GRD-Net: Graph Neural Network Model for Retinal Detachment Classification Through Fundus Image

Retinal detachment can lead to irreversible vision loss if not treated promptly. Early identification of RD is crucial for infected individuals and ophthalmologists who provide therapy. Fundus imaging screening is a more effective technique for detecting retinal detachment and aiding in diagnosis. To facilitate the RD diagnosis process, this work developed a graph convolutional network-based framework named GRD-Net to automatically classify retinal detachment and healthy fundus images. This work utilizes the simple linear iterative clustering approach for graph construction based on the Euclidean distance, and graph features are extracted using our proposed GRD-Net model. These graph features are then encoded to provide prediction results on fundus images. As far as we know, this is the first effort to include graph information in the RD classification problem. The GRD-Net architecture offers the best approach for early screening of retinal detachment with limited datasets and minimal memory requirements. The proposed model has been assessed using four publicly available repositories, including the Retinal Image Bank, Cataract Image Dataset, Retina Dataset, and Kaggle. Our proposed framework surpassed the prior methodologies with improved classification performances, and implementing this approach on a global scale might significantly decrease the prevalence of visual impairment caused by retinal detachment through prompt identification and referral.