Ocular Solute Movement Direction and Intracranial Clearance via Vitreous Gadolinium-based Contrast Agent MR Imaging: Potential as a Novel Biomarker for Glymphatic Dysfunction

Abstract

Purpose

The ocular and brain glymphatic systems may share common physiological pathways. We hypothesized that the anteroposterior movement of gadolinium-based contrast agent (GBCA) that has leaked into the vitreous could serve as a biomarker for brain glymphatic function. This study aimed to retrospectively investigate the association between the intravitreal GBCA distribution on MRI and recently proposed imaging markers of impaired brain waste clearance.

Methods

We analyzed 156 eyes from 78 adult participants who underwent 3T MRI 4 hr after standard-dose GBCA administration. On 3D-real IR images, we calculated a “contrast shift index” (the signal difference between anterior and posterior vitreous volumes of interest: VOIa-VOIp) to quantify the anteroposterior GBCA distribution. The primary outcome was a composite endpoint of positivity for either putative meningeal lymphatics at the posterior sigmoid sinus (PML-PSS) or enhanced basal ganglia perivascular spaces (PVS-BG). Multivariable logistic regression with cluster-robust inference was used to assess predictors, including the contrast shift index, mean vitreous contrast distribution, age, sex, and axial length of the eye.

Results

A positive contrast shift index, indicating preferential anterior GBCA distribution, was significantly and independently associated with the composite outcome of impaired brain clearance (P = 0.006). Age (P < 0.001) and male sex (P = 0.009) were also independent predictors. A predictive model incorporating these factors demonstrated high discrimination, with an area under the receiver operating characteristic curve (AUC) of 0.872. Axial length of the eye and mean vitreous contrast distribution were not significant independent predictors.

Conclusion

The anteroposterior distribution of GBCA in the vitreous is a novel, non-invasive imaging biomarker associated with impaired brain clearance function. This “contrast shift index” may reflect systemic glymphatic dysregulation common to both the eye and brain, offering a new avenue for assessing neurodegenerative risk.

Introduction

In the brain, the “glymphatic system,” a waste elimination mechanism via the microcirculation of cerebrospinal fluid (CSF), was proposed in 2012.¹ Recently, a similar “ocular glymphatic system” has been hypothesized in the eye, potentially serving as a drainage pathway for fluid and waste from the retina and optic nerve.^2–4 The glymphatic system significantly influences central nervous system physiological processes like nutrient distribution, waste clearance, and drug delivery.^5–7 While anterior chamber aqueous humor dynamics are well-studied, vitreous fluid dynamics, especially regarding drug distribution, remain largely unclear.^8,9 However, recent research suggests a new pathway where aqueous humor, after entering the vitreous, is absorbed into the retina via AQP4 in retinal Müller cells, indicating anterior-to-posterior water movement within the vitreous.¹⁰ This “retinal glymphatic pathway” within the ocular glymphatic system offers a new perspective on ocular fluid dynamics (Fig. 1).

Fig. 1.

A schematic diagram illustrating both the classical and emerging perspectives on aqueous humor circulation is presented. Traditionally, aqueous humor exits the posterior chamber, passes through the pupil into the anterior chamber, and then drains into Schlemm’s canal via the trabecular meshwork. Additionally, it may flow through intercellular spaces at the iris root, ciliary muscle fibers, and other uveal structures—collectively referred to as the uveoscleral route (indicated by red arrows). The newly proposed perspective involves the postiridial flow and the retinal glymphatic pathway (shown by blue arrows). In this model, the inflow of aqueous humor through the retinal glymphatic pathway is mediated by aquaporin-4 channels expressed in Müller cells within the retina.Co, cornea; L, lens; R, retina; S, sclera; U, uvea; V, vitreous body. This figure is adapted from Ueki, S.; Suzuki, Y. “New Perspective on Aqueous Humor Circulation: Retina Takes the Lead.” Int. J. Mol. Sci. 2025, 26, 2645. https://doi.org/10.3390/ijms26062645 under the terms and conditions of the Creative Commons Attribution (CC BY) license.

