On-going retinal neuronal degeneration despite clinical stability in glaucoma

Abstract

Glaucoma, a leading cause of irreversible blindness, involves progressive retinal ganglion cell loss. Early detection of this neurodegeneration is critical, yet current clinical methods often identify disease only after substantial damage has occurred. This preliminary study used high-resolution adaptive optics–optical coherence tomography (AO-OCT) to track individual retinal ganglion cells in patients and healthy volunteers over 3 to 5 years. Results revealed higher neurodegeneration in glaucoma compared to normal aging. We observed 4.16 to 6.86% annual decrease in cell density in glaucoma, far exceeding the 0.27% cell loss rate observed in the healthy volunteer group. Critically, AO-OCT detected progression significantly earlier, identifying cell loss 0.74 years before conventional OCT and 1.49 years before visual field testing. These findings demonstrate AO-OCT as a more sensitive tool capable of monitoring glaucomatous damage with superior precision, potentially representing a major advancement over current clinical imaging.

INTRODUCTION

Loss of retinal ganglion cells (RGCs) and their axons is the defining hallmark of optic neuropathies, such as glaucoma, where irreversible structural damage precedes perceptible vision loss (1–3). It has been shown in several studies that RGCs can be lost before significant anomalies are detected clinically on a visual field test (4, 5). Accounts of US (4.22 million) and worldwide (79.6 million) glaucoma incidence are likely underestimates (6, 7) owing to low visual field test sensitivity and other factors (e.g., health care use). Glaucoma’s slow, insidious progression poses a substantial challenge for early intervention, a problem compounded by the lack of sensitive biomarkers for detecting disease onset and tracking progression at a cellular level (8).

Clinically, glaucoma progression is monitored using parameters such as visual field mean deviation (MD), visual field pattern SD (PSD), visual field index, macular ganglion cell complex (GCC) thickness, and circumpapillary retinal nerve fiber layer (RNFL) thickness (8, 9). Clinical stability is defined as the absence of statistically detectable progression on those standard clinical measures over the study period. However, the diagnostic utility of these metrics is limited by low sensitivity for detecting early disease progression. For example, Zhang et al. (3) found that it took 5 years for glaucoma progression to be detected using MD, visual field index, or RNFL, while even the more sensitive GCC metric required 2.5 years. This delay allows a critical period in which irreversible vision loss may go undetected and untreated.

Advances in high-resolution retinal imaging, particularly adaptive optics (AO)–optical coherence tomography (AO-OCT), offer a transformative alternative by providing noninvasive, three-dimensional visualization of individual RGCs (10, 11). Cellular-level resolution provide a more direct and potentially sensitive metric to track RGC loss resulting from glaucomatous structural damage as compared to conventional measurements of retinal layer thickness or visual function (12, 13).

Since the seminal work demonstrating visualization of the RGC mosaic in living human eyes (10), reports of AO-OCT imaging in glaucoma have been largely cross-sectional or limited in the number of longitudinal follow-up visits (14). Our group previously used AO-OCT to document glaucomatous damage and its association with hemifield defects (12). Other studies have investigated inner retinal morphology at one or two follow-up visits, focusing on features like inner nuclear microcysts (15) and RGC loss near arcuate defects (13). While these studies established the potential of AO-OCT, longitudinal studies that monitor RGCs over several years are critical to understand disease progression and develop robust imaging biomarkers.

Applying AO-OCT to track glaucoma longitudinally has four primary challenges. First, to demonstrate benefit over existing methods and to link structural and functional changes, the small AO field-of-view images need to be precisely matched with functional data, accounting for the anatomical displacement between RGC receptive field and their corresponding visual field location at the photoreceptor level (16). Second, AO-OCT needs to distinguish extremely slow age-related cell loss from glaucomatous change (13). Detection and discrimination of these subtle differences necessitate exceptionally high sensitivity, and even with faster glaucomatous progression, the level of methodological rigor needed is extremely high. Third, to produce highly accurate quantification while avoiding bias from regional density variations, acquired AO-OCT volumes need to be registered with cellular-level precision (i.e., several pixels spatial difference) over several years. Fourth, glaucoma-related tissue remodeling makes tracking individual RGCs with accuracy exceptionally difficult. From our own cross-sectional study (12), we observed that ganglion cell layer (GCL) soma diameter increase (12.3 to 14.2 μm) coupled with density losses (40 to 50%) may suggest a potential compensatory mechanism, wherein remaining cells may undergo hypertrophy following the loss of adjacent cells. The challenges listed above are monumental and may benefit from a multipronged approach for the promise of sensitive glaucoma disease progression detection to be realized.

Here, we attempt to present such a multipronged approach for measuring GCL soma loss and detecting neuronal degeneration in glaucoma. We hypothesize that AO-OCT, by providing direct cellular-level measurements of GCL soma morphology, may detect glaucomatous neuronal loss with greater sensitivity than conventional clinical assessments, including structural OCT and functional visual field measurements. We used a multimodal AO system [with AO scanning laser ophthalmoscope (AO-SLO) and AO-OCT channels] to ensure precise reimaging of the same retinal locations over time, as shown in Fig. 1. We account for the displacement between RGCs and photoreceptors in the macula when the AO-OCT results are compared to visual function data. We developed a new registration algorithm for longitudinal analysis (Fig. 2) and used a deep learning algorithm (17) to achieve highly accurate cell soma segmentation. Furthermore, we structured our study to assess GCL soma loss in glaucoma with respect to hemifield defects. This allowed us not only to probe the disease where there is known differential cell loss but also to use the participant as their own reference to understand age-dependent and glaucoma-related changes. Overall, our preliminary study provides a robust stratagem for disambiguating the dual sources of potential cell loss (i.e., age-related decline and glaucomatous pathology) within the common GCL regions imaged across longitudinal visits, thereby validating the increased sensitivity of AO-based biomarkers to potentially advance glaucoma monitoring.

