Revolutionizing Retinal Health: Transfer Learning for Early Detection of Pathologies in Limited OCT Image Datasets

The study effort focuses on the urgent need to diagnose retinal disorders at an early stage using revolutionary methods in medical image processing. The work demonstrates the effectiveness of transfer learning in the medical field by using a dataset of fewer than 1,000 retinal OCT pictures, despite the large number of about 30 million OCT scans performed each year. The dataset has been carefully compiled using a hierarchical grading method that involves undergraduate students, medical students, ophthalmologists, and senior retinal specialists. It consists of four categories: DME, CNV, NORMAL, and DRUSEN. The approach used involves the utilisation of a pre-trained ResNet18 Convolutional Neural Network. The main emphasis is on adjusting the final layer to precisely address the objective of classifying retinal disease. The study achieves an impressive level of precision, reaching 94%, highlighting the promise of transfer learning as a formidable method for effectively and accurately detecting retinal abnormalities at an early stage. This work not only makes a valuable contribution to the area of medical imaging, but also serves as a complete reference for newcomers in the subject who are interested in implementing Convolutional Neural Networks and transfer learning.