STUDY DESIGN AND PARTICIPANTS

A subset of normal eyes with varied refractive statuses was randomly selected from the population-based Beijing Eye Study 2011. ^, For eyes with RE, defined as spherical equivalent values between +1.00 and −4.00 diopters (D), a best-corrected visual acuity (BCVA) of 20/25 or better was required. For eyes with RE ranging from −4.00 D to −6.00 D, a BCVA of 20/33 or better was required. Highly myopic eyes, identified as having an RE of ≤ −6.00 D or an AL of ≥ 26.00 mm, were included irrespective of their visual acuity. One eye per participant was selected. Eyes presenting with retinal diseases or optic neuropathies, including diabetic retinopathy, ocular trauma, retinal detachment, retinal vein occlusions, age-related macular degeneration, glaucoma, or nonglaucomatous optic neuropathy, were excluded. Additionally, eyes with poor OCT image quality, insufficient for segmentation, or inadequate fundus imaging for analysis were also excluded.

The included eyes were categorized into 5 groups: nonmyopia (RE +1.00 D to −0.50 D), mild myopia (RE −0.50 D to −3.00 D), moderate myopia (RE < −3.00 D but not high myopia), simple high myopia (SHM), and PM. High myopic eyes, with RE of ≤ −6.00 D or an AL of ≥ 26.00 mm, were further classified as SHM or PM, with PM defined by the META-PM Study Group criteria as the presence of myopic retinal degenerative lesions, including diffuse chorioretinal atrophy, patchy chorioretinal atrophy, or macular atrophy (categories 2, 3, and 4). All fundus photographs were independently assessed by 2 experienced ophthalmologists (YM and YXW), who were blinded to the participants’ refractive statuses.

QUANTIFICATION OF POSTERIOR CURVATURE

All eyes were scanned using spectral-domain optical coherence tomography (SD-OCT; Spectralis OCT, Heidelberg Engineering, Heidelberg, Germany) in a cube scan mode. The scans consisted of 31 continuous horizontal B-scans, covering an area of 30° × 30°, centered on the fovea. The image penetration resolution was 3.87 µm/pixel along the z-axis, 13.71 µm/pixel along the x-axis, and 292.07 µm/pixel along the y-axis, with each scan line averaged over 100 scans for enhanced quality.

OCT images were segmented using a multiple-surface OCT segmentation algorithm developed with a deep-learning approach and an interior point method. This algorithm has demonstrated a test error of less than one pixel. ^, Automatic segmentations were reviewed meticulously, and any errors were manually corrected by a trained examiner (ZP) using OCT Explorer software. For curvature analysis, the outer boundary of the retinal pigment epithelium (RPE) was extracted. A screening area of 30° × 25° was selected to exclude the optic nerve head. Before analysis, all scans were smoothed using locally weighted scatterplot smoothing (LOWESS). ^,

The posterior eye curvature was represented using 3 parameters. The first parameter, Curvature ^mean , represented the mean global curvature across all 31 B-scans. To calculate this, each B-scan within the target region was sampled at 50-pixel intervals and fitted to a circle using a least-square fitting method. The global curvature of each B-scan was then calculated as the inverse of the radius of the fitted circle. The second parameter, Curvature ^max , represented the highest 5% of pointwise curvature values. This was determined using a pointwise curvature calculation, where the curvature at each pixel was computed using a calculus-based formula, allowing for a detailed analysis of local curvature variations. The third parameter, Curvature ^macula represented the curvature of the macula and was calculated as the average of the pointwise curvature of the central scan passing through the macula ( Figure 1 ). Notably, we did not standardize the OCT images to a reference eye for a more accurate curvature calculation, ^, as the highly deformed eyes, especially those with high myopia in our study, may not align with a reference eye.

The method to calculate posterior eye curvatures. A. A representative fundus image with a red rectangle illustrating the curvature analysis area. B. The strategy to calculate global curvature with least squares fitting. Curvature ^mean was derived from the global curvature calculation approach. C. The strategy to calculate pointwise curvature with calculus. Curvature ^max and curvature ^macula were derived from the pointwise curvature calculation approach. D. A representative heatmap of pointwise curvature, with red regions indicating higher curvature and black regions indicating lower curvature. The 3 dotted white lines correspond to the OCT B-scan images shown on the right, which depict retinal cross-sections at the marked positions in the heatmap.

In addition to numerical curvature values, the spatial distribution of curvature was analyzed. The position of peak curvature was determined by averaging the locations of points with the top 10% highest pointwise curvature values.

STATISTICAL ANALYSIS

Curvature indexes were calculated using Python V.3.11 (Python Software Foundation, Wilmington, Delaware, USA; https://www.python.org/ ). Statistical analyses were performed using SPSS v.29.0 and Prism 10. All data were presented as means ± standard deviations. A P value <.05 was taken to be statistically significant. The linear regression was conducted to demonstrate the correlation of curvature parameters with age, AL, RE, BCVA, subfoveal choroidal thickness (SFCT), intraocular pressure (IOP), central corneal thickness (CCT), and anterior chamber depth (ACD). An Independent t-test was performed to evaluate the curvature difference between genders. A one-way ANOVA analysis was performed to assess whether differences exist in curvature parameters, SFCT, AL, and BCVA across various stages of myopia, followed by a trend test. Post hoc analyses were conducted to compare curvature parameters, SFCT, AL, and BCVA among pairwise myopia groups. Two sample Z-tests were performed to assess whether the differences in the curvature and ophthalmic parameters between SHM and PM were significantly greater than other pairs. To investigate the association of curvature parameters with PM occurrence in the high myopia population, logistic regression analysis was conducted, both unadjusted and adjusted for age, AL and RE. The areas under the receiver operating characteristic curve (AUROC) of the 3 curvature parameters, axial length, and refractive error were calculated and compared using MedCalc (Version 23.0.9). The cut-off value was determined by the maximal value of the Youden index (sensitivity + specificity −1). To assess the relationship between the spatial distribution of peak curvature points and different myopia groups, Pearson Chi-square tests were performed. A Cochran-Armitage test was further performed to evaluate whether the percentage of peak curvature points located in the fovea exhibited a trend across myopia severity groups.