Fig. 1. Multimodal assessment of a patient with glaucoma (4179) and an inferior hemifield defect.

(A) Spectralis SLO and OCT macular scans with overlay of AO-OCT imaging locations. (B) En face projections from the GCL layer of the averaged AO-OCT volumes clearly reveal differences in GCL soma density and diameter across retinal eccentricities. The clinical data show inferior retinal layer thinning in (C) macular GCL thickness grid (8 × 8), (D) posterior pole retinal thickness hemisphere asymmetry map, and (E) peripapillary RNFL thickness classification. The corresponding superior visual field defect is seen in (F) 10-2 and (G) 24-2 visual field PD maps. Note that retinal locations are inverted relative to the visual field (e.g., the superior visual field corresponds to the inferior retina).

Fig. 2. Example of longitudinal AO-OCT imaging of the same retinal location.

(A) Imaging region for six visits over 3 years for a patient with glaucoma (4179) shows unregistered and registered average images where registration used the intermediate capillary plexus layer en face projections. (B) Average AO-OCT en face images at the same retinal location show well-resolved somas longitudinally. (Also see movie S2.) (C) GCL soma segmentation map shows cell density decrease and soma enlargement over time. Segmented GCL somas are color coded by cell diameter. Scale bars, 50 μm.

RESULTS

Participants and clinical observations

The glaucoma and age-matched healthy volunteer cohorts (Table 1) showed no significant differences in terms of age, axial length, and intraocular pressure. All patients with glaucoma used pressure-lowering drops as directed by their physician (O.J.S.), which kept their intraocular pressures stable and within normal human physiological range over the course of this study: 11.6 ± 2.0 mm Hg (1541), 16.4 ± 1.3 mm Hg (1733), 12.8 ± 1.8 mm Hg (4179), 15.9 ± 1.4 mm Hg (4856), and 15.7 ± 2.1 mm Hg (8533).

Table 1. Imaging participants summary for the glaucoma and healthy volunteer groups.

G, glaucoma; HV, healthy volunteer.

Subject ID	Type	Age* (years)	Eye	Axial length (mm)
1541	G	66	OD	23.64
1733	G	57	OS	25.98
4179	G	63	OS	24.06
4856	G	71	OD	22.74
8533	G	67	OS	24.64
6242	HV	64	OD	23.28
6289	HV	60	OD	24.90
6809	HV	73	OS	24.52
9165	HV	77	OS	24.48
9235	HV	68	OD	22.25

Baseline differences between glaucomatous and healthy eyes

At the first visit, GCL soma density from AO-OCT showed varied distribution across the macula for both healthy volunteers and patients with glaucoma. The GCL soma density for healthy eyes was (mean ± SD): 24,114 ± 4199, 16,885 ± 3774, and 4289 ± 874 cells/mm² at 3°, 6°, and 12°, respectively. As expected, higher soma density was observed at 3°T compared to 12°T (12, 18). Glaucoma eyes demonstrated reduced densities of 14,437 ± 9736, 8670 ± 3184, and 3043 ± 1022 cells/mm² at corresponding locations, as shown in Fig. 3A. The symmetry index, calculated from GCL soma density ratio [superior-inferior (S-I)] at the same retinal eccentricity: symmetry index = min[density(S,I)]/max[density(S,I)] was significantly lower in the patients with glaucoma (0.56 ± 0.30; P = 0.02) than in the healthy volunteers (0.82 ± 0.14), as shown in fig. S1A. No significant variation in symmetry index was observed across different eccentricities in either of the cohorts. The GCL soma diameter in patients with glaucoma was significantly larger (P < 0.05) across all eccentricities compared to healthy volunteers, as shown in Fig. 3B. The largest soma diameters in patients with glaucoma were measured at 12° (15.6 ± 0.9 μm) and were progressively smaller at 6° (13.6 ± 1.3 μm) and 3° (13.1 ± 1.0 μm). However, these values were consistently larger than those of the healthy volunteer group, which were measured to be 14.3 ± 1.3 μm at 12°, 12.8 ± 0.4 μm at 6°, and 11.7 ± 0.4 μm at 3°. The differences represented respective increases of 12, 11, and 9% compared to healthy volunteers, similar to our previous cross-sectional study (12).

Fig. 3. Baseline quantification for healthy (blue) and glaucoma (orange) cohorts.

