In this cross-sectional study, patients diagnosed with primary open-angle glaucoma (POAG) and glaucoma suspects who were enrolled in the Diagnostic Innovation in Glaucoma Study (DIGS; clinicaltrials.gov identifier NCT00221897, IRB #140 276) ^, between January 2015 and December 2021 were included. Eligible participants underwent both OCT-A ONH (Angiovue, Optovue Inc, Fremont, California, USA) and OCT ONH (Spectralis, Heidelberg Engineering GmbH, Heidelberg, Germany) within a 6-month interval. Assessments followed a standardized protocol, incorporating scheduled follow-up visits that comprised clinical examinations, imaging procedures, and functional tests. Written informed consent was obtained from all participants before enrollment. The study protocols were approved by the University of California, San Diego Human Subjects Committee, and the methods employed adhered to the tenets of the Declaration of Helsinki.

Eyes were classified as glaucomatous if they exhibited repeatable abnormal visual field (VF) test results, accompanied by signs of glaucomatous optic neuropathy. This was defined as optic disc excavation, focal thinning, notching of the neuroretinal rim, or localized or diffuse RNFL defects, as assessed through masked grading of optic disc photographs by two independent graders or by clinical examination. An abnormal VF test result was defined as at least two consecutive tests showing a pattern SD (PSD) outside the 95% normal confidence limits or a Glaucoma Hemifield Test result outside normal limits. Glaucoma suspects were identified based on elevated intraocular pressure (IOP) (≥22 mm Hg) or signs of glaucomatous optic neuropathy, in the absence of repeatable VF abnormalities indicative of glaucomatous damage.

Participants older than 18 years, with open angles on gonioscopy and a best-corrected visual acuity of 20/40 or better at study entry were eligible to participate. Patients with a history of ocular trauma or intraocular surgery, except for uncomplicated cataract or glaucoma surgery, as well as coexisting retinal disease, uveitis, non-glaucomatous optic neuropathy, and poor-quality OCT-A or OCT images were excluded. Additionally, individuals diagnosed with systemic conditions such as Parkinson’s disease, Alzheimer’s disease, dementia, or a history of stroke were also excluded.

Myopia was classified based on AL, measured using partial coherence interferometry (IOLMaster500, ZEISS, Jena, Germany). Myopia was categorized into three groups: no myopia (AL ≤ 24.0 mm), mild myopia (24.0 mm < AL ≤ 26.0 mm), and high myopia (AL > 26.0 mm). This approach was chosen because the axial elongation is potentially associated with structural changes in the optic disc and retina, emphasizing the limitations of defining myopia solely by refractive error. Refractive error indeed may not accurately reflect axial elongation and its related anatomical alterations. Additionally, in eyes with axial elongation that have undergone refractive or cataract surgery, postoperative changes in refractive error may result in the misclassification of myopia when based on refraction alone.

Deep ONH structure measurements were obtained using the Spectralis spectral domain OCT (SD-OCT) (Spectralis HRA + OCT, software version1.10.12.0; Heidelberg Engineering GmbH, Heidelberg, Germany). The details of the SD-OCT imaging procedure have been previously reported. Briefly, 24 radial B-scans of the ONH image were used to automatically segment the 48.

BMO points as well as the internal limiting membrane (ILM). Image quality and segmentation accuracy were evaluated, and any segmentation errors detected by the automated Spectralis software were manually corrected by experienced examiners from the Imaging Data Evaluation and Analysis Reading (IDEA) Center.

Methods for manual segmentation of OCT ONH images have been previously described. ^,,, In summary, raw OCT volumes were imported into custom 3D visualization and segmentation software (Devers Eye Institute, ATL 3D Suite, Portland, OR) which employs a deep learning-based modified U-Net framework for automatic segmentation of each radial B-scan. The segmented anatomical landmarks included the ILM, posterior surface of the RNFL, posterior surface of the Bruch’s membrane/retinal pigment epithelium complex, BMO, neural canal wall, anterior scleral surface, and ASCO. The ASCO was identified on both sides of the canal by visually projecting the plane of the peripapillary anterior scleral surface through the neural canal wall and marking their intersection. All automated segmentations were reviewed, and manual corrections were performed when necessary, by a single operator (HY).

Detailed descriptions and methods for parameterizing the following deep ONH parameters have been previously reported. ^,,

BMO and ASCO size and shape (ovality index). A plane was fitted to the 48 segmented BMO and ASCO points, respectively, satisfying a least mean square error restraint in each case. The BMO points were then projected onto the best-fit BMO plane, where a best-fit ellipse was determined. Using this ellipse, the BMO area and ovality index (ratio of the long axis length to the short axis length of the ellipse) were calculated. Similarly, ASCO area and ovality index were determined within the ASCO plane. ^,,

Neural Canal Minimum Cross-sectional Area (NCMCA). As previously described, ^, NCMCA was calculated in a plane perpendicular to the neural canal axis after projecting BMO and ASCO points onto the plane and identifying the overlapping area between the two projections. The NCMCA size and ovality index were then computed within the projection plane, following the same methods used for BMO and ASCO.

