International Conference on Computer and Knowledge Engineering

Home / 12th International Conference on Computer and Knowledge Engineering

Segmentation of Hard Exudates in Retinal Fundus Images Using BCDU-Net

Authors :

Nafise Ameri¹ Nasser Shoeibi² Mojtaba Abrishami³

1- Ferdowsi university of mashhad 2- Eye Research Center, Mashhad University of Medical Sciences, Mashhad, Iran 3- Eye Research Center, Mashhad University of Medical Sciences, Mashhad, Iran

Keywords :

Fundus image،Diabetic retinopathy،Hard exudate،Deep learning،U-Net extended cannulation network،Lesion segmentation،Hard exudate segmentation

Abstract :

The importance of Diabetic Retinopathy (DR) screening and the difficulty in achieving an early diagnosis of DR at a reasonable cost requires attention to the development of computer-aided diagnostic tools. Computer-Aided Diagnosis (CAD) of retinal detachment imaging can reduce mass screening of the diabetic population and help physicians make the best use of their time. For this purpose, the deep learning technique and the developed U-Net canonization network have been used. Using this network, it receives retinal images and shows the segmentation of the hard exudate lesion as a binary image. The result of this research has been evaluated on the IDRID dataset with three important indicators of dice coefficient, sensitivity, and accuracy achieve at 76.81%, 72.24%, and 99.30%, respectively, and the effectiveness of the approach was confirmed.