Charting New Frontiers using Transfer Learning in OCT Image Analysis for Retinal Health

The project aims to address the pressing necessity of early-stage diagnosis of retinal illnesses by the use of groundbreaking techniques in medical image processing. The study showcases the efficacy of transfer learning in the medical domain, utilising a dataset of less than 1,000 retinal OCT images, despite the substantial volume of over 30 million OCT scans conducted annually. The dataset has been meticulously assembled utilising a hierarchical grading process that incorporates input from undergraduate students, medical students, ophthalmologists, and senior retinal experts. The classification has four categories: DME, CNV, NORMAL, and DRUSEN. The methodology employed entails the utilisation of a pre-trained ResNet18 Convolutional Neural Network. The primary focus is on fine-tuning the last layer to accurately target the goal of categorising retinal illness. The study demonstrates a remarkable degree of accuracy, achieving a 94% success rate. This emphasises the potential of transfer learning as a powerful approach for early detection of retinal disorders with high effectiveness and precision. This work not only provides a useful addition to the field of medical imaging, but also acts as a comprehensive reference for those new to the subject who are interested in applying Convolutional Neural Networks and transfer learning.