Peripheral Retinal Vessel Density in Children and Adolescents with Myopia: A Deep Learning Assessment

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Investigates the impact of axial length (AL) on peripheral retinal vasculature in pediatric populations.
• Explores the potential of deep learning (DL) for detecting early vascular alterations associated with myopia.

Purpose of the Study:

• To determine the relationship between axial length and peripheral retinal vessel density.
• To assess the efficacy of deep learning algorithms in identifying subclinical vascular changes in myopia.

Main Methods:

• Utilized non-mydriatic ultra-widefield imaging for retinal image acquisition.
• Employed Nested U-Net and ResNet34 deep learning models for vascular segmentation and AL classification.
• Quantified vessel density in peripheral retinal fields (60-30° and 100-60°).

Main Results:

• Analyzed 679 eyes from 396 children and adolescents across normal, medium, and high AL groups.
• Observed higher vessel density in temporal retina and lower density in inferior retina.
• Demonstrated a significant decrease in peripheral vessel density with increasing AL (normal > medium > high).
• Achieved high accuracy (AUC=0.9651) in AL classification using deep learning, with Grad-CAM highlighting key vascular regions.

Conclusions:

• Increased axial length correlates with diminished peripheral retinal vessel density.
• Deep learning analysis of ultra-widefield images can detect early, subclinical vascular changes.
• These findings suggest a potential role for DL in the early detection of high myopia risk.