A Unified Optic Nerve Head and Optic Cup Segmentation Using Unsupervised Neural Networks for Glaucoma Screening

Area of Science:

• Ophthalmology
• Medical Image Analysis
• Computational Neuroscience

Background:

• Accurate segmentation of the optic nerve head (ONH) and optic cup is crucial for glaucoma diagnosis.
• Existing methods struggle to segment both ONH and optic cup simultaneously or require extensive feature engineering and ground truth data.
• Supervised learning methods necessitate numerous informative features and ground truth for training.

Purpose of the Study:

• To develop a novel, simultaneous segmentation method for the optic nerve head (ONH) and optic cup.
• To improve the accuracy and efficiency of glaucoma assessment through automated retinal image analysis.
• To overcome limitations of previous segmentation techniques by utilizing saliency maps and self-organizing neural networks.

Main Methods:

• Utilized a saliency map (color, intensity, orientation contrasts) as input for a winner-take-all neural network.
• Employed color visual features as input for a self-organizing map neural network for simultaneous segmentation.
• Evaluated the method on a diverse dataset of 205 ocular fundus images from multiple sources.

Main Results:

• Achieved an average overlapping error of 9.6% for ONH segmentation and 25.1% for optic cup segmentation.
• Computed the cup-to-disc area ratio (CDR) for glaucoma assessment.
• Demonstrated a mean CDR difference of 0.11 and standard deviation of 0.09 compared to ground truth.

Conclusions:

• The proposed method effectively segments the optic nerve head and optic cup simultaneously.
• The achieved accuracy in segmentation and CDR calculation supports its utility in computer-aided glaucoma diagnosis.
• This approach offers a promising alternative to existing methods for automated glaucoma screening and monitoring.