Purpose: To quantify the structural changes in choroidal vessels and to observe choroid microstructural changes in different age and sex groups in a healthy Chinese population. Methods: Enhanced depth imaging optical coherence tomography (EDI-OCT) was employed to analyze the luminal area, stromal area, total choroidal area, subfoveal choroidal thickness (SFCT), choroidal vascularity index (CVI), large choroidal vessel layer (LCVL), choriocapillaris–medium choroidal vessel layer, and LCVL/SFCT of the choroid in the subfoveal macular area within 1500 μm of the macula. We analyzed the age- and sex-related changes in the subfoveal choroidal structure. Results: A total of 1566 eyes from 1566 healthy individuals were included. The mean age of the participants was 43.62 ± 23.29 years, the mean SFCT of healthy individuals was 269.30 ± 66.43 μm, LCVL/SFCT percentage was 77.21 ± 5.84%, and the mean macular CVI was 68.39 ± 3.15%. CVI was maximum in the 0–10 years group, decreasing with age, and the lowest values occurred in the >80 years group; LCVL/SFCT was the lowest in the 0–10 years group, increasing with age and reaching a maximum in the >80 years group. CVI showed a significant negative correlation with age, and LCVL/SFCT showed a significant positive correlation with age. There was no statistically significant difference between males and females. Interrater and intrarater reliability was less variable with CVI than with SFCT. Conclusions: The choroidal vascular area and CVI decreased with age in the healthy Chinese population, of which the age-related decrease in vascular components maybe dominated by the decrease in choriocapillaris and medium choroidal vessels. Sex had no effect on CVI. The CVI of healthy populations showed better consistency and reproducibility when compared with SFCT.
Fundus fluorescein angiography (FFA) is the gold standard for assessing diabetic retinopathy (DR) severity, but whether different contrast diffusion times affect staging remains unclear. Following two clinical DR staging criteria, we established DR staging models on FFA images respectively adopting Swin-Transformer and ConvNeXt architectures and investigated the impact of FFA diffusion time on staging performance. A total of 7,508 FFA images were annotated according to both staging criteria, and divided into venue, recirculation, and late phase. Single-phase and combined-phase images were used for multi-class and binary classification tasks respectively. Model performance was evaluated using accuracy, precision, recall, and F1-score. A generalized linear model with Bonferroni correction was adopted for inter-phase statistical comparisons. Experimental results demonstrated ConvNeXt architecture outperformed Swin-Transformer in overall accuracy. No statistically significant impact of diffusion time images on staging performance was observed for either model (P > 0.05). Although staging accuracy numerically declined across successive phases, the differences were not significant, possibly attributed to reduced contrast and obscured lesions caused by hyperfluorescent leakage in late phase. Our findings verify that different contrast diffusion times have only a marginal, non-significant impact on DR staging accuracy, offering theoretical guidance for intelligent DR diagnosis using FFA images.
Fundus fluorescein angiography (FFA) is the gold standard for diabetic retinopathy (DR) staging, yet whether angiographic phase affects deep learning performance remains unknown. Using 7508 FFA images from 863 eyes, stratified into venous, recirculation, and late phases, we developed Swin Transformer- and ConvNeXt-based DR staging models under the International five-grade and Chinese six-grade classification systems. Multiclass and binary (NPDR vs. PDR) classification tasks were evaluated. This study provides the first systematic quantification of angiographic phase effects on FFA-based DR staging. Phase-related performance was analyzed using generalized linear mixed-effects models with beta regression. ConvNeXt generally outperformed Swin Transformer, particularly in multiclass classification. Across all experimental settings, performance showed a consistent numerical decline from the venous to the late phase. Under the International five-grade system, ConvNeXt accuracy declined from 86.67% to 81.87%; however, Bonferroni-adjusted comparisons revealed no statistically significant phase-related differences (all adjusted p > 0.05), with small effect sizes (|SMD| < 0.2). Binary classification remained highly stable across phases, with accuracies exceeding 91%. Grad-CAM visualizations demonstrated progressively diffuse model attention in later phases. These findings support phase-flexible FFA acquisition for binary DR screening, whereas venous- or recirculation-phase images remain preferable for high-precision multiclass staging, guiding phase-aware AI development.