Artificial Intelligence in Myopic Maculopathy: A Comprehensive Review of Identification, Classification, and...

This review investigates the usefulness and effectiveness of artificial intelligence (AI) tools in the detection of myopic maculopathy lesions using traditional imaging techniques like fundus photography and optical coherence tomography (OCT). The role of machine learning (ML) and deep learning (DL) algorithms in the diagnosis, classification, and follow-up of highly myopic cases is discussed. A comprehensive analysis of articles published between 2018 and 2024 from PubMed, Science Direct-Elsevier, and Google Scholar identified 13 studies directly relevant to the topic. The majority of the studies were conducted in China and focused on patients with myopic macular degeneration and high myopia. The most popular AI algorithms included ResNet-18, ResNet-50, ResNet-101, DeepLabv3+ and DarkNet-19, Efficient Net (B0/B7), VOLO-D2, Efficient Former, ALFA-Mix+, and XGBoost. Reported statistical metrics ranged from 80% to 97.3% for accuracy, 80% to 99.8% for the area under the curve (AUC), 83.0% to 97.0% for sensitivity, 63.0% to 97.21% for specificity, and 0.8358 to 0.9880 for the kappa value. The findings reveal that AI models can play a supportive role in disease diagnosis, achieving performance metrics comparable to those of general ophthalmologists. Furthermore, the utilization of larger datasets of OCT and fundus images improves generalizability and diagnostic accuracy. The integration of multimodal imaging techniques, such as OCT, color fundus photographs, and ultra-wide field photographs, enhances diagnostic clinical value and enables more comprehensive disease monitoring.