(A) AO-OCT–derived GCL soma density with eccentricity showed agreement with histology (shaded curve) for healthy volunteers (18) and significantly lower for patients with glaucoma. The histologic curve represents data from the temporal meridian only, with the shaded area indicating the SD from the mean density. (B) GCL soma diameter measured from AO-OCT images increased with eccentricity and was significantly larger (P < 0.05) in patients with glaucoma compared to healthy volunteers. (C) GCL thickness derived from clinical OCT was reduced in patients with glaucoma. (D) More diseased hemifield shows reduced visual field PD (function) corresponding to hemifield defect. (E) The correlation between soma density (AO-OCT) and GCL thickness (clinical OCT) was strong in healthy volunteers (blue line) but weaker in glaucoma (orange line). (F) Normalized GCL soma density (with respect to healthy volunteers at each location) showed correlation with PD indicating simultaneous structural and functional loss with regional differences. Statistical significance was determined using a Student’s t test (*P < 0.05 and **P < 0.01).

Clinically, both structural and functional measurements in the retina were assessed using macular OCT scans in both cohorts and standard visual field tests in the glaucoma cohort. We extracted GCL thickness from macular OCT scans and pattern deviation (PD) values from visual field reports for each visit, corresponding to six AO imaging regions. We selected PD for the primary structure-function analysis because it isolates localized functional deficits, corrects for diffuse loss from media opacities (e.g., cataract), and provides a metric that corresponds more directly to the focal retinal area imaged by AO-OCT. GCL thickness decreased with increasing eccentricity in both cohorts, with patients with glaucoma demonstrated significantly lower thickness at 3° (P = 0.008) and 6° (P = 0.012) eccentricities compared to healthy volunteers, but not at 12° (P = 0.691; Fig. 3C). Visual field PD values in patients with glaucoma were consistent with their hemifield defects (Fig. 3D). For instance, patient 4179 exhibited a substantially reduced superior PD value (−22.9 dB) that corresponds to the 3° inferior retinal location. Clinical OCT and visual field measures at corresponding AO-OCT imaging locations are summarized in tables S1 and S2 for the baseline visit.

The GCL soma density derived from AO-OCT showed a strong correlation with local GCL thickness from clinical measurements in the healthy volunteer group (r = 0.89) but only a weak correlation in the glaucoma group (r = 0.68), as shown in Fig. 3E. Other clinical metrics, such as RNFL, GCC, and IPL thickness, showed a weaker correlation with AO-OCT compared to GCL thickness in both healthy (r = 0.48, 0.84 and 0.82, respectively) and glaucoma (r = 0.14, 0.66, and 0.59, respectively) cohorts, as shown in fig. S1 (B to D). These findings suggest that, although GCL thickness is the best surrogate for estimating GCL soma density among conventional OCT (thickness) measures, it is a less direct surrogate for GCL soma density in the early to moderate stages of glaucoma that our study cohort represents. The AO-OCT structural measurements of GCL soma density shows moderate to strong correlation with some functional visual field PD measurements, with coefficient of determination ranging from 0.29 to 0.77 (P < 0.05 at 3° and 6° and P = 0.11 at 12°). A slope less than 1 indicates that structural losses precede functional losses at 12°. The elevated slope at 3° in Fig. 3F is likely attributable to a single high PD value observed in one patient. Despite regional differences, the data demonstrated a moderate to strong correlation between PD and GCL soma density in patients with glaucoma.

Long-term morphological assessment

We achieved overlapping regions longitudinally over 3 years, spanning 1° ± 0.2° for healthy volunteers and 0.9° ± 0.3° for patients with glaucoma (from 1.5° acquisition fields of view or ∼66% overlap). However, ∼10.6% (19 of 180) of the imaging targets deviated from the baseline visit, resulting in regions with no longitudinal overlap. These deviations, primarily caused by fixation changes, still yielded quantifiable data. Further, patient 1541 exhibited substantial cell loss in the inferior region at both 3° and 6°, resulting in an insufficient number of cells to accurately quantify soma density.

The healthy volunteers had an average annual GCL cell loss rate of 0.27% (see movie S1). Spatially, the loss showed no significant difference between the superior and inferior hemifields across any eccentricity (3°: P = 0.193; 6°: P = 0.174; 12°: P = 0.367; paired t test). The average GCL cell loss rate varied substantially among patients with glaucoma, ranging from 1.15 to 27.9%/year (mean ± SD = 9.6 ± 7.0 for all patients and locations), with clear regional variation (see movies S2 and S3). Locations were designated as “more diseased” or “less diseased” based on the identified hemifield defect for each patient with glaucoma; specifically, the three superior or inferior retinal locations (at 3°, 6°, and 12°) corresponding to the visual field hemifield with greater PD loss were classified as more diseased. We found that the GCL cell loss rate was higher in the more diseased hemifield (−6.86%/year) compared to the less diseased hemifield (−4.16%/year), as shown in Fig. 4A. We observed a significant increase in soma size in patients with glaucoma compared to healthy volunteers (all locations combined, P = 0.03; Figs. 4B and 5) of 1.52%/year in the more diseased hemifield and 0.95%/year in the less diseased hemifield. In contrast, healthy volunteers showed no significant change (P = 0.27) in soma size over the study period.

Fig. 4. Longitudinal follow-up imaging of healthy volunteers and patients with glaucoma.

(A) Normalized GCL soma density measured by AO-OCT shows a minimal cell loss rate in healthy volunteers but a substantially higher cell loss rate in patients with glaucoma, with regional variations corresponding to hemifield defects. (B) GCL soma diameters showed a significant increase (P = 0.03) in patients with glaucoma compared to healthy volunteers, with higher rates of enlargement in the more diseased hemifield than the less diseased hemifield, consistent with dynamic cellular remodeling; healthy volunteers showed no significant change (P = 0.27) over the study period. (C) Normalized GCL thickness from clinical OCT scans at six retinal locations for a healthy volunteer shows age-related cell loss and, for patients with glaucoma, shows slightly higher reduction corresponding to hemifield defects. (D) Normalized PD measured at corresponding locations for patients with glaucoma showed a slight reduction in the more diseased hemifield. The value “y” represents the slope of the linear fit.