MRI, safe for human studies, has seen the development of sensitive pulse sequences (e.g., heavily T2-weighted 3D-fluid attenuated inversion recovery [hT2w-3D-FLAIR], 3D-real IR) to detect low concentrations of gadolinium-based contrast agents (GBCAs) in fluids.^11–15 These techniques have revealed “peripheral retinal leakage (PRL),” where IV-GBCA leaks from the inferotemporal peripheral retina (near the ora serrata) into the vitreous and diffuses over time.¹⁶ Ten minutes post-IV-GBCA, punctate or band-shaped leakage is seen; at 4 hr, it diffuses like a cloud; and at 24 hr, it’s nearly uniform throughout the vitreous.¹⁶ This finding resembles fluorescein angiography leakage in healthy eyes,¹⁷ and is age-dependent, suggesting increased blood-retinal barrier permeability with age, affecting vitreous homeostasis.^18–20 Brain glymphatic system and blood-brain barrier (BBB) permeability is also age-dependent, with more GBCA leakage into cortical perivenous spaces in individuals over 37.^21–23 The BBB permeability also varies with sex, location, and age.^24–26

We’ve observed numerous 3D-real IR images 4 h post-IV-GBCA in suspected endolymphatic hydrops patients.^27–30 The GBCA “cloud” diffusion direction in the vitreous varied, sometimes appearing more anterior or posterior, with some asymmetry. Although not explicitly stated in the initial PRL report, the images show posterior movement of peripheral retinal enhancement from 10 minutes to 4 hr, with discernible left-right asymmetry in movement distance (Fig. 2).¹⁶ While GBCA leakage from the inferotemporal peripheral retina is typically seen 10 minutes post-IV-GBCA, most endolymphatic hydrops patients have data only at 4 hr. However, anteroposterior vitreous GBCA movement can likely be assessed from these 4-hr images, considering the constant leakage point.

Fig. 2.

3D-real IR images were acquired at 4 time points: before (a), and at 10 minutes (b), 4 hr (c), and 24 hr (d) after intravenous administration of a gadolinium-based contrast agent (IV-GBCA) in a female patient in her 70s with suspected endolymphatic hydrops. Axial slices at the level just below the lens edge are shown for all time points (a–d). At 10 minutes post-contrast (b), punctate or band-like patterns of GBCA leakage are visible in both eyes, primarily originating near the ora serrata on the temporal side (arrows in b). By 4 hr post-contrast (c), this enhancement spreads into the vitreous, forming cloud-like patterns (arrows in c). Notably, these clouds display asymmetric movement posteriorly, with the cloud on the left side appearing further posterior than that on the right, suggesting a possible difference in postiridial and retinal glymphatic flow between the eyes. At 24 hr (d), enhancement within the vitreous becomes uniform.GBCA, gadolinium-based contrast agent; IR, inversion recovery; IV, intravenous. Reprinted 3D-real IR images from Naganawa S, Ito R, Kawamura M, Taoka T, Yoshida T, Sone M. Peripheral Retinal Leakage after Intravenous Administration of a Gadolinium-based Contrast Agent: Age Dependence, Temporal and Inferior Predominance and Potential Implications for Eye Homeostasis. Magn Reson Med Sci. 2023 Jan 1;22(1):45-55. doi: 10.2463/mrms.mp.2021-0100, under the terms and conditions of the Creative Commons Attribution-NonCommercial-NoDerivatives (CC BY-NC-ND) license.¹

IV-GBCAs are known to leak into CSF and basal ganglia perivascular spaces (PVS) in healthy subjects without significant BBB disruption.^{11,12,31–35} CSF GBCA signal intensity significantly increases at 4 hr post-IV-GBCA, then decreases by 24 hr.^34,36,37 This CSF GBCA clearance rate (4 to 24 hr) is presumed to reflect brain glymphatic system function. Delayed clearance is often seen in patients with the “putative meningeal lymphatics at the posterior wall of the sigmoid sinus (PML-PSS)” and the positive enhancement of basal ganglia perivascular spaces (PVS-BG). While the ocular and brain glymphatic systems are not anatomically continuous, they share a common physiological basis in AQP4-mediated fluid dynamics. It is therefore plausible that systemic factors, such as aging, could simultaneously affect the function of both systems. In the present study, we hypothesized that the direction of movement of the “cloud” of GBCA leaked into the vitreous, i.e., ocular glymphatic function, might be related to brain glymphatic function, particularly brain waste elimination function.

The purpose of this retrospective study was to investigate the association between the anteroposterior distribution of intravitreal GBCA and MRI markers of impaired brain waste clearance (PML-PSS and PVS-BG), considering clinical and anatomical factors such as mean vitreous contrast distribution, age, sex, and axial length of the eye.

Materials and Methods

Participants and MR imaging

Between August 2017 and September 2018, we retrospectively reviewed 156 eyes from 78 adult participants (34 males, 44 females; age 23-81 years, median 49 years, Q1:41, Q3:64, IQR:23) who underwent 3D-real IR, positive perilymph image: hT2W-3D-FLAIR and positive endolymph image: shorter inversion time version of the hT2W-3D-FLAIR for HYDROPS (hybrid of reversed image of positive endolymph signal and native image of positive perilymph signal) image generation acquisition,³⁸ and MR cisternography (MRC) 4 hr after IV-GBCA of standard dose due to suspected endolymphatic hydrops. This patient cohort was selected because the diagnostic imaging protocol for endolymphatic hydrops routinely included the 3D-real IR sequence at 4 hr post-GBCA administration, providing a consistent dataset for this analysis. Among the 89 initial participants, 11 patients with a history of cataract surgery, intraocular lens implantation, or other prior lens or vitreoretinal procedures were excluded. No patients reported a history of ocular disease during medical history taking, but intraocular pressure and visual acuity were not measured.

