Publication

Abstract : Glaucoma, an irreversible neurodegenerative disease leading to vision loss, remains a significant public health concern due to challenges in early detection and resource-intensive diagnostic procedures. Leveraging advancements in artificial intelligence (AI), particularly Convolutional Neural Networks (CNNs), presents a promising avenue for automating glaucoma detection, especially in regions lacking access to ophthalmologists. This study, explores the efficacy of glaucoma detection systems based on MobileNet architecture for mobile deployment, focusing on key factors such as dataset variability, size, and training time. This paper presents findings from two distinct datasets, namely chaksu and g1020, each trained on separate models termed the chaksu model and g1020 model, respectively. These models, evaluated against the Akshi dataset, demonstrate impressive performance, suggesting the feasibility of utilizing complete fundus images for Glaucoma detection, circumventing the complexities of segmentation tasks involving disc, cup, and blood vessels. In this study, LIME (Local Interpretable Model-agnostic Explanations) was integrated to enhance interpretability and shed light on the decision-making process of our glaucoma detection system based on MobileNet architecture. This research underscores the potential of mobile technologies coupled with AI in democratizing glaucoma screening, potentially mitigating the challenges posed in resource-constrained settings.