Fig. 5. AO-OCT–derived longitudinal individual measurements.

Longitudinal data of GCL soma density and diameter from AO-OCT measurements in healthy (blue) and glaucoma (orange) cohorts at 3°, 6°, and 12° eccentricity. In patients with glaucoma, solid and open symbols denote less diseased and more diseased hemifields, respectively.

The less diseased hemifield in patients with glaucoma showed no significant difference (P = 0.19) in GCL thinning (0.87%/year) compared to healthy volunteers (0.91%/year), as illustrated in Figs. 4C and 6. In contrast, the more diseased hemifield exhibited significantly (P < 0.01) greater GCL thinning of 3.0%/year compared to healthy volunteers. The observed loss of ganglion cells and GCL thinning corresponds to a decline in visual sensitivity, consistent with the physiological relationship between ganglion cell numbers and visual sensitivity (19). Overall, changes observed with PD showed no significant trend in either more diseased (P = 0.24) or less diseased (P = 0.51) hemifields of patients with glaucoma (Figs. 4D and 6). Similarly, total deviation changes showed no significant trends in either more (P = 0.52) or less diseased (P = 0.43) hemifields, with complete patient PSD and MD values detailed in the Supplementary Materials. All clinical OCT, visual field, and AO-OCT measures at corresponding AO-OCT imaging locations are summarized in figs. S2 to S5 and tables S3 to S8.

Fig. 6. Clinically derived longitudinal individual measurements.

Longitudinal data of GCL thickness and visual field PD from clinical measurements in healthy (blue) and glaucoma (orange) cohorts at 3°, 6°, and 12° eccentricity. In patients with glaucoma, solid and open symbols denote less diseased and more diseased hemifields, respectively.

The GCL was segmented from AO-OCT volumes for all participants, locations, and time points (see Supplementary Text). Total soma volume was calculated from the segmented masks. AO-OCT and clinical OCT GCL thickness measurements were statistically similar across most regions, except for the 3° location, where the difference in field-of-view size between the two modalities (3° versus 1.5°) where the GCL thickness gradient is steepest yielded different thickness values (fig. S6). Volumetric decomposition of the AO-OCT segmented GCL (figs. S7 and S8 and tables S9 and S10) into portions occupied by somas and “noncellular” components (vasculature, other structures, and other cell types) revealed that the soma volume portion decreased by 6% and the noncell portion increased by 6% over time in the glaucoma cohort. The proportion of cellular and noncellular components was unchanged over time in healthy volunteers. Further, of the 11% overall GCL volume reduction (in cubic millimeters) for the glaucoma cohort, soma volume in absolute values (in cubic millimeters) was reduced by 30%, while noncell volume decreased by only 2%. This indicates that despite some compensatory soma enlargement (Fig. 5), GCL thinning in glaucoma is predominantly due to RGC loss. To confirm that en face diameter increases reflect true volumetric enlargement rather than axial flattening, longitudinal ellipticity was computed from three-dimensional segmented soma masks. Ellipticity did not differ significantly between patients with glaucoma and healthy volunteers (Mann-Whitney U, P = 0.67) and showed no significant change over time in patients with glaucoma, whether in the more diseased or less diseased hemifield (P = 0.58 and P = 0.85, respectively) or healthy volunteers (P = 0.55), indicating that soma shape is preserved across the observation period (see table S11). The absolute ellipticity values observed in both healthy volunteers and patients with glaucoma were consistent with those reported in our previous study (20).

Comparative slope analysis

To examine potential regional differences of glaucomatous degeneration, we performed a longitudinal slope analysis to determine the relationship between the normalized annual loss rate of GCL soma density, GCL thickness, and PD, chosen as the most sensitive metrics from AO-OCT, clinical OCT, and functional tests, respectively. All patients with glaucoma with six visits were included in the analysis; the complete datasets and time points are provided in the supplementary tables. The results indicate that the rate of loss exhibits regional variability depending on the hemifield defect, as shown in Fig. 7. Overall, our analysis revealed that the annual loss rate was lower and had a weak linear coefficient of determination with clinically derived GCL thickness (r² = 0.27) and PD (r² = 0.32). In contrast, we observed a larger slope and stronger linear coefficient of determination with GCL soma density (r² = 0.78). Together, the absolute value of the slope and coefficient of determination results suggest that GCL soma density is more sensitive and may provide a better prediction of the state of heterogeneous GCL soma loss in patients with glaucoma compared to conventional measures of GCL thickness and PD. Normalized GCL soma density loss for patient 1733, who was not included in the slope analysis because they were only imaged twice, is shown in fig. S9.

Fig. 7. Regional variability in glaucomatous GCL soma loss.

Normalized annual loss rates of GCL thickness, PD, and soma density in patients with glaucoma shows regional variability when compared between less diseased (L) and more diseased (M) hemifields. AO-OCT–derived GCL soma density exhibits greater coefficient of determination (r² = 0.78) of loss rate compared to PD (r² = 0.32) and GCL thickness (r² = 0.27) across eccentricities, suggesting potentially higher prediction capability for detecting heterogeneous cell loss.

