AI-Based Quantitative Assessment of Retinal Vascular Morphology in Circumscribed Choroidal Hemangioma

This study aimed to apply a deep learning-based artificial intelligence (AI) system for quantitative analysis of retinal vascular morphology in patients with circumscribed choroidal hemangioma (CCH), and to explore differences based on lesion location.

This retrospective case-control study included 45 treatment-naive CCH eyes and their contralateral healthy eyes as controls, recruited from 45 patients (mean age: 44.91 ± 11.98 years; 10 female patients). Retinal photographs were analyzed using AI software to extract vascular parameters, including vessel density, caliber, tortuosity, and fractal dimension. Inter-group comparisons and conditional logistic regression were conducted. Subgroup analysis was performed based on the lesion's position relative to the optic disc.

Compared with controls, CCH eyes showed significantly reduced vascular density (p < 0.005) within 3 mm and 5 mm of the fovea. Venular caliber was significantly increased across multiple concentric zones (1.0-2.5 papillary diameter, PD), while arteriole-to-venule ratio (AVR) was decreased (p < 0.001). Tortuosity and fractal dimension of both arterioles and venules were significantly reduced (all p < 0.05). Logistic regression confirmed vessel caliber (p = 0.001), AVR (p = 0.001), tortuosity (p = 0.006), and fractal dimension (p = 0.004) as significant parameters associated with CCH. Lesions within 1.0 PD of the optic disc were linked to lower arteriolar caliber and AVR (p < 0.05).

Retinal vascular morphological alterations are evident in CCH and vary with lesion location. Key parameters, such as vascular density, venular caliber, AVR, and fractal dimension, may serve as potential imaging biomarkers for evaluating and monitoring CCH-related retinal changes.