Optic disc detection based on fully convolutional network and weighted matrix recovery model.

Area of Science:

• Ophthalmology
• Medical Imaging
• Computer Vision

Background:

• Accurate optic disc contour detection is vital for diagnosing and treating eye diseases.
• Fundus image complexity and blood vessel interference pose challenges to optic disc segmentation.
• The optic disc is typically a salient region in fundus images.

Purpose of the Study:

• To propose a weakly-supervised method for accurate optic disc detection in fundus images.
• To leverage fully convolutional neural networks (FCN) and weighted low-rank matrix recovery (WLRR) for improved segmentation.
• To address the challenges posed by complex image structures and blood vessel disturbances.

Main Methods:

• Feature extraction and pixel clustering using the Simple Linear Iterative Clustering (SLIC) algorithm to form a feature matrix.
• Integration of top-down semantic priors from FCN and bottom-up background priors for optic disc region.
• Construction of a prior information weighting matrix to guide the decomposition of the feature matrix into sparse (optic disc) and low-rank (background) components.

Main Results:

• The proposed method accurately segments the optic disc region in fundus images.
• Experimental results on DRISHTI-GS and IDRiD datasets demonstrate superior performance compared to existing weakly-supervised methods.
• The combined FCN and WLRR approach effectively handles complex image structures and blood vessel interference.

Conclusions:

• The developed weakly-supervised method provides accurate optic disc segmentation.
• This approach offers a promising solution for automated diagnosis and treatment of eye diseases.
• The integration of FCN and WLRR enhances the robustness and precision of optic disc detection in challenging fundus images.