All MRI scans were performed on a 3T MRI unit (Skyra, Siemens Healthineers, Erlangen, Germany) using a 32-channel head coil. After scout imaging, MRC, hT2W-3D-FLAIR for the positive perilymph image and a shorter inversion time version of the same sequence for the positive endolymph image to generate HYDROPS images, and 3D-real IR were acquired, all with the same field of view and at 1 mm slice thickness in the axial plane. Detailed imaging parameters are shown in Table 1.

Table 1.

Pulse sequence parameters.

Parameter	MR-cisternography	Heavily T2-weighted 3D-FLAIR for positive perilymph image	Heavily T2-weighted 3D-IR with shorter inversion time for positive endolymph image	3D-real IR
Sequence type	SPACE with restore pulse	SPACE with inversion pulse	SPACE with inversion pulse	SPACE with inversion pulse
Slab orientation	Axial	Axial	Axial	Axial
TR (ms)	4400	9000	9000	15130
TE (ms)	544	544	544	549
Inversion time (ms)	N.A.	2250	2050	2700
Fat suppression	CHESS	CHESS	CHESS	CHESS
Flip angle (degree)	90/constant 120	90/constant 120	90/constant 120	90/constant 145
Section thickness/gap (mm)	1.0/0.0	1.0/0.0	1.0/0.0	1.0/0.0
Pixel size (mm)	0.5 × 0.5	0.5 × 0.5	0.5 × 0.5	0.5 × 0.5
Number of slices	104	104	104	256
Echo train length	173	173	173	256
FOV (mm)	165 × 196	165 × 196	165 × 196	165 × 196
Matrix size	324 × 384	324 × 384	324 × 384	324 × 384
Parallel imaging/Accel. factor	GRAPPA/2	GRAPPA/2	GRAPPA/2	GRAPPA/3
Band width (Hz/Px)	434	434	434	434
Number of excitations	1.8	2	2	1
Scan time (min)	3	7	7	11

CHESS, chemical shift selective; FLAIR, fluid attenuated inversion recovery; GRAPPA, generalized auto-calibrating partially parallel acquisition; 3D-real IR, 3D real inversion recovery with phase sensitive reconstruction; NA, not applicable; SPACE, sampling perfection with application-optimized contrasts using different flip angle evolutions;.

Evaluation of endolymphatic hydrops followed the Nakashima grading scale,³⁹ where the degree of endolymphatic hydrops in the left and right cochlea and vestibule was documented as none, mild, or significant in the radiological reports. In this study, if significant hydrops was present in at least one location in either the left or right cochlea or vestibule, the participant was classified as endolymphatic hydrops positive; otherwise, they were classified as negative.

Observers

After evaluating endolymphatic hydrops, the following items were evaluated with an interval of more than 5 years, without observing the inner ear region of the images. One neuroradiologist with 35 years of experience (S.N.) was responsible for the image evaluation. However, for the evaluation of positive visualization of the PML-PSS and the high signal of PVS-BG, another neuroradiologist with more than 30 years’ experience (T.T.) also participated. If there were discrepancies between 2 observers, consensus was obtained after discussion. The agreement between the 2 observers was assessed by Cohen’s κ coefficient.

Image analysis details

Images were displayed and evaluated on a picture archiving and communication systems (PACS) viewer (RapideyeCore; Canon Medical Systems, Tochigi, Japan) for all the evaluation except for the eyes. The other PACS system (Synapse SAI viewer, Fujifilm, Tokyo, Japan) was employed for the evaluation of the eyes. The variations in PACS devices are due to the point in time when image assessments were performed.

