GRENZE International Journal of Engineering and Technology
Vol. 12
(2026), Issue 2
Automating Retinal Health Assessment: A Lightweight Deep Learning Solution for OCT Image Analysis
Authors
Abhinav Pandey, Deeksha Sirohi, Abhinav Gupta, Anshika Singh, Shreya
Abstract
The world's top causes of permanent vision loss include retinal illnesses including Drusen, Diabetic Macular Edema, and Choroidal Neovascularization. Although optical coherence tomography (OCT) has emerged as the gold standard for diagnostic imaging, the manual processing of the vast number of OCT images is laborious and prone to inter-observer variability. Deep learning has showed great promise in automating this process, especially with regard to Convolutional Neural Networks (CNNs). Unfortunately, a lot of high performance models need a lot of computing power, which restricts their use in actual clinical situations. For the automated multi-class categorization of retinal disorders using OCT images, this research suggests a lightweight and effective deep learning architecture. We make use of the Keras framework-implemented MobileNetV2 architecture, a model intended for high accuracy at minimal computational cost. Over 84,000 OCT pictures classified as CNV, DME, Drusen, and Normal are used in a sizable, publicly accessible dataset for training and validating the model. Transfer learning, which preserves computer economy while achieving high classification accuracy, was used to train our model over 15 epochs. This study shows that lightweight CNNs, such as MobilenetNetV2, are a dependable and effective diagnostic tool for ophthalmologists.
Pages:
2036 - 2042