Fig. 6: Distribution of latent features and clinically meaningful morphological indices in MEH cohort.
From: Understanding pre-training data effects in retinal foundation models using two large fundus cohorts
a Shows t-SNE visualisation with latent features across age subgroups extracted by FMs trained with Masked Autoencoder, b shows t-SNE visualisation after eliminating confounding effects by specifying sex and ethnicity (e.g. Female, Asian or Asian British). c, d Show t-SNE visualisation with latent features extracted by FMs trained with DINOV2. e Demonstrates the distribution density of clinically meaningful morphological indices over age subgroups, while f shows the distribution density after specifying sex and ethnicity. A Kruskal–Wallis H-test followed by Holm–Bonferroni correction (n = 2) was conducted to assess statistical significance. Both artery fractal dimension (p = 7.04E−5) and vein fractal dimension (p = 1.26E−5) show significant differences, even after controlling the sex and ethnicity, artery fractal dimension (p = 0.004) and vein fractal dimension (5.11E−4). These indicate strong data variations across age subgroup.
