Artificial intelligence assisted quantitative analysis of fundus structural and vascular alterations reveals fundus...

Myopia is characterized by progressive axial elongation and structural remodeling of the posterior segment, ultimately leading to irreversible visual impairment in high myopia. However, early, noninvasive biomarkers that capture the continuum of fundus alterations across myopia severity remain insufficiently defined. Recent advances in artificial intelligence (AI) enable high-throughput quantitative analysis of fundus images, providing new opportunities for identifying imaging-based biomarkers.

This cross-sectional observational study included 539 eyes from 274 participants with varying degrees of myopia. Based on cycloplegic spherical equivalent (SE), eyes were categorized into low, moderate, high, and super-high myopia groups. Color fundus photographs were analyzed using an AI-assisted framework to quantify fundus tessellation (FT), peripapillary atrophy (PPA), optic disc morphology, and retinal vascular parameters, including fractal dimensions, vessel geometry, and vessel density. Logistic regression and receiver operating characteristic (ROC) analyses were performed to identify independent biomarkers and evaluate their discriminative performance. Linear regression was used to assess associations with SE.

Fundus structural and vascular parameters showed significant differences across myopia severity groups. FT-related parameters increased progressively with increasing myopia severity (all P < 0.001), with macular FT area demonstrating the strongest association. Retinal vascular complexity and vessel density decreased with increasing myopia severity, reflected by reduced fractal dimensions and vessel density (all P < 0.001). In multivariable analysis, macular FT area (OR = 2.925, P < 0.001) and PPA height (OR = 1.501, P = 0.001) were independently associated with high myopia, while vertical optic cup diameter showed an inverse association (OR = 0.664, P < 0.001). Vascular parameters did not retain independent significance after adjustment. ROC analysis showed that macular FT area achieved the highest discriminative performance (AUC = 0.819), and a combined model yielded an AUC of 0.869. Linear regression demonstrated a strong association between FT parameters and SE.

Fundus structural alterations, particularly fundus tessellation and peripapillary atrophy, are the most robust imaging biomarkers associated with myopia severity. While refraction and axial length remain the primary measures of myopic severity, FT provides complementary information about posterior segment remodeling. Prospective studies are needed to evaluate whether FT predicts future complications such as myopic maculopathy or retinal detachment. Retinal vascular changes appear secondary and contribute less to disease discrimination. AI-assisted quantitative fundus analysis provides a noninvasive and scalable approach for identifying individuals at risk of high myopia and offers new insights into the structural-vascular remodeling underlying myopia progression.