Yixue Yin, Xinran Guo, Shuo Xu, Xiuyan Zhang, Hongsheng Bi, Jike Song
Automated quantitative SLO analysis was completed in 716 of 821 initially available early postoperative images. Group-level differences were observed across clinically recorded cataract subtype groups, but these findings were exploratory and may reflect non-standardized subtype assignment, unmeasured spherical-equivalent refractive error, age, axial length, image characteristics, and other confounding. They should not be interpreted as subtype-specific structural phenotypes or as evidence of diagnostic, prognostic, or clinical decision utility. Prospective studies with standardized cataract grading and cohort-specific algorithm validation are required.
OBJECTIVE: To explore the application of automated quantitative scanning laser ophthalmoscopy analysis to early postoperative images and to characterize group-level variation in the resulting measurements across patients grouped according to clinically recorded age-related cataract subtype.
METHODS: This retrospective observational imaging study included 716 eyes from 571 patients with age-related cataract who underwent phacoemulsification combined with intraocular lens implantation between August 2023 and August 2025. Early postoperative scanning laser ophthalmoscopy (SLO) images were analyzed using EVisionAI version 4.0 as a standardized automated quantitative image-analysis tool. Eyes were grouped as nuclear cataract (394 eyes), cortical cataract (217 eyes), or posterior subcapsular cataract (PSC; 105 eyes) according to routine preoperative slit-lamp records. Standardized LOCS III grading and cohort-specific validation of EVisionAI against manual annotations were not available in this retrospective dataset. Automated image-derived parameters included tessellated fundus-related features, optic disc morphology, peripapillary atrophy (PPA)-related parameters, and retinal vascular parameters. Patient-level baseline characteristics were compared using conventional statistical tests, as appropriate. Eye-level comparisons of automated image-derived SLO parameters were performed using generalized estimating equations to account for inter-eye correlation, with Wald χ2 statistics reported. Multiple comparisons were adjusted using the Benjamini-Hochberg false discovery rate procedure. Multivariable GEE models adjusted for age, sex, axial length, and cataract subtype were used to evaluate associations with selected imaging parameters. Standardized effect-size, age-overlap, and single-eye sensitivity analyses were also performed.
RESULTS: After false discovery rate correction, multiple automated image-derived fundus parameters differed significantly among the retrospectively defined cataract subtype groups, including cup-to-disc ratio parameters, optic disc morphology, PPA-related parameters, and retinal vascular parameters. Compared with PSC eyes, nuclear and cortical cataract eyes showed relatively higher cup-to-disc ratio-related parameters and larger PPA-related parameters, whereas PSC eyes showed relatively greater retinal vessel length. Across age groups, tessellated fundus density, macular tessellated fundus density, optic disc-related tessellated fundus parameters, and PPA-related parameters increased significantly with age. Among the 14 contrasts involving the seven principal outcomes, standardized effect estimates were small to moderate (absolute standardized β, 0.258-0.619), and none met the descriptive threshold for a large effect. In multivariable GEE models, age remained associated with multiple tessellated fundus- and PPA-related parameters after adjustment for sex, axial length, and cataract subtype. Axial length also remained associated with optic disc tilt angle, PPA-related parameters, and tessellated fundus parameters. The age-overlap and single-eye sensitivity analyses showed generally consistent directions for the principal associations, although one optic disc tilt-angle contrast was attenuated after age restriction.
CONCLUSION: Automated quantitative SLO analysis was completed in 716 of 821 initially available early postoperative images. Group-level differences were observed across clinically recorded cataract subtype groups, but these findings were exploratory and may reflect non-standardized subtype assignment, unmeasured spherical-equivalent refractive error, age, axial length, image characteristics, and other confounding. They should not be interpreted as subtype-specific structural phenotypes or as evidence of diagnostic, prognostic, or clinical decision utility. Prospective studies with standardized cataract grading and cohort-specific algorithm validation are required.