A Multi Label Glaucoma Classification Using a Lightweight Attention Mechanisms

Glaucoma is a leading cause of irreversible blindness worldwide, with early detection and accurate diagnosis being critical for effective treatment. However, its asymptomatic nature in the early stages often leads to delayed interventions. Manual screening methods are not only time-consuming but also prone to errors, necessitating automated and reliable diagnostic tools. This research proposes a novel deep learning-based approach for the multi-class classification of glaucoma into three stages: normal, early, and advanced. The methodology employs preprocessing techniques to enhance retinal image quality, followed by pattern recognition using a custom-designed deep convolutional neural network. Comprehensive experimentation was conducted using publicly available datasets, including ORIGA and Harvard to validate the model's performance. The proposed approach aims to significantly reduce diagnostic errors, providing ophthalmologists with a reliable tool for accurate staging of glaucoma. This enables the development of tailored treatment strategies, potentially preventing further visual impairment and improving patient outcomes.