Joint optic disc and cup boundary extraction from monocular fundus images

Area of Science:

• Ophthalmology
• Medical Imaging
• Computer Vision

Background:

• Accurate optic disc and cup segmentation is vital for glaucoma screening and diagnosis.
• Existing methods struggle with depth information in monocular fundus images, leading to separate segmentation of optic disc and cup.
• Current techniques often rely on color and vessel kinks, lacking explicit depth cues.

Purpose of the Study:

• To develop a novel method for joint segmentation of optic disc and cup boundaries from monocular color fundus images.
• To explicitly model depth information within the segmentation process.
• To improve glaucoma screening by enabling accurate segmentation without expensive 3D imaging.

Main Methods:

• A boundary-based Conditional Random Field formulation for simultaneous optic disc and cup segmentation.
• Estimation of depth from fundus images using a coupled, sparse dictionary trained on image-depth map pairs.
• Integration of color gradients and estimated depth for improved segmentation accuracy.

Main Results:

• The estimated depth showed a strong correlation (0.80) with ground truth.
• The proposed method outperformed state-of-the-art techniques on five public datasets.
• Achieved high segmentation performance (Dice coefficient 0.87-0.97 for disc, 0.83 for cup) and good glaucoma classification (AUC 0.85).

Conclusions:

• A novel method for jointly segmenting optic disc and cup boundaries by modeling depth drop.
• The technique enables large-scale glaucoma screening using single fundus images, bypassing the need for costly 3D imaging.