Optic disk detection and segmentation of retinal images using an evolution strategy on GPU

In this paper we show an optic disk (OD) detection and segmentation approach based on evolution strategy (ES) implemented on GPU using CUDA (Compute Unified Device Architecture). The approach has two main steps: Coarse detection and contour edges refinement. Coarse detection estimate a position approximation using an ES which bright pixels amount and the vascular structure edge pixels are considers in its objective function. The contour edge refinement uses a geometrical approach for circle deformation in order to adjust the edge circle with OD edges. For this, the pixel with the largest intensity value variation along a normal line is considered. The proposed method was evaluated using the STARED and DIAREDB public repository, processing normal and disease patient states retinal images. In the experimental results we show that the computational time for optic disk detection task has a speedup factor of 5x and 7x compared to an implementation on a mainstream CPU and identifies the optic disk position with an accuracy of 96%.