An Automated Glaucoma Detection from Fundus Images based on Deep Learning Network

Abstract

A condition known as glaucoma, is an eye illness brought on by high intraocular pressure, may lead to total blindness. On the other hand, prompt glaucoma screening-based therapy may keep the individual from losing all vision. Professionals manually analyze retina to pinpoint the areas affected by glaucoma using precise testing procedures. However, because of complicated glaucoma testing methods and a lack of resources, delays in detection are often experienced that may raise the global rate of visual impairment. Moreover, the significant resemblance between the lesion and eye color also makes the manual categorization procedure more difficult. Hence, there exists an urgent need to develop an effective smart approach that can precisely detect the Optic Disc as well as Optic Cup lesions at the early stage in order to address the difficulties of manual methods. Therefore, a Deep Learning based strategy called EfficientDet-DO with EfficientNet-B0 serving as its foundation has been proposed in this paper. There are three phases in the conceptual methodology for the localization and categorization of glaucoma. First, the EfficientNet-B0 feature extractor computes the feature representations from the suspicious examples. Next, the top-down and bottom-up key points merging operations are repeatedly carried out by the Bi-Directional Feature Pyramid system modules of EfficientDet-DO using the calculated characteristics from EfficientNet-B0. The resulting localized areas of a glaucoma lesion and its accompanying classification are anticipated in the last stage.