The presence or absence of high signal in PML-PSS and PVS-BG was subjectively evaluated on the heavily T2-weighted 3D-FLAIR images obtained 4 hr after IV-GBCA. The criteria were the same as in previous studies.³⁶ Briefly, they are as follows. A positive visualization of the PML-PSS was defined as a distinct high signal band-shaped structure between the posterior wall of the sigmoid sinus and the cerebellar hemisphere, which was at least 1 mm thick and at least 5 mm in length along the venous sinus, on either side on the axial hT2w-3D-FLAIR image. The presence or absence of high signal in the PVS-BG was defined in reference to a previous report as follows.³⁶ In the axial MRC slice, which included the AC-PC, the presence of PVS-BG was confirmed if the perivascular space in the basal ganglia had a dilation of at least 1 mm in diameter. The detected PVS-BG was confirmed as having a high signal if the PVS-BG was equal to or higher than the intensity of the pyramidal tract on the hT2w-3D-FLAIR images obtained 4 hr after IV-SD-GBCA. A high signal, presumed to be the contrast enhancement of the PVS-BG, was considered positive if it was observed on at least one side of the brain. These method, which demonstrated a high level of agreement between two observers in the previous study (Cohen’s κ: PML-PSS 0.808, PVS-BG 0.810), was performed by 2 experienced neuroradiologist described above. This ensured consistency and objectivity in the evaluation.³⁶ The window width/level of hT2w-3D-FLAIR images was set to 140/50.

Furthermore, after an interval of more than 6 months of the PVG-BG and PML-PSS evaluation, 2 spherical volumes of interest (VOIs) with a diameter of 3 mm were drawn within the eyeball on 3D-real IR images at the slice level of the inferior lens margin (Fig. 3: Example image of region of interest (ROI) setting) in the enlarged view of the eyes. At this time, the radiological reports were not referenced, nor were the inner ear region, basal ganglia area, and sigmoid sinus area of the images observed. Referring to the MRC at the same slice level, the anterior VOI (VOIa) on the 3D-real IR image was set immediately posterior to the inferior lens margin on the axial slice at the center of the lens, ensuring the VOI did not overlap with the lens. Another posterior VOI (VOIp) was set to be contiguous with the posterolateral aspect of VOIa, such that the center points of VOIa and VOIp were equidistant from the inferotemporal peripheral retina and the posterior end of the ciliary body at this slice. The images were scrolled to confirm that the spherical VOIs did not extend outside the vitreous body. Signal values for VOIa and VOIp were measured separately for the left and right eyes. The difference in signal values between VOIa and VOIp for each eye was defined as contrast shift index (VOIa-VOIp), serving as an index of the anteroposterior difference in GBCA distribution within the vitreous. During 3D-real IR measurements, only the eyeball region was displayed magnified, and the PML-PSS and PVS-BG areas were obscured. The 2 VOI settings were designed to detect differences in the anteroposterior movement of PRL GBCA and to ensure reproducibility. This specific VOI placement was chosen to sensitively capture the anteroposterior signal gradient, avoid partial volume artifacts, and account for the primary GBCA leakage source near the ora serrata. Spherical, rather than planar circular, regions of interest were used to minimize the influence of slight slice differences between left and right eyes.

Fig. 3.

An example of the VOI setting in the right vitreous. In the axial slice of the lowest edge of lens, a spherical VOI with 3 mm diameter was placed posterior to the center of lens for the VOIa. For the VOIp, spherical VOI with 3 mm diameter was positioned laterally posterior to VOIa, with the peripheral edges in contact. Additionally, it is arranged so that the center points of VOIa and VOIp are equidistant from the temporal retinal margin of this slice and the posterior end of the ciliary body (as indicated by the thick dashed line and thick solid line). Images were also scrolled to confirm that there is no sphere extending outside the vitreous body.VOI, volume of interest; VOIa, VOI in the anterior part; VOIp, VOI in the posterior part.

The mean vitreous contrast distribution (VOIave) was calculated for each eye as the average of the VOIa and VOIp values, serving as an indicator of the degree of GBCA leakage into the vitreous.

In 3D-real IR, the axial length of each eye was measured (Fig. 4). Axial length was defined as the distance from the anterior-most edge of the corneal center, drawing a straight line perpendicular to the lens, to the posterior-most edge of the vitreous on the inner retinal surface (near the macula). Axial length was measured to evaluate its potential influence as a confounding factor, as anatomical variations associated with myopia could affect intravitreal solute dynamics.

Fig. 4.

Example of axial length measurement in the right eye. In the axial slice of the central part of the lens acquired from 3D-real IR, a straight line is drawn perpendicularly from the most anterior edge of the corneal center to the last edge of the vitreous at the retinal inner surface (near the macula), defining this distance as the axial length.

To investigate the possibility of within-subject correlation while treating the left and right eyes of a single participant separately, Subject IDs were assigned to participants to enable the examination of robust variance across subject clusters.

The integrated positive outcome was defined at the participant level as positivity of either basal ganglia perivascular spaces (PVS-BG) or putative meningeal lymphatics at posterior sigmoid-sinus (PML-PSS) on MRI (positive or negative). These 2 markers, representing dysfunction in meningeal lymphatic drainage and perivascular clearance respectively, were combined into a single composite endpoint to more comprehensively capture overall brain clearance impairment and to increase statistical power for the analysis. This subject-level outcome label was then assigned to both eyes of the corresponding participant.

