Background The use of artificial intelligence (AI) for glaucoma detection through fundus retinal images has increased significantly in recent years. This study aims to identify recent research trends in AI-based glaucoma detection systems. Methodology We conducted a systematic literature review of 46 main studies to identify research trends regarding artificial intelligence-based glaucoma detection systems between 2020 and 2025. The recognized factors include the utilization of datasets, feature extraction and classification methods, evaluation metrics, and contributions to glaucoma detection based on fundus images. Results The review found that 16 out of 46 use real-world datasets, often obtained from eye hospitals and related institutions. Among public datasets, the most widely used were REFUGE, DRISHTI-GS, ORIGA, RIM-ONE, ACRIMA, and HRF. Regarding methodology, while convolutional neural network (CNN) remain the dominant backbone, recent studies have increasingly moved beyond basic CNN architectures by employing advanced strategies such as residual network (ResNet), EfficientNet, Vision Transformers, attention mechanisms, hybrid segmentation-classification pipelines, transfer learning, and explainable AI components for feature extraction and classification. An analysis of the evaluation matrices with various public datasets shows that the same model works differently for each type of dataset, indicating differences in the data distribution. Conclusions The reviewed studies contributed to several aspects, including data augmentation, application of transfer learning methods, image segmentation, explainable AI, and diagnosis time. This trend indicates a growing contribution of substantial significance to the effective identification of glaucoma using fundus. This study strengthens this field by providing a comprehensive evaluation of recent advancements in artificial intelligence-driven glaucoma detection research.
Saidah et al. (Tue,) studied this question.