Automated detection of neovascularization for proliferative diabetic retinopathy screening

Area of Science:

• Ophthalmology
• Medical Imaging
• Computer Vision

Background:

• Proliferative diabetic retinopathy (PDR) is a leading cause of acquired blindness.
• Neovascularization, the growth of new blood vessels, is a key indicator of PDR.
• Accurate classification of neovascularization is crucial for timely treatment and vision preservation.

Purpose of the Study:

• To develop and validate a novel method for classifying neovascularization in proliferative diabetic retinopathy (PDR).
• To differentiate between neovascularization in the optic disc (NVD) and neovascularization elsewhere (NVE) for improved diagnostic accuracy.
• To reduce false positive rates in neovascularization detection for better patient screening and prioritization.

Main Methods:

• Extraction of optic disc (OD) region and blood vessels from retinal images.
• Classification of major blood vessel segments within a 1-OD diameter region as NVD.
• Classification of minor blood vessel segments outside the 1-OD diameter region as NVE using optimal feature sets.

Main Results:

• Achieved high classification performance for NVD (74% sensitivity, 98.2% specificity, 87.6% accuracy) and NVE (61% sensitivity, 97.5% specificity, 92.1% accuracy).
• Demonstrated 86.4% sensitivity and 76% specificity for PDR screening using combined public and local datasets.
• The method effectively reduces false positives in neovascularization classification.

Conclusions:

• The proposed method accurately classifies NVD and NVE, aiding in the diagnosis of PDR.
• This automated approach can significantly improve the efficiency and accuracy of PDR screening.
• The NVD and NVE detection methods are valuable tools for patient prioritization in diabetic retinopathy management.