ABSTRACT
Diabetic Macular Edema (DME) is a prevalent condition linked to diabetes that can cause vision complications if not identified and addressed promptly. This study presents a novel approach of the YOLO (You Only Look Once) deep learning method for the purpose of detecting diabetic macular edema (DME). Our approach leverages the advancements, in the YOLO v8 framework integrating a core network, streamlined detection module and modified loss function to improve object detection capabilities. We trained and verified our model using a dataset of approximately 10,000 labeled retinal images, including standard image categories and bounding box annotations for normal and DME conditions. The evaluation enabled us to gauge the algorithm's effectiveness in binary classification. Performance tests were conducted based on various criteria like precision, recall, and F1-score. Our study findings suggest that YOLO v8 demonstrates performance in detecting DME compared to previous versions, which could potentially assist in early detection, in clinical environments. Our new method outlined in this study achieved an accuracy of 94.4%.