Predictors for the the integrated positive outcome

Candidate predictors included: Age (years), Sex, endolymphatic hydrops (EH) (positive or negative), and eye-specific measurements for each eye: Axial length (AxLen), VOIave, and the anteroposterior contrast shift index VOIa–VOIp (“Shift”).

Primary and extended models

We prespecified 2 logistic models (eyes treated as independent observations, as per protocol):

• Model 1 (core): Age + Shift

• Model 2 (full): Age + Shift + Sex + EH + AxLen + VOIave

In a sensitivity analysis, we additionally tested an interaction term (Age × Shift) on top of Model 2.

Statistical analysis

Interobserver agreement between observer 1 and observer 2 was assessed for 3 binary variables: PVS-BG (positive = 1, negative = 0), PML-PSS (positive = 1, negative = 0), and integrated positive outcome (defined as positive if either PVS-BG or PML-PSS was positive). Cohen’s kappa coefficients (κ) were calculated for each variable to quantify agreement beyond chance. All κ values and their 95% confidence intervals (CIs) were computed using Python with the scikit-learn package. The 95% CIs were estimated using bootstrap resampling with 10000 iterations. Kappa values were interpreted according to the Landis and Koch criteria: <0.00, poor agreement; 0.00–0.20, slight agreement; 0.21–0.40, fair agreement; 0.41–0.60, moderate agreement; 0.61–0.80, substantial agreement; and 0.81–1.00, almost perfect agreement.

Univariate comparisons used the Mann–Whitney U test for continuous variables and χ² test for categorical variables. Multivariable modeling used a binomial generalized linear model (GLM, logit link) with cluster-robust (sandwich) covariance by subject ID to account for within-subject correlation between fellow eyes. We report adjusted odds ratios (ORs) with 95% CI and cluster-robust P-values.

Model discrimination was quantified by the area under the receiver operating characteristic curve (ROC-AUC). To mitigate information leakage between fellow eyes, we performed GroupKFold cross-validation (k = 5) with the group variable set to subject ID and summarized the distribution of cross-validation (CV)-AUC across folds. For clinical operating points, Youden’s J was used to select the optimal probability threshold, and we report sensitivity, specificity, accuracy, positive predictive value (PPV), and negative predictive value (NPV) at that threshold.

We overlaid the 2 ROC curves (Model 1 and Model 2) in a single figure. Distributional differences by the integrated outcome were shown with violin-plus-box plots.

All analyses were conducted in Python 3.11.8 with the following libraries and versions: NumPy 1.24.0, pandas 1.5.3, SciPy 1.14.1, statsmodels 0.13.5 (GLM with cov_type = "cluster"), scikit-learn 1.1.3 (GroupKFold CV, logistic regression, ROC utilities), and matplotlib 3.6.3 (graphics). Code executed in a controlled environment; data handling used pandas, univariate tests used scipy.stats, GLM fitting used statsmodels.api.GLM(family = Binomial), cross-validation used sklearn.model_selection.GroupKFold, and ROC/Youden analyses used sklearn.metrics (roc_curve, roc_auc_score, confusion_matrix).

This retrospective study has been approved by the institutional medical ethics committee with a waiver of informed consent (2025-0162).

Results

Representative case images are presented (Figs. 5, 6, 7).

Fig. 5.

A 3D-Real IR image acquired 4 hr after intravascular administration of GBCA, comparing a case in a man in his 40s with a posterior GBCA distribution is dominant, (a), to a case in a woman in her 60s with an anterior distribution is dominant, (b). Both images show an enlarged view of the left globe. In both cases, the highest contrast enhancement is at the inferotemporal peripheral region of the retina. GBCA, gadolinium-based contrast agent.

Fig. 6.

Image of the right eyeball of a male patient in his 70s, showing a case of GBCA leakage into the vitreous body moving posteriorly. In the 3D-real IR image obtained 4 hrs after IV-GBCA administration (slice at the level of the lower edge of the lens, a), the leaked GBCA is recognized as a cloud-like area of high signal intensity, moving and dispersing posteriorly (indicated by arrows). The corresponding MR cisternography (b), where the contrast agent is not visible, but the shape of the vitreous body is clearly recognizable. MR cisternography (c) and heavily T2-weighted 3D-FLAIR image (d) at the level of basal ganglia. MR cisternography (e) and heavily T2-weighted 3D-FLAIR image (f) at the level of sigmoid sinus. Both the visualization of the PVS-BG and PML-PSS were negative on the heavily T2-weighted 3D-FLAIR image in this participant.FLAIR, fluid attenuated inversion recovery; GBCA, gadolinium-based contrast agent; IR, inversion recovery; IV, intravenous; PVS-BG, basal ganglia perivascular spaces; PML-PSS, putative meningeal lymphatics at the posterior sigmoid sinus.