BMO/ASCO Offset Magnitude. The BMO and ASCO coordinates were projected onto the BMO reference plane, and best-fit ellipses were fitted to determine the centroids of BMO and ASCO. ^,,, The magnitude of BMO/ASCO offset was derived by projecting the vector connecting the ASCO and BMO centroids to the BMO reference plane (the BMO/ASCO centroid vector), and transposing it from the projected ASCO centroid to the BMO centroid. ^,

Neural Canal Obliqueness. Neural canal obliqueness was defined as the angle between the neural canal axis vector (determined by the BMO/ASCO centroid vector as described above) and the vector perpendicular to the BMO plane, originating at the BMO centroid, as described previously. ^,

The AngioVue OCT-A system (software version 2018.1.1.63) was employed to obtain 4.5 × 4.5 mm² ONH-centered scans, consisting of 304 B-scans with 304 A-scans per B-scan. Retinal and choroidal layer segmentation was automatically performed using the AngioVue software, generating an en-face choroidal vessel density map. OCT-A image quality was evaluated following the standardized protocol of the IDEA Center, ensuring that only good quality scans were included in the analysis. Images were excluded if they exhibited any of the following criteria: (1) low scan quality with a quality index (QI) below 4; (2) poor image clarity; (3) residual motion artifacts, such as irregular vascular patterns or distorted disc boundaries in the en-face angiogram; (4) significant image truncation or localized signal attenuation due to vitreous opacity; or (5) segmentation errors that could not be corrected.

In this study, juxtapapillary MvD was defined as the complete absence of choriocapillaris within the parapapillary beta zone. ^, The parapapillary beta zone, characterized by the presence of Bruch’s membrane and the absence of the retinal pigment epithelium (RPE), facilitated the visualization of choroidal vessels and the underlying sclera. The dropout had to extend across at least four consecutive horizontal B-scans, have a minimum diameter of 200 µm in at least one scan, and be in direct contact with the OCT optic disc boundary, as automatically detected by the Optovue software or manually refined by an independent reader. In cases where the automated determination of BMO was inaccurate, a trained observer, masked to the clinical details of participants, manually adjusted the OCT-based BMO delineation by identifying Bruch’s membrane termination, as previously described. The assessment of MvD presence, as well as the quantification of MvD area and angular circumference, was conducted independently by two ophthalmic specialists (AJ and EM), who were masked to both the clinical characteristics of participants and the timing of scan acquisitions. If challenges or ambiguities arose in identifying MvD or its exact location, a consensus was reached through a group discussion involving three experienced graders (AJ, EM, and TN).

The clinical disc margins and parapapillary zone were identified by simultaneously viewing stereoscopic optic disc photographs and scanning laser ophthalmoscopic-like images obtained along with OCT-A scans. The measurement of MvD area and angular circumference followed previously established methods. ^,,,, In brief, the MvD area was manually delineated on en-face OCT-A choroidal vessel density maps using the line tool in ImageJ software (vs1.53; National Institutes of Health, Bethesda, Maryland, USA). To determine the MvD angular circumference, two reference points were established at the outermost borders of the MvD area where it intersected the BMO border, which serves as the anatomical landmark of the optic disc margin. The angular circumference was then determined by drawing two lines connecting the BMO center to the circumferential margins of the MvD. The MvD measurements were adjusted using Littmann’s formula, which corrects for ocular magnification effects in OCT-A. ^,, The default parameters of the Avanti SD OCT include an AL of 23.95 mm and an anterior corneal curvature radius of 7.77 mm. If the MvD extended beyond large retinal vessels, the area containing those vessels was included as part of the MvD region. However, when the retinal vessels were located at the border of the MvD, the area occupied by the vessels was excluded from the MvD measurement. Reflectance or shadowing artifacts from large vessels visible on horizontal and en-face images were excluded from quantitative analysis by two independent graders (AJ and EM). In eyes exhibiting more than one MvD, the area and angular extent of each MvD were measured separately, and their values were summed to obtain the total MvD area and total angular extent for that eye.

Patient and eye-level demographic and clinical characteristics were presented as mean (95% CI) and count (%) for continuous and categorical variables, respectively. Stratified analyses allowed for comparisons based on myopia status as well as the presence of MvD. Subject-level continuous parameters were compared via t -tests and one-way analysis of variance (ANOVA), while subject-level categorical parameters were compared using Fisher’s Exact test and Chi-Squared test. Linear mixed-effects models, fitted with a random intercept to account for between-subject variability and adjust for correlated repeated measures, were utilized to cf eye-level continuous parameters. Mixed-design ANOVA was also utilized to cf eye-level continuous parameters, when appropriate.

Interobserver agreement in detecting the presence of MvD was assessed by using Cohen’s Kappa, and an intraclass correlation coefficient (ICC) was used for the MvD area and angular circumference measurements.

Generalized linear mixed-effects models were used to evaluate the binary presence of MvD, while linear mixed-effects models were applied to continuous outcomes, including MvD area and angular circumference. Univariable mixed-effects models were used to analyze direct associations between clinical and demographic characteristics of interest, including deep ONH structures, and MvD outcomes. Multivariable mixed-effects models were then fit for each deep ONH structural parameter, adjusting for potential confounders, including age, race, IOP, VF mean deviation (MD), and AL. To minimize multicollinearity and variance inflation, each multivariable model included one deep ONH structural parameter at a time. Associations with binary MvD presence are reported as odds ratios, while continuous outcomes are presented as regression coefficients. All mixed-effects models incorporated a random intercept to adjust for between-subject variability and within-subject correlations. Statistical analyses were performed using R (vs 4.5.1, The R Foundation for Statistical Computing, Vienna, Austria). Two-sided P -values less than.05 were considered statistically significant.