Automated detection of diabetic macular edema involving cystoids and serous retinal detachment. (July 2020)

• Record Type: Journal Article
• Authors: Maurya, Prashant Kumar; Gupta, Vishal; Singh, Maninder; Singh, Amit Kumar; Kumar, Basant; Mohan, Anand
• Is Part Of: Optics & laser technology. Volume 127(2020)
• Publication Date: 2020-07
• Subjects: OCT -- DME -- Cystoid -- CME -- Subretinal; Optics -- Periodicals; Lasers -- Periodicals; Electronic journals; 621.366
• DOI: 10.1016/j.optlastec.2020.106157
• Languages: English
• ISSNs: 0030-3992

Abstract

Highlights: We propose a fully automated method for the detection of diabetic macular edema using OCT images. Uses Gradient information of the retinal boundaries to detect the layers of the retina. Investigate the presence of cystoid macular edema along with serous retinal detachment. Abstract: Diabetic Macular Edema (DME) is a complex and irreversible ocular disorder which leads to permanent vision loss. DME mostly occurs due to the formation of cyst spaces and storage of serous fluid in retina. We present a fully automated method for the detection two forms of diabetic macular edema using Optical Coherence Tomography (OCT) images. Segmentation of the retinal layers in the OCT image is important in the evaluation of the ocular diseases. Gradient information of the retinal boundaries is used to detect the layers of the retina. Retinal thickness between the vitreous-NFL (Nerve Fiber Layer) and the RPE (Retinal Pigment Epithelium) layer is measured and compared with the pre-defined values of normal retinal thickness to detect the edema in the eye. To detect cystoid macular edema (CME), the cystoid regions which are non-reflective spaces between the vitreoretinal layer and the inner -outer segment (IS-OS) layer are extracted by processing the pixel values inside the restricted search region between these layers. For detecting the serous retinal detachment (SRD), we extract the subretinal fluid region from the retinal image by changing pixel values below the IS-OS layer with zeroes and retracing the light-to-dark boundary. We also suggest criteria to validate the above result and reject any unwanted region, if detected.