Leveraging Machine Intelligence Driven Models for Diagnosis and Detection of Retinal Detachment Through Color Fundus Images

Abstract

Retinal detachment (RD) is a chronic ophthalmic emergency that may cause irreversible vision loss if left untreated. RD is characterized by separating the retinal pigment epithelium from the neurosensory layer, accumulating fluids in the sub-retinal space. The incidence of RD in Eastern nations, such as India, China, Singapore, and Korea, ranges from 7.98 to 17.9 episodes per 100,000 inhabitants. The annual incidence of RD in Western nations like the Netherlands, America, and Scotland varies from 12.05 to 18.2 cases per 100,000 persons. The occurrence of RD is substantially higher in men compared to females, with a male-to-female ratio of 1.3:1. Therefore, it is crucial to promptly detect and treat RD to address this worldwide epidemic of avoidable visual impairment. Early recognition of RD is challenging because it suddenly develops and grows substantially from small sizes at the retina s periphery. Most individuals who experience the early symptoms of RD, such as flashes of light, floaters, and curtain-like shadows, often ascribe these concerns to the natural aging process. Early identification of RD needs an expert ophthalmologist to evaluate the whole retinal region through dilated fundus following mydriasis. The manual screening process is tedious and laborious. Additionally, there is a possibility of subjective mistakes occurring. These limitations impede the creation of screening systems for RD diseases in underdeveloped nations that lack diagnostic equipment and competent ophthalmologists. Identifying RD in fundus images during the early stages is challenging due to the complex structures of RD lesions and their varied locations. The RD was previously recognized using retinal lesions contrast and intensity characteristics. Nevertheless, these strategies were ineffective because of the fluctuating medical environment characterized by varying contrast and brightness levels in the fundus images.