led data collection and analysis, and drafted the manuscript. QM and CB supervised the project, provided methodological guidance, and critically revised the manuscript. NZ, LZ, MX, and YN delivered the interventions, acquired data, and assisted with curation. YD developed analysis pipelines, processed data, and prepared figures. WL coordinated participant recruitment and outcome assessments. XF further refined the study concept and manuscript structure, contributed to data interpretation, and critically revised the manuscript. All authors interpreted the results, revised the manuscript for important intellectual content, and approved the final version.

Funding

This study was supported by the National Natural Science Foundation of China (11372223, 11102135); the Natural Science Foundation of Tianjin (17JCZDJC36000, 18JCZDJC35900); the scientific research projects of the General Administration of Sport of China (22KJCX077, 2022pqky-01, 24ZDKJCX11); the Tianjin Sports Bureau scientific research projects (22BZ02, 24BZ02); and the Tianjin Key Medical Discipline Construction Project (Grant No. TJYXZDXK-3–002 A-3).

Data availability

The datasets generated and analysed during the current study are not publicly available due to ethical restrictions related to participant confidentiality, but are available from the corresponding author upon reasonable request.

Declarations

Ethics approval and consent to participate

This study complies with the Declaration of Helsinki. The study has been approved by the Ethics Committee of Tianjin University of Sport, with approval number [TJUS2024-023]. Written informed consent was obtained from all participants prior to data collection.

Consent for publication

All participants provided written informed consent for publication of anonymized data.

Competing interests

The authors declare no competing interests.