AO-OCT detects neuronal degeneration significantly earlier than clinical standards

To further illustrate the relative sensitivity of AO-OCT imaging, clinical OCT imaging, and functional visual field measurements, neurodegeneration analysis, as defined in Materials and Methods, was performed on four patients with glaucoma (1541, 4179, 4856, and 8533) who had six visits over 3 years using GCL thickness, visual field PD, and GCL soma density measurements. Events of neurodegeneration (or progression) were identified on the basis of significant negative slope with P < 0.05 over time (trend based) as described by Zhang et al. (3). For example, patient 4179 exhibited a 9.4%/year decrease in GCL thickness and 14.5%/year decrease in GCL soma density at the 3I location, as shown in Fig. 8A. For this patient, the AO-OCT–derived GCL soma density detects progression (blue arrow) 0.58 years earlier than the clinical measure of GCL thickness (orange arrow).

Fig. 8. Glaucoma neurodegeneration detection using AO-OCT and comparison to clinical metrics.

(A) Analysis of GCL thickness, soma density, and PD reveals that GCL soma density detects progression (blue arrow) 0.58 years earlier for patient 4179 at 3I location compared to GCL thickness (orange arrow), demonstrating the higher sensitivity of AO-OCT compared to clinical parameters. In contrast, progression detection using the functional metric PD remains undetected for 3 years. (B) Kaplan-Meier plots of glaucoma fraction of progression show early detectability with AO-OCT compared to clinical OCT and visual field metrics.

A Kaplan-Meier analysis was performed on a per-location basis to determine the cumulative fraction of progression for GCL soma density, GCL thickness, and PD among patients with glaucoma (Fig. 8B). By the end of the study period, the fraction of progression across 24 imaging locations in four patients with glaucoma reached 42% for GCL soma density, 17% for GCL thickness, and 8% for PD. GCL soma density outperformed both GCL thickness and PD in term of progression detection rate. GCL soma density detected progression 0.74 years earlier than GCL thickness and 1.49 years before visual field PD over the duration of the study.

DISCUSSION

This multiyear longitudinal study demonstrates that AO-OCT, when coupled with a rigorous analysis approach, provides an unprecedented view into the cellular changes of glaucoma, enabling detection of disease neuronal degeneration with significantly greater sensitivity than the current clinical imaging techniques. Our principal findings are that (i) AO-OCT can clearly distinguish the higher RGC loss of glaucoma from the lower aging related loss; (ii) direct cell counting with AO-OCT circumvents the limitations of surrogate GCL layer (or GCC) thickness and subjective functional visual field measurements, which are less sensitive; and (iii) AO-OCT–derived biomarkers may detect statistically significant neuronal degeneration years earlier than conventional methods.

A key strength of our study is the ability to distinguish pathological RGC loss from the natural course of aging. Our measurement of a 0.27% annual loss rate in healthy volunteers (Fig. 4A) is in excellent agreement with histological and imaging-based studies that place age-related RGC and axon loss in the range of 0.1 to 0.6%/year (13, 21, 22). Against this slow, physiological baseline, the glaucomatous eyes in our cohort exhibited pathological decay rates that were 3 to 30 times higher. This is a critical validation of the technology’s sensitivity. Crucially, in our small cohort, conventional OCT-derived GCL thickness was not sensitive enough to distinguish the thinning in the less diseased hemifield of patients with glaucoma from that of healthy volunteers (Fig. 4C) which underscores a key limitation of analyses based on bulk tissue layer thickness measurements. Although, in a larger cohort, the GCL thickness difference between healthy and glaucomatous eyes would likely reach significance at more locations. AO-OCT, however, detected significant cell loss in these same regions, highlighting its superior sensitivity for detecting early disease progression.

Our work also contributes to the understanding of cellular morphological changes in glaucoma. The observation of increased GCL soma diameter in glaucomatous eyes (Fig. 4B) confirms a previous cross-sectional report (12) and may represent cellular remodeling, as the remaining cells expand to fill the voids left by neuronal loss. The weak correlations between annual loss rates and traditional metrics (GCL thickness and PD) versus the robust correlation with AO-OCT–derived soma density in Fig. 7 underscores the potential of cellular-resolution imaging to overcome the floor effects and measurement variability that limit conventional structural assessments in glaucoma monitoring (23, 24). We note that the slope difference between more and less diseased hemifield regions was numerically larger for GCL soma density than for GCL thickness. However, in the absence of disease, GCL thickness showed no significant change over time in healthy volunteers, while soma density showed a small but detectable age-related decline, indicating that the greater dynamic range of GCL soma density across the full range from healthy aging to advanced glaucoma supports its utility as a sensitive metric. The interpretation of relative sensitivity is complex and will require confirmation in larger, prospective studies. Furthermore, volumetric decomposition of the GCL demonstrates that RGC soma volume reduction substantially outpaces bulk GCL thinning in glaucoma (figs. S7 and S8 and movie S4), with some evidence in the results of structural remodeling (compensatory soma enlargement and expansion of noncellular material), which is not detectable with conventional OCT but can be resolved with AO-OCT.

