Deep Learning-Based Retinal Vascular Analysis with Quantitative Metrics for Hypertensive Microvascular Damage

Background/Objectives: The goal of this study was to determine whether deep learning (DL)-assisted retinal vascular analysis detects subclinical microvascular remodeling associated with hypertension-mediated target organ damage (TOD) in patients with absent or mild hypertensive retinopathy (HR). Methods: This prospective study included age-comparable hypertensive patients with TOD (n = 54), hypertensive patients without TOD (n = 54), and healthy controls (n = 54). Color fundus photographs underwent DL-based segmentation using the Retinal Microvascular Health Assessment System, ophthalmologist quality review, and Fiji morphometric analysis. Central retinal arteriolar and venular equivalents (CRAE and CRVE), arteriovenous ratio (AVR), and arterial and venous network metrics were quantified. Results: AVR decreased progressively from 0.814 ± 0.038 in controls to 0.740 ± 0.043 in hypertensive patients without TOD and 0.655 ± 0.042 in those with TOD (p < 0.001). CRAE and CRVE differed among groups (p < 0.001 and p = 0.002). Arterial fractal dimension and vessel area density (VAD) declined stepwise (both p < 0.001), whereas arterial skeletonized vascular density (SVD) and diameter were primarily reduced in TOD. Venous VAD and diameter differed (p = 0.022 and p = 0.006), but venous SVD did not (p = 0.146). After adjustment among hypertensive participants, TOD remained associated with lower AVR, CRAE, and all arterial network metrics except tortuosity (all p < 0.001), but not with venous parameters. Conclusions: Retinal arterial alterations were associated with established extra-retinal TOD in patients with absent or mild HR. Residual cardiovascular confounding limits interpretation, and longitudinal studies are needed to establish temporal relationships, predictive value, and clinical utility.