Blood Vessel Segmentation Using U-Net for Glaucoma Diagnosis with Limited Data

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Glaucoma is a leading cause of global blindness.
• Early detection and diagnosis are critical for preserving patient vision.
• Retinal blood vessel analysis is important for diagnosing eye conditions.

Purpose of the Study:

• To develop and optimize a blood vessel segmentation model for retinal fundus images.
• To evaluate the performance of a U-Net based model using different loss functions.
• To enhance early glaucoma detection through improved image analysis.

Main Methods:

• Utilized the U-Net architecture for blood vessel segmentation.
• Trained the U-Net model with three distinct loss functions.
• Employed hyperparameter tuning to optimize model performance for each loss function.
• Evaluated models on retinal fundus images from the SALUS study.

Main Results:

• Achieved accuracy exceeding 93% across the best models.
• Obtained Dice scores around 83% and Intersection over Union scores over 70%.
• Demonstrated reliable identification of both large and small blood vessels in retinal images.

Conclusions:

• The developed U-Net model shows high efficacy in retinal blood vessel segmentation.
• The model's ability to identify various vessel sizes supports improved glaucoma diagnosis.
• This segmentation model offers a promising tool for enhanced glaucoma management and patient care.