Wednesday 12 March 2025
Researchers have made a significant breakthrough in the field of medical imaging, developing a new technique that can accurately diagnose diabetic retinopathy and glaucoma from fundus images. These conditions are among the leading causes of blindness worldwide, and early detection is crucial for preventing vision loss.
The technique, known as Class Extension with Limited Data (CELD), uses deep learning algorithms to analyze fundus images and identify features associated with diabetic retinopathy and glaucoma. The approach is particularly effective in dealing with imbalanced data sets, where one class has significantly more instances than the others.
To develop CELD, researchers combined three publicly available datasets of fundus images, including 3,111 images from healthy eyes, diabetic retinopathy patients, and individuals with glaucoma. They then trained a deep neural network to classify the images into two classes: healthy or diabetic retinopathy. Subsequently, they fine-tuned the model to recognize glaucoma, transforming it into a three-class classifier.
The results were impressive, with CELD achieving an overall accuracy of 91% in diagnosing diabetic retinopathy and glaucoma. The technique was particularly effective in identifying diabetic retinopathy, with an F1-score of 0.8971 compared to the state-of-the-art model’s score of 0.5797.
The researchers also used perturbation techniques to analyze the importance of different features in fundus images for accurate diagnosis. They found that the green channel plays a critical role in distinguishing between diabetic retinopathy and healthy eyes, while image quality is essential for glaucoma detection. The optic disc region was also identified as a key feature for diagnosing glaucoma.
The development of CELD has significant implications for the early detection and prevention of vision loss due to diabetic retinopathy and glaucoma. The technique can be used in clinics to rapidly and accurately diagnose these conditions, enabling timely treatment and reducing the risk of blindness.
Moreover, CELD’s ability to deal with imbalanced data sets makes it a valuable tool for diagnosing rare eye diseases. As the technique continues to evolve, it has the potential to improve patient outcomes and reduce healthcare costs associated with vision loss.
The researchers’ work highlights the importance of developing machine learning algorithms that can effectively handle imbalanced data sets. Their approach demonstrates the potential of deep learning in medical imaging and could pave the way for new applications in other fields where data is often skewed.
Cite this article: “Breakthrough in Medical Imaging: Accurate Diagnosis of Diabetic Retinopathy and Glaucoma with CELD Technique”, The Science Archive, 2025.
Medical Imaging, Diabetic Retinopathy, Glaucoma, Deep Learning, Fundus Images, Imbalanced Data Sets, Class Extension With Limited Data, Machine Learning, Vision Loss, Blindness Prevention