Fig. 7.

Image of the right eyeball of a female patient in her 70s, showing a case of GBCA leakage into the vitreous body with a preferential anterior distribution. In the 3D-real IR image taken 4 hrs after IV-GBCA administration (slice at the level of the lower edge of the lens, a), the leaked GBCA is recognized as a cloud-like area of high signal intensity, remaining in the anterior portion (indicated by arrows). The corresponding MR cisternography (b), where the contrast agent is not visible, but the shape of the vitreous body is clearly recognizable. MR cisternography (c) and heavily T2-weighted 3D-FLAIR image (d) at the level of basal ganglia. MR cisternography (e) and heavily T2-weighted 3D-FLAIR image (f) at the level of sigmoid sinus. Both the visualization of the PVS-BG (arrow, c and d) and PML-PSS (arrows in e and f) were positive on heavily T2-weighted 3D-FLAIR image in this participant.FLAIR, fluid attenuated inversion recovery; GBCA, gadolinium-based contrast agent; IR, inversion recovery; IV, intravenous; PVS-BG, basal ganglia perivascular spaces; PML-PSS, putative meningeal lymphatics at the posterior sigmoid sinus.

The interobserver agreement was almost perfect for all three variables. Cohen’s kappa coefficients (95% CIs) were 0.810 (0.662–0.923) for PVS-BG, 0.827 (0.681–0.944) for PML-PSS, and 0.949 (0.870–1.000) for integrated positive outcome. Based on the Landis and Koch criteria, all κ values indicated almost perfect agreement between observers.

Univariate analyses of 156 eyes revealed that eyes from integrated positive participants exhibited significantly higher Age and larger Shift compared to integrated-negative participants (both P ≤ 0.006 by Mann–Whitney U). Axial length and VOIave also showed significant differences, while Sex and EH did not (Table 2, Fig. 8).

Table 2.

Univariate analyses between 2 groups.

Parameter	Negative group (mean ± SD/%)	Positive group (mean ± SD/%)	P-value	U statistic	Test
Age (years)	44.43 ± 11.69	60.61 ± 12.97	1.16 × 10−11	1116	Mann-Whitney U
Axial length (mm)	24.62 ± 1.55	23.64 ± 1.54	1.28 × 10−4	4101.5	Mann-Whitney U
Intravitreal Gd amount (VOIave)	−38.40 ± 11.39	−26.19 ± 22.83	1.04 × 10−4	1932	Mann-Whitney U
Contrast shift index (VOIa-VOIp)	−4.08 ± 8.38	4.44 ± 19.76	7.84 × 10−5	1914	Mann-Whitney U
Male (%)	38.1%	50.0%	0.183		Chi-square
With EH (%)	42.9%	30.6%	0.156		Chi-square

EH, endolymphatic hydrops; VOIa-VoIp, signal difference between anterior and posterior vitreous volumes of interest.

Fig. 8.

Violin and box plots comparing distributions of clinical parameters between the negative and positive groups (integrated outcome: positive if either PVS or PML-PSS is present). The violin plot depicts the full data distribution via kernel density estimation, with wider sections representing higher data frequency. Each plot includes a boxplot showing the median (horizontal line within the box), the interquartile range (box edges: Q1 and Q3), and whiskers extending to values within 1.5 × IQR. Points outside the whiskers are plotted as outliers. (a) Comparison of mean age between integrated negative and positive groups. Mean age is significantly higher in positive group (P = 1.16 × 10^–11). (b) Comparison of axial length of the eye between integrated negative and positive groups. Mean axial length is significantly longer in negative group (P = 1.28 × 10^–4). (c) Comparison of mean vitreous contrast distribution (VOIave) between integrated negative and positive groups. Mean vitreous contrast distribution is significantly higher in positive group (P = 1.28 × 10⁻⁴). (d) Comparison of mean contrast shift index (VOIa-VOIp) between integrated negative and positive groups. Mean contrast shift index (VOIa-VOIp) was higher in the positive group than the negative group (P = 7.84 × 10^–5).PVS-BG, basal ganglia perivascular spaces; PML-PSS, putative meningeal lymphatics at the posterior sigmoid sinus; VOIa-VoIp, signal difference between anterior and posterior vitreous volumes of interest; VOIave, mean vitreous contrast distribution.

Multivariable logistic regression analysis, with cluster-robust inference to account for within-subject correlation, demonstrated that Age (P < 0.001) and the contrast shift index (Shift; VOIa-VOIp, P = 0.006) were independently associated with the integrated positive outcome. Male sex was also identified as a significant predictor (P = 0.009). In the full model (Model 2), axial length, VOIave, and the presence of endolymphatic hydrops (EH) did not demonstrate independent predictive value (Table 3).

