Aparna Paliya and Amit Singhal, Dr. Bhimrao Ambedkar University, India

Eye cancers such as retinoblastoma, ocular melanoma, and eyelid tumors present diagnostic difficulties due to their rarity, diversity, and the increased risk of vision loss or death if detected late. Recent advancements in deep learning, particularly through convolutional neural networks (CNNs), have significantly improved the precision and automation in the detection and segmentation of ocular tumors. This review article examines the latest applications of Convolutional Neural Networks (CNNs) in detecting eye cancer, drawing on findings from ten recent studies conducted between 2021 and 2024. It highlights the importance of CNNs in improving diagnostic accuracy and efficiency in ophthalmology, suggesting they could significantly change clinical practices. Results reveal developments in CNN architectures that lead to better detection rates on the retinoblastoma and melanoma among different eye cancer types by analysing retinal images (e.g. fundus imaging, OCT, MRI) and histological records. It also introduces the clinical impact of these improvements referring to embedding the CNNs in routine tests. Recent advancements in deep learning, particularly through convolutional neural networks (CNNs), have significantly improved the precision and automation in the detection and segmentation of ocular tumors. The models analyzed include traditional architectures like LeNet and VGG16, as well as more advanced methods such as variations of U-Net, multi-view CNNs, ConvNeXt, and hybrid systems like FedCNN paired with XGBoost. Nonetheless, issues remain concerning dataset diversity, clinical validation, interpretability, and the practical use of these technologies. This review highlights both the notable advancements achieved and the gaps that still need to be filled to enable the incorporation of CNN-based systems into standard ocular cancer treatment.

Keywords