Automated Segmentation of Macula in Retinal Images Using Deep Learning Methodology

A significant reason for blindness in working age people in developing nations is diabetic retinopathy. Proper treatment of this eye disorder might be challenging as it starts manifesting symptoms too late. Regular screening is the only approach to reduce the progression of sight impairment. Macula appears as a dark and ovoid region close to the center of the retina. The localization and segmentation of the macula is very essential in analyzing the impacts of macular degeneration since it plays a vital role in human vision. Specialists can determine the severity of diabetic retinopathy by looking for lesions connected to the vascular irregularities. The manual process for identifying diabetic retinopathy symptoms is time-consuming and requires expertise clinicians and equipments which are insufficient as compared to the growing number of individuals with diabetes. Therefore, the necessity to develop an automated diabetic retinopathy screening tool is recognized. In this study, an attention mechanism-based encoder–decoder framework with residual extended skip unit has been introduced to automatically segment the macula area in fundus images. This approach has outperformed the previous techniques when analyzed on three public databases, namely MESSIDOR, DIARETDB1, DIARETDB0 and recorded 94.06% overall accuracy.