Table 3.

Multivariable logistic regression analysis.

Variable	Coefficient	Std Error	z value	P-value	95% CI lower	95% CI upper
Intercept	−0.2	4.59	−0.044	0.965	−9.19	8.79
Age	0.093	0.023	4.01	0.00006	0.048	0.139
Axial length	−0.164	0.155	−1.06	0.29	−0.467	0.139
VOIave	0.025	0.023	1.1	0.27	−0.019	0.069
VOIa-VOIp	0.074	0.027	2.75	0.006	0.021	0.128
Gender (Male)	1.18	0.456	2.59	0.0095	0.289	2.08
With EH	−0.852	0.472	−1.8	0.071	−1.78	0.073

EH, endolymphatic hydrops; VOIa-VoIp, signal difference between anterior and posterior vitreous volumes of interest.

A summary of the modeling analyses highlights the robustness of these findings. A parsimonious model including only Age and Shift already provided strong discrimination (in-sample AUC ≈ 0.837) with good cross-validated performance (Model 1). The full model (Model 2), including Sex, EH, AxLen, and VOIave, yielded a modest improvement in discrimination (in-sample AUC ≈ 0.872), while the addition of an Age × Shift interaction term offered negligible performance gain (Fig. 9). Youden-optimized operating points for the models achieved balanced sensitivity and specificity, supporting the clinical utility of these markers. The minimal increase in AUC from the core to the full model suggests that Age and the “Shift” index capture the vast majority of the predictive information, highlighting their primary importance as biomarkers.

Fig. 9.

ROC curve for the logistic regression model predicting integrated positive status (PVS or PML-PSS positive). The resulting area under the curve (AUC) was 0.837 for model 1 (Age + Shift) and that was 0.872 for model 2 (Age + Shift + Sex + EH + Axial Length + VOIave), indicating high discrimination performance for both model. The dashed line represents the line of no discrimination (AUC = 0.5).EH, endolymphatic hydrops; PVS, perivascular spaces; PML-PSS, putative meningeal lymphatics at the posterior sigmoid sinus; ROC, receiver operating characteristic; Shift, contrast shift index (VOIa–VOIp).

Overall, these results support Age and VOIa–VOIp as practical, generalizable markers for anticipating positivity of brain waste-clearance impairment on MRI, with a simple 2-variable model offering a favorable trade-off between parsimony, interpretability, and performance.

Discussion

To our knowledge, this study is the first to demonstrate that the contrast shift index (VOIa-VOIp) is associated with an indicator of impaired brain clearance function. A positive contrast shift index, indicating a preferential anterior distribution of GBCA, may reflect dysfunction of the retinal glymphatic pathway, which is responsible for posterior-directed fluid movement from the vitreous into the retina. The observed link between this ocular phenomenon and brain clearance markers suggests a potential common underlying pathophysiology. This could involve systemic factors that affect AQP4 expression or function in both the eye and the brain, leading to concurrent impairment of fluid and solute transport in both compartments. It would be interesting to investigate how the contrast shift index behaves in ocular diseases such as glaucoma, high myopia, and age-related macular degeneration.

It is presumed that there is a close relationship between the water dynamics of the brain and sensory organs, particularly the eyes and inner ears,⁴⁰ and functional similarities between the brain’s glymphatic system and the ocular glymphatic system have also been pointed out.^41,42 The “retinal glymphatic pathway,” where aqueous humor flowing into the vitreous is absorbed into the retina via aquaporin-4 (AQP4) in retinal Müller cells, may play an important role in the anterior-to-posterior movement of solvents and solutes within the vitreous, and this function is presumed to decline with age.⁴³

Glaucoma is a major ocular disease characterized by progressive optic neuropathy, leading to irreversible visual impairment and blindness.^3,44 Elucidating its pathogenesis and developing early diagnostic and therapeutic methods remain critical challenges in ophthalmology.⁴⁴ Although elevated intraocular pressure (IOP) is widely recognized as a major risk factor for glaucoma development, how IOP elevation affects the retina and optic nerve, particularly the involvement of vitreous fluid which constitutes the majority of the eyeball, remains largely unclarified. Monitoring changes in the biomechanical properties and molecular level of the vitreous, a complex biological fluid, could offer valuable insights into understanding how damage occurring in the anterior segment propagates to the posterior segment.⁴⁴ Furthermore, there is a condition known as “normal tension glaucoma,” where glaucomatous optic neuropathy progresses despite IOP being within the normal range, and its detailed mechanisms are also unknown.^3,44 This suggests that despite the conventional understanding that elevated IOP is a major risk factor, many patients are already diagnosed with structural damage at the time of glaucoma diagnosis, necessitating the search for other more sensitive and specific biomarkers. In patients with normal tension glaucoma, the brain’s DTI-ALPS index was found to be reduced and was also associated with visual function and visual field narrowing.⁴⁵ Thus, the evaluation of the DTI-ALPS index, which is presumed to reflect brain glymphatic system function,^32,46–48 potentially serving as a biomarker for glaucoma, is an intriguing research finding that suggests a connection between the brain and sensory organs.⁴⁵ Further studies are awaited to investigate whether a preferential anterior distribution of PRL is a risk factor for glaucoma, and whether there is a correlation with the brain’s DTI-ALPS index.