The findings indicate that cellular-level structural markers may be more effective than tissue-level structural markers in glaucoma assessment. The key distinction lies in the nature of the measurement itself: AO-OCT provides a discrete, quantitative count of surviving neurons, whereas conventional OCT provides a bulk measurement of layer thickness. The observation that the strong correlation between RGC soma density and GCL thickness present in healthy eyes weakens in glaucomatous eyes (Fig. 3E) points to the complex pathological changes occurring within the GCL. GCL thickness, as measured by clinical OCT, is an aggregate measure, which includes not only RGCs but also essential glial support cells like astrocytes and Müller cells. In healthy retinas, RGC soma density is a significant determinant of GCL thickness, resulting in the strong correlation observed. However, in glaucoma, the relationship becomes more complex as multiple cellular and structural changes occur simultaneously. Glial cells may undergo reactive gliosis, and other pathological processes such as cellular swelling, extracellular matrix remodeling, or inflammatory responses may alter GCL thickness independently of RGC number. Dynamic changes in the GCL include changes in surviving RGCs, such as soma enlargement (12) or dendritic atrophy (25), as the retina attempts to compensate for cell loss and rebuild synaptic connections. Consequently, the overall GCL thickness may not decline in direct proportion to the true number of apoptosing RGCs. The weakened cross-sectional correlation between GCL thickness and soma density in glaucoma is quantitatively consistent with the longitudinal volumetric findings, where soma volume declined by 30% while overall GCL volume declined by only 11%, with the noncellular fraction increasing by 6%. This structural remodeling partially attenuates the thickness signal relative to true neuronal loss, explaining why GCL thickness is a less direct surrogate for soma density in glaucoma. The structure-function correlation we observed between GCL soma density and PD (Fig. 3F), however, confirms that direct cell counting remains a robust and functionally relevant measure of neuronal health (19). The modest structure-function correlation observed (r = 0.54 to 0.88) is expected given the inherent high test-retest variability of standard visual fields (26, 27). Therefore, the ability to detect a notable structure-function relationship at this highly localized, cellular scale further validates the sensitivity of AO-OCT in capturing biologically relevant damage.

These findings are consistent with the definition of glaucoma as a chronic, progressive optic neuropathy resulting in RGC degeneration and death, rather than a disease defined solely by elevated intraocular pressure (28, 29). This work provides a potential cellular explanation for a well-recognized clinical phenomenon: continued structural and functional worsening in some patients despite intraocular pressure lowering to a predefined target. Large longitudinal studies have demonstrated that although intraocular pressure reduction reduces the risk and rate of progression, it does not fully arrest disease activity in all eyes. In the Early Manifest Glaucoma Trial, treated eyes progressed less frequently than untreated controls, yet a substantial proportion still demonstrated progression over time (30, 31). Similarly, the Collaborative Normal-Tension Glaucoma Study showed that lowering intraocular pressure significantly reduced progression risk, but a meaningful subset of treated eyes continued to deteriorate, suggesting persistent neurodegenerative processes once injury has been initiated (32). In this context, our observation of continued RGC soma loss in eyes that are clinically stable by standard-of-care global metrics, including all visual field and conventional OCT indices, highlights a dissociation between clinical stability and cellular stability, indicating that glaucomatous neurodegeneration may persist even when conventional measures appear unchanged and that this may be common. The relevance of this work lies in its ability to quantify ongoing neuronal loss directly; if such measures were available clinically, they could prompt consideration of further intraocular pressure lowering or alternative interventions in selected eyes. More fundamentally, these findings underscore the need for neuroprotective therapies in glaucoma and suggest that in vivo quantification of RGC density and morphology may represent an ideal biomarker for future neuroprotection trials, directly measuring neuronal integrity rather than relying solely on indirect structural or functional surrogates.

From a clinical translational perspective, the most impactful finding is the substantial lead time in progression detection (as defined in this manuscript) offered by AO-OCT. Our data show that AO-OCT identified significant RGC loss nearly a year before current clinical standard-of-care methods (GCL thickness and PD) and in cases where conventional methods failed entirely (Fig. 8B). This capability could be transformative for both new treatment trials and patient management. While the practical requirements of this technology now limit its widespread clinical use, this work demonstrates a notable step toward more sensitive disease monitoring. For developing neuroprotective therapies, using a sensitive cellular endpoint could markedly shorten trial durations, reduce sample sizes, and lower costs (33–35).

High-resolution AO-OCT imaging for detecting glaucoma progression faces several technical challenges that now limit translation but present opportunities for areas of future innovation. Now, cellular-level imaging is constrained by a relatively small field of view, a limitation primarily driven by the slow acquisition speeds inherent to the AO-OCT systems. However, emerging technical advances, particularly the development of megahertz-speed AO-OCT platforms (36–38), hold considerable promise for addressing these limitations by enabling larger field of view imaging and facilitating the creation of detailed montages that can capture more extensive retinal areas while maintaining cellular resolution (39, 40). Beyond expanding the imaging area, faster acquisition speeds would substantially reduce susceptibility to eye motion artifacts, which now compromise image quality and necessitate use of robust registration algorithms. Furthermore, the integration of artificial intelligence–powered image processing algorithms may revolutionize high-resolution imaging workflows by reducing or potentially eliminating the need for extensive frame averaging, as has been demonstrated for other retinal cell imaging with AO-OCT (41), thereby accelerating data acquisition while maintaining the signal-to-noise ratios necessary for reliable cellular visualization. Collectively, these technological innovations—spanning hardware and computational advances—represent a pathway for high-resolution AO-OCT imaging device translation for more clinically practical and accessible glaucoma screening and monitoring.

