Enhancing OCT Image Classification for Retinal Disease Diagnosis: A Novel Approach Using Squeeze-and-Excitation ResNet

Millions of individuals are projected to be impacted by age-related macular degeneration (AMD), which is the primary cause of visual loss. Conversely, only a few number of highly qualified ophthalmologists are produced globally each year. Over the last decade, machine learning technologies have demonstrated significant potential to revolutionise clinical AMD care and help research for a better understanding of the disease. As a result, we developed a SeResNet architecture to categorise various AMD diseases. SeResNet is a technology for identifying retinal disorders such as Diabetic Macular Edema and Age-Related Macular Degeneration that focuses on recalibrating characteristics at the channel level which boosts decision-making, feature representation, and classification accuracy. The model was evaluated using 84,495 images categorised as CNV, DME, Drusen, or Normal. To determine which model performs best, we trained four alternative models. The results show that SeResNet performs very well, with recall and precision scores up to 1.00 across a variety of categories and an overall accuracy score of 99%. In particular, SeResNet helps for earlier detection of diseases and better treatment for patients by decreasing the requirement for on-site examinations. This paper highlights the utility of SeResNet model in ophthalmology. In addition, by analyzing the most sophisticated AI techniques for OCT imaging, it offers recommendations for more study and technical developments.