A laser flare photometer is a device used in ophthalmology to quantitatively evaluate the protein concentration in the aqueous humor. It helps to quantify intraocular inflammation, such as after cataract surgery or in uveitis, aiding in early detection, follow-up, and evaluation of treatment efficacy. By detecting subtle changes in flare that are difficult to capture with a slit lamp, it becomes possible to grasp a more detailed state of the eye. In the eyeball, flare values can be easily measured. Flare has also been shown to correlate with the permeability of the blood-aqueous barrier.⁴⁹ Furthermore, a correlation has been shown between the concentration of IgG and albumin in the aqueous humor and the flare value.⁵⁰ In the future, it will be necessary to investigate the relationship between flare and the contrast shift index. If the Contrast shift Index obtained from MRI images 4 hours after contrast administration could be estimated from the flare index, which is far easier to obtain than the Contrast Shift Index, it would be beneficial.

This study has several limitations as follows. It is a retrospective single-center study with a limited number of cases. All participants were patients suspected of endolymphatic hydrops; there were no healthy controls. Intraocular pressure and other ophthalmic parameters were not measured. A future investigation to establish the normal range of Contrast Shift should ensure normality through ophthalmologic examinations. Although the VOI settings were manual, this study attempted to enhance reproducibility by defining a detailed protocol based on anatomical landmarks such as the inferior lens margin. Nevertheless, bias cannot be completely excluded. It has been more than 5 years since the diagnosis of endolymphatic hydrops, and the inner ear portion was not observed when measuring the eyes, so it is presumed that there is no influence. Similarly, the presence or absence of PML-PSS and PVS-BG was a subjective determination by 2 observers. In the previous study, the Cohen kappa’s coefficient (κ) between 2 observers for subjective evaluation of the PML-PSS was 0.808 (95% CI: 0.629–0.987). That for subjective evaluation of the PVS-BG was 0.810 (95% CI: 0.632–0.987). In the present study, Cohen’s kappa coefficients were 0.827 (95% CI: 0.681–0.944) for PML-PSS, 0.810 (95% CI: 0.662–0.923) for PVS-BG, and 0.949 (95% CI: 0.870–1.000) for integrated positive outcome.³⁶ Kappa values for the PVG and PML-PSS were comparable to the previous study. By using the integrated positive outcome, we could obtain even higher kappa value between 2 observers. Although the signal intensity values of 3D-real IR on the MR scanner used in this study are reported to be stable as long as the same coil is used and similar locations are measured,^{23,28,51–53} the contrast agent concentration was not quantified in this retrospective study. Distinguishing between increased vascular permeability and impaired clearance function is difficult with imaging at limited time points after IV-GBCA administration. Changes in GBCA movement patterns within the vitreous may be influenced by vitreous degeneration in addition to ocular glymphatics (vitreous fluid flow), but data on vitreous composition were not included in this study.

Despite these limitations, the results of this preliminary study indirectly suggest a relationship between the direction of solute movement within the ocular vitreous and the clearance of solutes in the intracranial CSF, providing a further impetus to investigate the connection between brain and sensory organ clearance functions. Future prospective studies should not only validate these findings but also explore whether the contrast shift index can predict future cognitive or visual decline. Furthermore, investigating whether similar information can be extracted from non-contrast MRI sequences or clinical indicators that can be obtained more easily, such as the flare-index, would be crucial for broader clinical translation.

Conclusion

This analysis confirms that both age and the contrast shift index (VOIa-VOIp) are robust independent predictors of impaired brain waste clearance, as assessed by the presence of either PVS or PML-PSS, on a per-eye basis. Additionally, male sex was found to be an independent risk factor in this cohort. Other clinical parameters, including axial length, mean vitreous contrast distribution, and the presence of endolymphatic hydrops, were not statistically significant predictors. The predictive model yielded a high AUC of 0.872, supporting the clinical applicability of these factors for non-invasive risk assessment of the glymphatic system.