The primary limitation of our study is its small sample size. However, the study’s strength lies in its high-density, longitudinal data from a few individuals, which provide a demonstration of the technology’s capability for sensitive monitoring. Given the nature of longitudinal studies, time points are not exactly equally spaced across the modalities (mean intervisit intervals: 0.536 ± 0.061 years for AO-OCT, 0.578 ± 0.086 years for clinical OCT, and 0.646 ± 0.040 years for visual field test for the 0.5 year design interval; CV ≤ 15% across all modalities). However, given the change detected with each modality (Figs. 4 to 6), we do not expect that unequal sampling will alter one of the primary conclusions of the paper: AO-OCT detects significant cell loss earlier than the clinical OCT or functional visual field test. Future larger cohort and multisite studies may establish normative databases for AO-OCT metrics. Future work could focus on developing fully automated, end-to-end analysis pipelines to improve throughput and objectivity, making the technology more feasible for clinical trials and, eventually, routine practice. The development of more accessible and cost-effective AO-OCT systems may also be crucial for wider adoption. As demonstrated here, AO-OCT–derived metrics of RGC health are highly sensitive biomarkers. With further validation, these metrics may serve as powerful clinical trial endpoints, potentially shortening trial durations and accelerating the development of new neuroprotective therapies for glaucoma. Our study positions AO-OCT as a scientific instrument capable of quantifying cellular-level changes, highlighting its potential for glaucoma monitoring and as a tool for understanding cellular pathophysiology.

MATERIALS AND METHODS

Participants and clinical examination

Five patients with glaucoma (62.2 ± 5.2 years) and five age-matched healthy volunteers (65.3 ± 6.8 years) (Table 1) were enrolled in this longitudinal study. Written informed consent (protocol #17-062R) was obtained after the study protocol, and potential risks were explained to each participant. The study protocol was approved by the Institutional Review Board of Food and Drug Administration (FDA) and adhered to the tenets of the Declaration of Helsinki.

All participants had open angles documented by gonioscopy and underwent standard ophthalmic examination at each visit with optic disc photography, OCT (peripapillary and macula scans, Spectralis OCT; Heidelberg Engineering GmbH, Heidelberg, Germany), and standard automated perimetry (Humphrey Field Analyzer; Carl Zeiss Meditec Inc., Dublin, CA, USA). Both 24-2 and 10-2 SITA standard protocols were used for perimetry. All patients with glaucoma were beyond the learning curve phase. Glaucoma diagnosis was confirmed by an experienced glaucoma specialist (O.J.S.) based on the American Academy of Ophthalmology Practice Patterns. One eye of each participant was imaged for this study. Exclusion criteria included media opacities, hypertension, coexisting ocular conditions (e.g., diabetic retinopathy and age-related macular degeneration), a history of ocular surgery other than uncomplicated cataract or glaucoma surgery, unreliable visual field tests, or inability to maintain fixation. Unreliable tests were defined as those with fixation losses >20% or false-positive rates >15% (42). False-negative rates were not used for exclusion as they correlate with disease severity rather than patient attentiveness (43). All included visual field tests in this study demonstrated excellent reliability, with mean fixation losses of 2.6%, mean false-positive rates of 2.7%, and mean false-negative rates of 1.6%.

AO-OCT imaging with the FDA multimodal system

For four patients with glaucoma (1541, 4179, 4856, and 8533), AO-OCT images were acquired on six different imaging sessions separated by 181 ± 33 days (target: 182.5 days or 6 months). One additional patient with glaucoma (1733) was imaged twice with sessions 4.7 years apart. All age-matched healthy volunteers were imaged twice with sessions 4.5 ± 0.3 years apart. Patients with glaucoma were imaged frequently to precisely follow longitudinal changes, while two extended time points were sufficient to measure age-related changes in healthy volunteers. For each imaging session, AO-OCT volumes were acquired after pupil dilation and cycloplegia with 1% tropicamide, using the FDA multimodal AO system operating at 830 nm (∆λ = 60 nm) with a 210-kHz A-scan rate (44). The illumination power was maintained below 430 μW, adhering to the safety thresholds outlined by the American National Standards Institute Z80.36-2021 (45). The system focus was adjusted to approximately the GCL to maximize the contrast of RGC somas.

As in the previous cross-sectional study (12), AO-OCT volumes were acquired at six macular locations (Fig. 1): three 2.5° above and three 2.5° below the horizontal midline at 3°, 6°, and 12° temporal retinal eccentricities. The locations were strategically selected to reflect regional variations in RGCs relative to the hemifield differences (12). For each retinal location, 30 AO-OCT videos (10 volumes per video) were acquired with a field of view of 1.5° × 1.5° (∼0.2 mm² on retina) and 300 × 300 lateral pixel sampling. The same six retinal locations were reimaged for each participant across repeat sessions throughout the study. The identical locations were found for each individual during follow-up sessions with the aid of the AO-SLO imaging channel and system fixation coordinates using retinal landmarks (vessel/capillary patterns). Each imaging session lasted ∼90 min, including breaks for patient comfort, a duration comparable to prior AO-OCT studies.

Longitudinal image analysis

AO-OCT volumes were reconstructed using standard OCT processing steps, registered in three dimensions (46), and averaged. The averaged volumes were used for further morphological analysis as described previously (12). A high-performance distributed computing postprocessing pipeline in MATLAB (MathWorks, Inc., Natick, MA, USA) was implemented to expedite the processing of the large longitudinal datasets. Retinal images were scaled for eye axial length in the lateral dimension using the Bennett-Rabbetts eye model (47).

For longitudinal assessment, the averaged volumes from six sessions were further registered laterally using en face projections from the intermediate capillary plexus, which provided uniform structures throughout the field of view. The intermediate capillary plexus layer was chosen for its stable vasculature features that appear less susceptible compared to superficial vascular plexus to glaucomatous changes over time (48). To ensure consistency in registration, each projection image was further processed with a tubeness filter using ImageJ to enhance vessel features (49). The deformation field (using the MATLAB imregdemons function) was iteratively computed with each projection image selected as a reference. The image that resulted in the smallest sum of deformation fields was chosen as the final reference for longitudinal registration. The transformation matrices derived from the longitudinal registration were then applied to the averaged AO-OCT volumes to identify the overlapping region before automated counting, as shown in Fig. 2A. GCL somas within the overlapping regions of registered volumes were detected and segmented using an automated machine learning algorithm (17) and corrected manually by one of three graders who were randomly assigned volumes. Manual correction of the automated segmentation was performed by one expert grader and two trained independent graders in a partially blinded fashion. All volumes were randomly allocated among graders who were aware of the cohort (glaucoma or healthy) but blinded to participant identity and all other clinical data. Subsequently, an expert grader reviewed all the volumes to ensure consistency and accuracy in segmentation results. We manually excluded areas containing blood vessels to ensure accurate cell density calculations. Segmented cells are termed “GCL somas” in this manuscript, as this layer is primarily composed of RGCs in the central retina (varying from ∼97% at 3° to 78% at 12°), with the remainder being displaced amacrine cells (18). Our method does not distinguish between these two cell types or among RGC subtypes. To calculate GCL soma density, the three-dimensional GCL soma center locations were projected onto a single en face plane, and Voronoi analysis was applied to the soma mosaic. The GCL soma density was calculated as the ratio of the total number of Voronoi cells to the cumulative area encompassed by these Voronoi patches. Only registered volumes with minimal artifacts and clearly visible cells [mean Michelson cell contrast > 0.45 (50)] were selected for quantification, a set of criteria that 94.4% (204 of 216) of all acquired volumes met. The symmetry index was quantified by calculating the GCL density ratio at the same retinal eccentricity at matched eccentricities, as defined above (12). The GCL soma diameter was calculated from the en face projection of each segmented soma at its central depth plane (17). Further, ellipticity (vertical diameter/horizontal diameter) was extracted from the corrected three-dimensional GCL soma segmentation masks.

The clinical visual field qualification criteria was fixation losses <20% and false-positive errors <15% (51). The clinical OCT qualification criteria was an image quality measurement of >20 (52). We interpolated each participant’s OCT layer thickness and visual field PD values to the regions that correspond to the six AO imaging locations, as detailed elsewhere (12). Although the window size for the clinical OCT layer thickness (3°) and visual field PD analysis (2° for 10-2; 6° for 24-2) were larger than AO-OCT field size, these differences are not expected to generate inaccuracies for the smooth layer contours of our cohort. For visual field, we used the projection between RGCs and photoreceptors modeled by Watson (16). For all analyses of the heterogeneous manifestation of glaucoma in our cohort, we used local OCT and visual field data corresponding to the AO-OCT imaging locations, rather than global clinical metrics.

Progression analysis was conducted using OCT-derived GCL thickness and AO-OCT–derived density parameters. At each visit, parameters were analyzed from baseline to the current visit, starting from the second follow-up visit. Following Zhang et al. (3), progression was identified using Kaplan-Meier survival analysis for patients with at least five follow-up visits, when a significant (P < 0.05) negative slope was detected, indicating layer thinning for the GCL thickness measure and cell loss for the GCL soma density measure. The first visit showing a significant trend was noted as the progression detection time point. The Kaplan-Meier analysis was conducted on a per-location basis rather than a per-patient basis to account for individual spatial differences in cell loss across the macula.

Method assessment and statistics

To evaluate the inter-rater reliability of the GCL soma counting method, measurements were compared between two independent graders and one expert grader. For this analysis, a total of 12 volumes were randomly selected, comprising two volumes from each cohort (healthy volunteer and glaucoma) at each of the three eccentricities (3°, 6°, and 12°).

Statistical comparisons of baseline characteristics between patients with glaucoma and healthy volunteers were performed using the Mann-Whitney test. For longitudinal analysis, all imaging parameters (soma density, soma diameter, GCL thickness, and PD) were normalized to their baseline values to facilitate comparison of rates of change. Group comparisons were performed using the Student’s t test. For the soma size longitudinal analysis, normalized soma size change data were pooled across all imaging locations, and a Pearson’s correlation was performed on the aggregated measurements. All statistical analyses were conducted using Excel with Analysis ToolPak add-in and GraphPad Prism software (v. 9.0.2).

Supplementary Materials

The PDF file includes:

Segmentation analysis

Intergrader assessment

Figs. S1 to S9

Tables S1 to S12

Legends for movies S1 to S4

Other Supplementary Material for this manuscript includes the following:

Movies S1 to S4